text
stringlengths 28
334k
|
|---|
<s>[INST] You know, they're kinda thinking ahead, I guess. Well, what about also just imagine a world where India and China doing fentanyl trade with Mexico or whatever transport. Imagine Chinese military moves, makes an agreement, [/INST] a NATO type of agreement with Mexico. That's pretty possible. Again, we're seeing militarized Mexico. It's another aspect of Mexico that, again, I haven't seen talked about a lot here in the US. The main promise that the current president had was he was gonna make the police, the federal police and the security issues in Mexico civilian. He was gonna do exactly the opposite as his main rival, Felipe Calderon, the guy that started off the drug war officially. And what does he do? He dissolves the civilian leadership of the federal police, dissolves the federal police, creates the National Guard, which is a military unit, and he puts the military in charge of that. Now the military has a full monopoly over all federal policing. When you cross into Mexico, you'll see them wearing these white camouflage uniforms. Those are National Guard people, but they're the military. So now you're seeing a militarized Mexico with some of these leaks that happened during the Wakamaya leaks. You're now seeing that Mexico has been hosting members of the Haitian military, and they've been training them up to go back to police their country. That's not something that Mexico has been known for hosting other nations and then training them in such a way. So, it's an interesting maneuver. Mexico has been historically neutral about getting involved in foreign conflicts, about voting in resolutions as far as invading or not invading or doing all these things. Mexico has been historically kind of neutral as it comes to some of these things. And now we're training foreign military forces to go and do that role somewhere else. We have the military building airports and building infrastructure in Mexico, and a lot of their higher-ups getting very wealthy around it. And they basically have a monopoly over who gets to have guns down there. There's one gun store in all of Mexico, and it's run by the military. And the only way you can buy a gun there is if you can buy a plane ticket to fly there and have enough money to sustain that right for that privilege. So you're seeing the military not being in its traditional role of just being the security force. Now it's policing. It's involved in, it's getting involved in politics in a big way. It's legislation that has passed to keep it on the streets in a policing role for more years now. So it should be looked at closer by anybody observing it from afar, the militarization of Mexico </s><s>[INST] and where it's going by the military. Because if you move towards a world where a World War III happens, it feels like Mexico will be the center. Because a hot war would be fought on the ground. And so you have a very difficult parallel between Mexico and Ukraine. Both don't have nuclear weapons. Both have relationships. So Ukraine has a relationship, or a pull towards the European Union and NATO. Mexico, at least currently, has a kind of slow pull towards China, India potentially, and Russia. And you have this divide between power centers in the world. And in terms of, just imagine hundreds of thousands of Mexican troops, hundreds of thousands of Chinese troops on the border, on the US border. Yeah. [/INST] On the Mexican side. And also the fact that that border doesn't mean anything to any sort of conflict that would happen regionally. Because that's a very easy to cross border. Doesn't matter how many walls you put across it. People are already here. This is not gonna be a war fought off in some overseas place. Like you're not gonna, this is something if it happens, if destabilization is utilized in Mexico to cause a conflict there. And it turns into a Vietnam, or a proxy war down there of a sort. Which I think in a way, you're already kind of seeing some of that through some of the conflicts going on down there. You have a new generation cartel that is being fed fentanyl from the Pacific side ports. And suspiciously, you wanna think that maybe it's favored by a foreign government of some sort in some way, shape or form, who knows. And then you have a historically in control seen in law cartel that may or may not be favored by the US in some way, shape or form. You can imagine a further conflict down there and people fostering it. And seeing the effects of basically setting a fire on the feet of the United States. Its second largest consumer of US products is Mexico. The massive wave of immigration that is going to be basically weaponized. You saw the collapse of the border security structure with a contingent of 3,000 Honduran, Guatemalan immigrants in that first wave of caravans coming to Tijuana. You saw, it was pretty bad. It was pretty bad and it could have gotten worse. Now what is gonna happen when that wave is no longer 3,000, but a million people being displaced by violence or being in fear of whatever conflict might originate down there. And just that massive wave of migration and move. I think that's an interesting thing that people should look at and how can you affect change </s><s>[INST] to try and stop some of these things that happen. Well, let me ask you at a philosophical, at a human level, what do you think about immigration? Illegal and illegal immigration from the direction of Mexico to the United States. So we have an estimated 11 million illegal immigrants in the United States and estimated 45 million legal immigrants in the United States. [/INST] A few things about that. When COVID hit, there was no shortages of produce in the supermarkets, which means that, I mean, illegal immigration is pretty much the backbone of all produce and some of the farming industries out there, most of it. So illegal immigration and illegal workers in those fields are essential workers in a way. And I think there's a weird relationship in the United States with some of these workers and how they're demonized and how they're called criminals. I think there was a state out there that passed anti-illegal immigrant worker legislation. The farmers had to look elsewhere for people to show up to work in some of these fields, which basically caused millions of dollars worth of losses for some of these farms. So anywhere you go out there in the United States, you go into the kitchens and there's gonna be paisano's there, high-level French restaurants. You'll see people from Puebla there that made their way legally and might have legalized or regularized their way into the country or in a sanctuary city. You go to the service industry, hotels, those are the people changing the blankets. Those are the people in the washrooms. You have them doing jobs that no American wants to do, realistically, and they're everywhere in this country. And they are the backbone of some of these industries that are essential in this country. </s> |
<s>[INST] but he was able to survive. [/INST] Yes. And you know that there have been assassinations, Not not a lot. And uh you know that </s><s>[INST] that we know of [/INST] good point. </s><s>[INST] This is very difficult. The question is how many lines our intelligence agencies are willing to cross to attain [/INST] to </s><s>[INST] achieve the goal? [/INST] I think none of these agency sees have the ultimate line. I think eventually the last line will be crossed if they believe it's necessary. </s><s>[INST] Well, I think you can justify a lot of things, especially in this modern world with nuclear weapons that you can justify, that you're saving the world. Actually let me ask a few difficult questions and we'll jump back to your time in America. But Vladimir Putin has been accused of ordering the poisoning and assassination of several people, including alexander Litvinenko early on all the way to alexei Navalny. Do you think these accusations are grounded in truth? And we will return to a couple more questions maybe above Vladimir Putin's early days in the KGB, which would be interesting. [/INST] Yeah, there's a there's a phrase that I'd like to say in response is called plausible deniability. I don't think Putin gave a direct command as they do that. It would just maybe muse it would be nice if something would happen and then somebody picks it up and does it </s><s>[INST] is there can you steal me on the case that Putin did not have direct or indirect involvement [/INST] with who would know, you know just, well </s><s>[INST] the international the reputation perhaps um perhaps catalyzed by Putin himself is that he is the kind of person that would directly or indirectly make those orders. Perhaps the case there is he's somebody to be feared and thereby [/INST] you out there. Uh But the act itself, uh the the the poisoning of Litvinenko and oh and then the assassination of the Bulgarian Markov and with with the umbrella and they all directly traced back to Russians soviet intelligence. And so that's enough to be feared right? Um My answer that I gave you is an educated guess. You know, I can't pretend to know this for sure. But </s><s>[INST] it's frustrating to me because there's a lot of people listening to this would say would even sort of with chuckle at the naive nature of the question. But if you actually keep an open mind you have to understand what is the way that intelligence agencies function? Is it possible to the head of an intelligence agency not to make direct orders of that kind where there's a distributed, you [/INST] know, the head of the intelligence agency would most likely give the order </s><s>[INST] even though it's compartmentalized. [/INST] Yeah. But but but not the head of State. </s><s>[INST] Maybe not the head of State. Although in the case this is the case in the United States as well but certainly is the case in Russia, there are close relationships between the head of the FsB and the G. R. U. And personal relationships not just even [/INST] the head of the FsB was now in jail. Mhm. </s><s>[INST] There's interesting details, especially coming out recently around the war in Ukraine. So let me actually ask about the war in Ukraine. What is your analysis of The war in Ukraine from 2014 to the full on invasion of Ukraine by Russia in 2022 and February 2022. What there's many questions who could ask one is, what are the sins of the government's involved? What are the sins of Russia? Ukraine? America china? Are those sins comparable? Who are the good guys and the bad guys? [/INST] That was more than one question. Now, let me just give you my the basics about this. Savvy observers saw this coming, there were very small minority because Vladimir Putin was pretty open about what he told the world. His mission was, was the re establishment of a strong Russia, the re establishment of something like the Russian empire. To unite all the Russian speaking uh people uh on the one country and the world ignored him. I mean, he was open. What was it was at a conference in in in in France, I believe when we when we set this out out in the open, uh and then what we had in the United States. We had wishful wishful thinking. You know, Obama had this reset with Russia. You know, we all get friendly. And then when when uh Putin invaded Crimea, we did nothing. So and it and it just escalated slowly but surely. Oh, it was pretty clear. And then they said was I think two years ago there was an essay published by uh Putin. Whether he wrote it or not doesn't matter, but that was also out in public where he was again quite clear what he was going to do now. How do you do this with force and and the the sins committed by the american government was that we ignored it. We weren't engaged in wishful thinking and we didn't stop it with sanctions before the shooting started </s><s>[INST] to push back. I don't think you're fully describing, you are describing the sins of the Russian government and Putin, I don't think you're fully describing the sins of the american government here because not only didn't you're doing, you're describing the miscalculation. So not only did they not pressure correctly with sanctions and so on. And and and clearly respond to the actual statements and the essays and the words spoken. [/INST] I know where you're going, but keep on. Yes, </s><s>[INST] but they also at the same time pressured pressured Russia. And they also as Putin himself said, there's a rat and they pushed the rat towards the corner by expanding NATO. [/INST] Mhm. </s><s>[INST] And uh [/INST] and arming Ukraine </s> |
<s>[INST] It's sad. I mean it says steve jobs leaving Apple by passing away sad, you know what the future of SpaceX and Tesla looks like without Elon musk is quite sad. It's very possible that those companies become something very different. They become something much more, you know like corporate and uh stale. Yeah, so maybe maybe most of progress is made. Their cycles may be newly almost comes along all those kinds of things. But it does seem that the american system of government has uh, built into it. The cycling. Yes, that makes it effective and it makes it last very long. It lasts a very long time, right? It continues to excel and lead the world. Sure, [/INST] sure. And let's hope it continues to know it's it's I mean we were into, you know, a third century and democracies on this scale rarely lasts that long. So that that's that's a point of pride. But it also means we need to be attentive to keep our house in order because it's not inevitable that this experiment continues </s><s>[INST] now. It's important to meditate on that. Actually. You mentioned that FDR franklin D Roosevelt is one of the great leaders in american history. Why is that [/INST] franklin Roosevelt had the power of empathy. No leader that I have ever studied or been around or spend any time reading about was able to connect with people who were so different from himself as franklin Roosevelt. He came from the most elite family. He never had to work for a paycheck in his life when he was president. He was still collecting an allowance from his mom. You couldn't be more elite than franklin Roosevelt. But he authentically connected. This was not propaganda. He was able to feel the pain and understand the lives of some of the most destitute americans in other parts of the country. </s><s>[INST] It's interesting. So through the one of the hardest economic periods of american history, he was able to feel the [/INST] pain. He was able to the number of immigrants. I read oral histories from or who have written themselves. Saul Bellow was one example the great novelist who talk about how as immigrants to the US all. Bell was a Russian, jewish immigrants. He said growing up in Chicago politicians were all trying to steal from us. I didn't think any of them cared until I heard FDR and I knew he spoke to me. Uh, and, and I think part of it was FDR really tried to understand people. That's the first thing was humble enough to try to do that. But second, he had a talent for that. And it's hard to know exactly what it was. But he had a talent for putting himself imagining himself in someone else's shoes, </s><s>[INST] what stands out to you as important. And he said he was, he went through the great Depression the so the new deal, which some people criticize some people see, I mean, it's funny to look at some of these policies and their long, um, the ripple effects. But at the time it's some of the most innovative policies in the history of America, you can say they're ultimately not good for America, but there nevertheless hold within them very rich and important lessons. But the new deal, obviously World War Two, uh, that entire process, is there something that stands out to you is um, particularly great moment that made FDR. [/INST] Yes, I think, uh, what FDR does from his 1st 100 days in office forward and this begins with his fireside chats is he helps americans to see that they're all in it together. And that's by creating hope and creating a sense of common suffering and common mission. It's not offering simple solutions. One of the lessons from FDR is if you want to bring people together, don't offer a simple solution because as soon as I offer a simple solution, I have people for it and against it. Don't do that explain the problem, frame the problem and then give people a mission. So Roosevelt's first radio address in March of 1933, the banking system is collapsing. We can't imagine it right? Banks were closing and you couldn't get your money out. Your life savings would be lost. Right? We can't imagine that happening in our world. Today, he comes on the radio, he takes five minutes to explain how banking works. Most people didn't understand how banking works, right? They don't actually hold your money in a fault. They landed out to someone else. And then he explains why if you go and take your money out of the bank and put it in your mattress, you're making it worse for yourself. He explains this, uh, and then he says, I don't, he doesn't, I don't have a solution. But here's what I'm gonna do. I'm gonna send in government officers to examine the banks and show you the books on the banks. And I want you to help me by going and putting your money back in the banks. We're all going to do this together. No simple solution, no ideological statement, but a sense of common mission. Let's go out and do this together. When you read as I have so many of these oral histories and memoirs for people who lived through that period. Many of them disagreed with some of his policies. Many of them thought he was too close to jews and they didn't like the fact had a woman in his cabinet and all that, but they felt he cared and they felt they were part of some common mission. And when they talk about their experience fighting in World War Two, whether in europe or Asia, it was that that prepared them. They knew what it meant to be an american when they were over there. Uh, so that, to me is a model of leadership and I think that's as possible today as it's ever been. </s> |
<s>[INST] Well, what I've noticed about Vladimir Putin and this is true about, for example, donald trump and all those kinds of um divisive figures that some for some reason people's opinion on the details of those people are very sticky. Once you decide this is a bad guy, there's like a black hole and people are not able to think like, one act at a time, you don't have to like that doesn't somehow justify this, that somehow doesn't uh remove all the evil things that are done. But you can analyze clearly each of the actions. And to me it is interesting to see how did this man rise through the ranks. Now, you're saying that to be a KGB agent, there's a lot of skills involved uh and perhaps raw um technical skill of spycraft is perhaps not related to the skill of raising, rising through the ranks. And you're saying that it's a politician, he was good at [/INST] lying and influencing uh that is something that that is significant as a significant talent and ability that an agent must have that helps you as a politician </s><s>[INST] continuing the kind of thread of the role of KGB in defining the the heart, soul and mind of Vladimir Putin. Let me return to your Ebay Usmanov, who was a soviet KGB agent that wrote a four step framework for ideological subversion on a national scale as practiced by the soviet union. So, the four steps are demoralization, destabilization crisis and normalization. He had a lot of other kind of systematic ways of describing this kind of stuff. So, can you speak to some of these ideas about the systematic large scale ideological subversion goals of the KGB is there truth to that kind of those ideas. [/INST] Yes, But I think I already sort of mentioned that, I think Desmond was a fraud. And I have I have again, can </s><s>[INST] you elaborate good [/INST] arguments? Let's put it this way. First of all, we we know that the KGB was involved in active measures, which is, you know, you can call it fake news, seating fake news into advert in the countries that are your adversaries. And and the Russians have been doing this lately by meddling in our election and and focusing on the left and the right fringe and influencing them to become more left and no more right. And uh and Vasili Mitrokhin has an as in one of his books, he has a whole chapter about active measures. Okay, so what he has to say about the department and I forgot what department that was, was the one department that was the least desirable for KGB agents because these were desk jobs for people who had to come up with fake stories that in countries where they didn't quite know too much about the country. Now there were some successes, like one of the two most famous successes uh, that I'm aware of is that uh uh the Canard that the AIDS virus was concocted in a CIA lab and there's a lot of people around the world believe that. And the other one was that uh j Edgar Hoover was a secret cross dressers cross dresser that that was that that is still known by a lot of americans who are of a certain age that this was the truth. But Metro can actually traces it back to a story that was placed in a sort of left wing but close to mainstream uh french magazine. And it was then taken taken up by uh more uh you know, larger newspapers and and well established papers. So so they had some successes. But this kind of a massive, well thought out campaign to destabilize the United States. I don't believe that the KGB was capable of doing that. Ma token seems to agree with me. I was trained, I would think, you know, I was one of the crown jewels of their agents. one would think that they used the best that they had to help me how to become an american and they didn't have a clue. So how do they, if you don't know how a country operates, how do you come up with this? This this this kind of a very detailed long term plan that's that's also timed two years this and one year that and all of that. Yes. </s><s>[INST] So we should, we should actually just clarify. So he has this whole idea, There's a 15-20 years I needed for demoralization um where you're Um you're basically infiltrating a country from a young people from a young age, manipulating their mind, your destabilizing then that's the second step that takes 2-5 years. You target the country's foreign relations, defense and economy, you create a crisis artificially and then you normalize it as as if it always was this way. So it's basically saying that the KGB is capable of at scale over many years manipulate an entire population of people. And this is kind of um there's a lot of people that believe in conspiracy theories that are amenable to this kind of idea. Now, my own experience is that there is in fact just a giant amount of incompetence and that this is something that's actually very difficult to pull off because it's incredibly, incredibly difficult to achieve this kind of manipulation. I think it would require, first of all, not much bureaucracy, not much slowing down. You have to have incredible in the modern world, digital systems that are able to do surveillance manipulation. There has to be a strategy that is carried out in secrecy across a huge number of people effectively. That also requires you hire the best people in the world. And I think it's difficult to execute on this kind of thing with if you can compartmentalize because there has to be great collaboration, there has to be a great where there's a unified vision, [/INST] coordination </s> |
<s>[INST] I think to me the important lesson is this is not a left, the right thing. [/INST] Not at all powerless. Yes. </s><s>[INST] And, but, and also the important lesson there, I think at least in the case of twitter in our society, it's a, it's a slippery slope. You don't get there overnight. You start, you start using those tools a little bit, a little bit to slow down this information to just a little bit that you start, uh, sending emails to each other a little bit and it becomes more and more um, you, you start forming justifications, you start getting a little more and more comfortable kind of talking about this stuff. Um, I think there are several ways to fight that one is having hard core integrity up front. So don't, don't even open the door, but I think realistically human nature is what it is. And so I think the only way is through transparency is this is why the nice I hate the fact that got politicized, I really hate that the right has have run with it. Like look the left is planning the rig elections and so on to me that it shouldn't be left or right shouldn't be about politics. It's that transparency is good. Other companies should do the same, facebook should do the same and in [/INST] fact that </s><s>[INST] transparency will protect facebook, it will protect google. Look like this is our situation, tell us what to do and we'll do our best. [/INST] I remember when I was writing the new right twitter's line was we're not gonna tell you guys what the metrics are by which we ban or censor people because then bad actors are going to navigate around them and it's like, what are you doing? Like just tell people in any establishment, what are the rules for which behavior is permissible? If I go to a store, if I return the sweater is a cash back, no refunds or if I get store credit, you know what I mean? So that they were having this place which is presented as a huge international space for public discourse and they're not telling you ahead of time. This is what we will tolerate. This is what we'll warn you about. This is what will kick you out overnight. That to me was crazy and outrageous. Um and I'm really pleased with, to what extent Ellen is like is being open with their policies and what I'm really want to commend him about is now I'm triggered because one of the things that he </s><s>[INST] took [/INST] over, he's like our first priority is getting with a child pornography and child </s><s>[INST] exploitation, right? [/INST] That was, he's like like racial slurs, homophobic slurs anti Semitic slurs, Yeah, yeah, that's cool kids kids getting harmed is number one and you know he fired the old task force because they weren't doing their job Eliza blue who you know you know she had been on this for a long time, but people who were victims of child pornography, child exploitation were emailing twitter being like these are my images, get them off and they're like too bad allowed on twitter, he starts trying to crack down on it, this is a very hard problem because these bad actors have mechanisms to evade uh you know being banned, they they want to get there for lack of a better term product out there. Forbes magazine who is an agent of the devil had a tweet and they tweeted this nine times, you know now that Elon's here twitter's child porn, nightmare has gotten much worse, they tweet this nine times I looked up, Anyone listeners can look up look at. Forbes and do a search, they never mentioned this problem before. So now that Ellen is doing something about it now it's a problem for, you know, it's a problem. Ellen is the problem. It's not the child porn that you guys had a problem with. And that to me is like, yeah, I understand that you think that Ellen is a bad guy because he's upset your apple cart. This isn't a political issue. This isn't a gotcha moment. This is all right. Here are some tips. We talked to 10 experts, digital experts and here are some techniques Mr Musk that you might want to take from us free of charge. That will help you solve this. That would be a great article. And </s><s>[INST] I just want to use this opportunity to say quite clearly and strongly that even though twitter and other parts of the internet, I are interpreting some of my statements to me and I'm right in this case meaning leaning right, right wing and in other cases leaning left, left wing. I'm not, I'm a political or at least I try to be in my thinking, take one issue at a time. I do take an opinion on each issue at a time, but I hate camps. I try to avoid political accounts in general. It just, it sucks that promoting transparency in this case or celebrating transparency um is somehow connected to being right wing. [/INST] No, it's being made into </s><s>[INST] supposed [/INST] euphemism for being right wing. </s><s>[INST] It's just it sucks uh it sucks even though I'm wearing a red suit and this is a very red themed conversation. Well [/INST] that's I mean the revolution was the color of blood. </s><s>[INST] I'm just gonna let it sit on that for a second. Okay. You mentioned New york Times bestseller list. You chose to self publish. Yes. And we just linger on that decision. What's what are the pros and cons of self publishing? [/INST] The cons are it is acceptable in our current business climate or cultural climate for corporate media outlets to pretend the book doesn't exist. So basically, and there's a reason for it. I can make the case to them pretty easily if someone's doing it themselves, who is this guy is some crackpot writing crazy stuff from his basement, right? It's a little different I think for me because I'm an established author uh c span gave me an hour on book tv for dear reader, I think I was the first one to get an hour and book tv for a book that I did myself. Um So there is space for that. Um It didn't go through a vetting process the way a book going through a corporate publisher did. So those are the minuses. The pros are uh I can drop it and publish it immediately. Uh If you go through corporate publisher, you have to wait a year. Um You can do what you can have the book you want instead of getting past the editor and some editors are very very good and there's a whole spectrum. Some of them not so good. Someone good </s> |
<s>[INST] actually test for that? [/INST] For imagination that led to the... That there was something internal going on, some kind of inner world, as people say, or you could say, it actually is an agent, it's making decisions, it has an internal representation. And whether you say that's experience or not is a different thing, but at least the feature that there's some abstraction it's doing, that's not obvious from looking at the physical </s><s>[INST] substrates. Do you think it's possible to do that kind of thing? One of the compelling things about the Turing test is that defining intelligence, defining any complicated concept as a thing, like observing it from the surface and not caring about what's going on deep inside, [/INST] because how do you know? That's the point. So the idea is exactly that. So what we're trying to do, the Turing test for goal-directedness is literally take some objects that clearly don't have any internal representation, grains of sand blowing on the beach or something, and a crab wandering around on the beach, and then generating an experiment where the experiment generates an entity that literally has no internal representation to sand. These are oil droplets, actually, what I've got in mind, a robot that makes oil droplets. But then what we want to try and do is train the oil droplets to be like crabs, give them an internal representation, give them the ability to integrate information from the environment, so they remember the past, are in the present, and can imagine a future. And in a very limited way, their kind of game engine, their limited simulation of the world, allows them to then make a decision. So then you would run a bunch of crabs, like over and over and over and over. </s><s>[INST] How many crabs, Lee? How many? Because you have to have a large number of crabs. What does your theory say? [/INST] Is there a mathematical? We're working on it. I mean, this is literally a crab limit. There's literally a... Excellent. There's literally a... </s><s>[INST] What's the herd limit? What's the herding cats have to do? Oh, that's random. Wait, it's cats in the title by Daniel Dennett, Herding Cats and the Free Will and Inflation. [/INST] What does herding cats mean? What does free will and inflation mean? I love this essay because it explained to me how I could live in a deterministic universe, but have free will, but have freedom. And also it helped me explain that time needed to be a real thing in this universe. So what basically Dan was saying here is like, how do these cats appear to just do what they want? And if you live in a deterministic universe, why do the cats do these things? Aren't they just isn't all obvious? And how does free will inflate the universe? And for me, I mean, probably I love the essay because my interpretation of the essay in assembly theory makes complete sense because you need an expanding universe in assembly theory to create novelty that you search for, that then when you find something interesting and you keep doing it because </s><s>[INST] it's cool and it gives you an advantage, then it appears in the past to be a goal. So what does in assembly theory, the expansion of the universe look like? What are we talking about? Why does the expansion universe give you more possibilities of novelty and cool [/INST] stuff? So for me, I don't think about the universe in terms of big bang in space. I think about in terms of the big memory expansion, that you only have the ability to store one bit of information, so then you can't do very much. So what the universe has been doing since forever, it's been creating more, it's been increasing the size of its RAM. Okay? So it's like 1 megabyte, 2 megabyte, 3 megabyte, 4 megabytes, all the way up. And so the more RAM you have, the more you can remember about the past, which allows you to do cooler things in the future. So if you can remember how to launch a rocket, then you might be able to imagine how to land a rocket and then relaunch, reland and carry on. And so you're able to </s><s>[INST] expand the space and remember the past, and so that's why I think it's very important. But not a perfect memory. It's an interesting question whether there's some forgetting that happens or might increase. Is the expansion of the forgetting at some point our accelerate [/INST] faster than remembering? I think that that's a very important thing that probably intelligence does and we're going to learn in machine learning about because you want machine learning right now or artificial right now doesn't have memory right, but you want the ability to, or not for, if you want to get to human-like consciousness, you need to have the ability, I suppose, to remember stuff and then to selectively forget stuff, so you can re-remember it and compress it. Arguably, the way that we come up with new physical laws, yeah, sorry, I think that there is a great deal to be gained from having the ability to remember things, but then when you forget them, you can then have a, you can basically do the simulation again and work out if you get to that compressed representation, so that it's in cycles. So cycles of remembering and forgetting are probably important, but there shouldn't be an excuse to have a universe with no memory in it. The universe is going to remember that it forgot, but just not tell you. No, no, it's all right. </s> |
<s>[INST] I still wonder what kind of man it takes, what kind of human being it takes to do that. Because I've been studying Vladimir Putin quite a bit and he's still, I believe he still has popular support that that's not fully manipulated because I know a lot of people in Russia and actually almost the entirety of my family in Russia are big supporters of Putin, everybody I talked to. So if that's not just like on social media, like the people that live in Russia seems to seem to support him. It feels like this will be in a George Washington way. Now would be the time that Putin, just like Yeltsin could relinquish power and thereby, in the eyes of Russians become in in like the long arc of history, be viewed as a great leader. You look at the economic growth of Russia, you look at the rescue from the collapse of the soviet union and Russia finding its footing and then relinquishing power in a way that that perhaps if Russia succeeds forms a truly democratic state, this would be how Putin can become one of the great leaders in Russian history, At least in the, in the, in the context of the 21st century, [/INST] I think there are two reasons why this is really hard for Putin and for others, one is the trappings of power are very seductive as you said before. They're corrupting. This is a real problem, right? If it's in the business context, I don't want to give up that private jet. Uh, if it's in Putin's context, it's billions of dollars every year that he's able to take for himself or give to his friends. It's not that he'll be poor if he leaves, he'll still be rich and he has billions of dollars stored away, but he won't be able to get the new billions. And so that's part of the trappings of power are, are a big deal. And then second in Putin's case in particular, he has to be worried about what happens next. Will he be tried? Well, someone, you know, try to come and arrest him with someone try to come and assassinate him. Um Washington recognized that leaving early LTD the corruption and limited the enemies that you made. And so it was a strategic choice. Putin is at this point, Brennan power too long. And this comes back to your core insight. It's a cliche, but it's true power corrupts. No one should have power for too long. This was one of the best insights. The founders of the United States had that power was to be held for a short time as a fiduciary responsibility, not as something you owned, right? This is the problem with monarchy with aristocracy that you own power, right? We don't own power. We were holding it in trust. </s><s>[INST] Yeah, There's uh there's some probably like very specific psychological study of how many years it takes for you to forget that you can't own power. That's right. There's, you know, that could be a much more rigorous discussion about the length of terms that are appropriate. But really There's amount like Stalin had power for 30 years. Like Putin is pushing those that many years already. Mhm There's a certain point where you forget the person you were before you took the power. That's right. You forget to be humble in the face of this responsibility and then there's no going back. That's right. That's how dictators are born. That's how the evil like authoritarians become evil or well, that's not used the word evil but uh counterproductive destructive to the, to the idea that they initially probably came to office with. [/INST] That's right. That's right. One of the core historical insights is people should move </s><s>[INST] jobs it supplies for Ceo is probably [/INST] absolutely they can go become Ceo somewhere else but don't stay Ceo one place too long. It's a problem with startups, right? The founder, you can have a billion founder and that founder doesn't want to let go. Yeah, it's the same issue </s><s>[INST] at the same time. I mean, this is where Elon musk and a few others like uh larry page and Sergey brin. That stayed for quite a long time and they actually were the beacon, they on their shoulders carried the dream of the company where everybody else doubted so, but that seems to be the exception right versus the rule. [/INST] And even Sergey for example, right has stepped back, right? He plays less of a day to day role and is not running google on the way. </s><s>[INST] The interesting thing is you step back in a quite tragic way from what I've seen, which is I, I think google's mission, an initial mission of making the world's information accessible to everybody is one of the most beautiful missions of any company in the history of the world. I think it's what google has done with a search engine and um, and other efforts that are similar like scanning a lot of books. Just incredible. It's similar to like Wikipedia. But what he said was that it's not the same company anymore. And I know maybe I'm reading too much into it because it's more maybe practically saying just the size of the company's much larger, the kind of leadership that square. But at the same time they changed the model from, you know, don't be evil too, is becoming corporatized and all those kinds of things. And it's sad. It um, [/INST] they're also recycles, right? History is about cycles right? There there there cycles to life. Their cycles to organizations. </s> |
<s>[INST] How do you take the language like Swift? Um you know, if we think about Gps but also a six, maybe we can dance back and forth between hardware and software. Uh you know, how do you design for these features to be able to program making a first class citizen to be able to do like Swift for tensorflow to be able to do machine learning on current hardware. But also future hardware like uh Cpus and all kinds of basics that I'm sure will be popping up more well. [/INST] So so a lot of this comes down to this whole idea of having the nuts and bolts underneath the covers that work really well. So you need, if you're talking to Tps you need, you know, M. L. I. R. X L. A. Or one of these compilers that talks to T. P. Us to build on top of. Okay, and if you're talking to circuits, you need to figure out how to lay down the transistors and how to organize and how to set up clocking and like all the domain problems that you get with circuits, then you have to decide how to explain to human, what is he? Why? Right. And if you do it right, that's a library problem. Not a language problem. And that works if you have a library or language which allows your library to write things that feel native in the language by implementing libraries, because then you can innovate in programming models without having to change your syntax. Again, like to invent new code formatting tools and like all the other things that language has come with and this, this gets really interesting. And so um, if you look at this space, the interesting thing, once you separate out syntax becomes what is that programming model? And so do you want the cuda style? I write one program and it runs many places. The do you want the implicitly parallel model? How do you reason about that? How do you give developers, you know, chip architects, the the ability to express their intent and that comes into this whole design question of how do you detect bugs quickly? So you don't have to take out a chip to find out what's wrong ideally. Right? How do you And and you know, this is a spectrum, How do you make it so that people feel productive? So their turnaround time is very quick, all these things are really hard problems and in this world, I think that not a lot of effort has been put into that design problem and thinking about the layering in other pieces. </s><s>[INST] Well, you've on the top of concurrency, you've written the swift concurrency manifest, I think it's it's kinda interesting anything that uh has the word manifesto and is very interesting, can you summarize the key ideas of uh each of the five parts you've written about? [/INST] So what is the manifesto? How about we start there? Uh So in the swift community we have this problem which is, on the one hand, you want to have relatively small proposals that you can kind of fit in your head, you can understand the details of a very fine grained level that move the world forward, but then you also have these big arcs. Ok. And often when you're working on something that is a big arc, but you're tackling in small pieces, you have this question of how do I know I'm not doing a random </s><s>[INST] walk, where [/INST] are we going? Like how does this add up furthermore, when you start that first, the first small step, what terminology to use? How do we think about what is better and worse than the space? What are the principles, what are we trying to achieve? And so what a manifesto in the swift community does it starts to say, hey, well let's step back from the details of everything. Let's paint a broad picture to talk about how, what we're trying to achieve. Let's give an example design point. Let's try to paint the big picture so that then we can zero in on the individual steps and make sure that we're making good progress. And so the swift concurrency manifesto is something I wrote three years ago. It's been a while maybe, maybe more trying to do that for for swift and concurrency. It starts with some fairly simple things like making the observation that when you have multiple different computers or multiple different threads are communicating, it's best for them to be asynchronous. Right? And so you need things to be able to run separately and then communicate with each other and this means a synchrony and this means that you need a way to modeling asynchronous communication. Many languages have features like this. A single weight is a popular one. And so that's what I think is very likely in swift. Um but as you start building this tower of abstractions, it's not just about how do you write this, you then reach into the, how do you get memory safety because you want correctness, you want debug ability, insanity for developers and how do you get uh that memory safety into um into the language. So if you take a language like go or C or any of these languages, you get what's called a race condition when two different threads or go routines or whatever, touch the same point in memory, right. This is a huge like, maddening problem to the bug because it's not reproducible generally. And so there's tools, there's a whole ecosystem of solutions that built up around this, but it's it's a huge problem when you're writing concurrent code. And so what swift, this whole value semantics thing is really powerful there because it turns out that math and copies actually work even in concurrent worlds. And so um you get a lot of safety just out of the box, but there are also some hard problems and it talks about some of that um when you start building up to the next level up and you start talking beyond memory safety, you have to talk about what is the programmer model? How does the human think about this? So a developer that's trying to build a program think about this and it proposes a really old model with the new spin called actors. Actors are about saying we have islands of single threaded nous logically. So you write something that feels like it's one programming, one program running in a unit and then it communicates a synchronously with other other things. And so making that expressive and natural feel good. Be the first thing you reach for and being safe by default is a big part of the design of that proposal. When you start going beyond that. Now you start to say, cool, well these things that communicate synchronously, they don't have to share a memory. Well if they don't have to share memory and they're sending messages to each other, why do they have to be in the same process? These things should be able to be in different processes on your machine and why just processes? Well why not different machines? So now you have a very nice gradual transition towards distributed programming. And of course when you start talking about the big the big future, the manifesto doesn't go into it. But accelerators are I think things you talk to a synchronously by sending messages to them. How do you program those? That that gets very interesting. That's not that's not in the proposal. </s> |
<s>[INST] you going? [/INST] Yeah. Right. Yeah. And one of them is winning. And and that's affecting, you know, the next sentence and so long, uh, there was this book, Machine Intelligence </s><s>[INST] on Intelligence [/INST] on Intelligence. Yeah. Bill Atkinson was was was a total devotee of that </s><s>[INST] book. I like whether it's consciousness or something else. I like the storytelling part that we it feels like, uh, first humans, it feels like there's a concrete, it's almost like literally programming. I don't know what the programming going on on the inside, but I'm getting a nice story here about what happened and it feels like I'm in control and I'm getting a nice clear story, but it's also possible there's a computation going on that's really messy. There's a bunch of different competing ideas. And in the end, it's just kind of generates a story for you to uh, a consistent story for you to believe. And that makes it all nice. [/INST] Yeah. And so I prefer to talk about things that I have some expertise, and then then things which which I'm only a uh on the sideline. </s><s>[INST] So there's a tricky thing, I don't know if you have any expertise in this, you might be a little bit on the sideline would be interesting to ask though, what are your thoughts on cellular automata in the Game of Life? Have you ever played with those kind of little games? [/INST] I think the Game of Life is his wonderful. And I uh and and shows all kinds of stuff about how things can evolve without the creator understanding anything more than the power of learning in a way. But to me, the most of important thing about the Game of Life is, is how it focus for me, what what it meant to have free will or not, because the game of Life is obviously totally deterministic. I find it hard to believe that anybody who's ever had Children cannot believe in free will. Right. On the other hand, this makes it crystal clear. Uh john Conway said uh he wondered whether it was immoral to shut the computer off after he got into a particularly interesting play of the Game of life. Um, wow, </s><s>[INST] Yeah. So there is to me the reason I love the Game of Life, it is exactly as you said, a clear illustration that from simple initial conditions with simple rules, you know, exactly how the system is operating is deterministic. And yet if you let yourself uh, if you allow yourself to lose that knowledge a little bit enough to see the bigger organisms that emerge and then all of a sudden they seem conscious. They seem not conscious, but living if [/INST] the universe is finite, we're all living in the game of life to slow down. I mean, it's sped up a lot. Uh </s><s>[INST] but do you think technically some of the ideas that you used for analysis of algorithms can be used to analyse the game of life? Can we make sense of it? Or is it too weird? And [/INST] yeah, I mean, I've got I've got a dozen exercises in volume for Fast ical six that actually worked rather well for that purpose. Bill gossipers came up with the algorithm that allows allowed Golly Too uh you know, to run thousands and thousands of times faster. You know the website called Golly T. O. L. L. Y. </s><s>[INST] Just, it stimulates the sell your item in a game of life. [/INST] Yeah. You got you got to check it out. </s><s>[INST] Can I ask you about john Conway [/INST] yes. In fact, I'm just reading now the the issue of mathematical intelligence or that came in last last week. And the whole issue devoted to uh you know, remembrance of him. </s><s>[INST] Did you know him? [/INST] I slept overnight in his house several times. I yeah, </s><s>[INST] he recently passed away. [/INST] You know, he he died a year ago. May I think it was of covid, </s><s>[INST] what are you, what are some memories of him of his work that stand out for? You did uh on a technical level that any of his work inspire you? On a personal level that did he himself inspire you in some way [/INST] and you know, absolutely to all of those things. But let's see, when did I first meet him? I guess I first met him at Oxford in 1967 when I was </s><s>[INST] okay. That's a long time ago. [/INST] Yeah. You were -20 years old or something, I don't know, 1967. But but uh, there was a conference where uh, I think I've spoken, I was speaking about something that no one has the canoes. Bendix algorithm now, but but he he gave famous talk about knots and and at the, and I didn't know at the time, but anyway, that talk now uh the stories of thousands and thousands of papers since since then. Uh and it was he was reported on something that he had done in high school, uh you know, almost 10 years earlier, uh, before this conference, but he never published it. And anti climax to stalk by building some not. So you have these these little plastic things that you that you can stick together. Uh It's something like lego but easier. Uh, so he made a whole bunch of, not in front of the audience and so on. And then disassembled. It was dramatic lecture before. He had learned how to give even more dramatic lectures later. So, All right. And </s><s>[INST] were you at that [/INST] lecture? And I was there, Yeah, because I had, you know, I was at the same conference um for some reason I was I happened to be in in Calgary uh at the same day that he was visiting Calgary And it was the spring of 72, I'm pretty sure. And and we had lunch together and he wrote down during the lunch on a napkin, all of the facts about what he called numbers. Uh and I and uh and he covered the napkin with the theorems about his his idea of numbers. And I thought it was incredibly beautiful. Um and and later in 1972, my sabbatical year began and I went to Norway and in december of that year, uh in the middle of the night, the thought came to me, you know, Conway's theory about numbers would be a great think teach students how to invent research and what the joys are of research. And and and so I said, and I had also read a book and dialogue, Bye bye Alfred renew. Uh We're kind of a Socratic thing where they're two characters were talking to each other about mathematics. And so I and so at the end in the morning I woke up my wife and said, Jill, I think I want to write a book about Conway's theory. And um you know, you know, I'm supposed to be right in the heart of computer programming, doing all this other stuff. But I got, but I really want to write this other book. And so we made this plan, but I said I thought I could write it in a week and we made the plan then. So in january, I rented a room in a hotel in downtown Oslo were on sabbatical in Norway uh and I ran at the hotel in downtown Oslo and did nothing else except right up Conway's theory. And I changed the name to surreal numbers. That so this book is now published a surreal number. And, and you know, we figured out, we always wondered what would you like to have a fair enough hotel rooms. So, so we figured out that she would visit me twice during the week. Things like this. You know, we would, you know, try to sneak in. This was hotel was run by a mission organization. These ladies were probably very strict, but anyway, so, so uh in the wild </s> |
<s>[INST] Is there the scratches do those come from? Are those programmed or they failures? Meaning is it? So if it's by design, then there's like a encoded timeline schedule that the body is just on purpose degrading the whole thing. And then there is the just the wear and tear of like the scratches in the disc that happened through time. Which which one is it? That's the source of aging. [/INST] It's more akin to wear and tear. There isn't a program getting back to evolution. There's no selection for aging were not designed to age. We just live as long as we need to. And then we're at the whim of entropy, basically. 2nd of thermodynamics stuff falls apart. We live a bit longer than age 40 Only because there are robust resilient systems, but eventually they fail as well. Current limit to the human lifespan where they completely fail at 122. But it's and I but I don't like to think of it as wear and tear because there's there's two aspects to it. There's a system that's built to keep us alive when we're young but actually goes, comes back to bite us as we get older and we call this this issue antagonistic player trophy, what's good for you when you're young can cause problems when you're older. So we've been looking what is the cause of the main causes of the noise. And we've come, we found two of them definitively. The first one is broken chromosomes. When a chromosome breaks, the cell has to panic because that's either going to cause a cancer or kill the cell. There's only two outcomes. It's pretty much a problem. And so what the cell does is it reorganizes the epic genome in a massive way. What that leads to is think of it as a tennis match or a ping pong game. The proteins of the balls and they now leave where they should be, which is regulating the genes that make the cell type, whatever it is. And after they have a dual function actually go to the break the chromosomal break and fix that. And then they come back. You might ask, well, why is it set up that way? Well, it's a beautiful system that coordinates gene expression, the control systems with the repair you want them coordinated problem is as we get older, this ping pong game, some of the balls get lost. They don't come back to where they originally started. Uh And that's what we think is the main noise for aging. And we've also, the other cause of aging that we found is self stress, we damaged nerves and they age rapidly. So that's the other issue. There's probably others um smoking chemicals for example, we know accelerates biological age pretty dramatically. Well, the question is, can you slow that down or can you reset them to get those ping pong balls to go back to where they originally started in the game? And we think we found a way to do that. </s><s>[INST] Can you give me hints whose fault is it? And the ball is not coming back. Is that the proteins themselves? Like, are they, are they starting again? I've been obsessed with the protein folding problem. From the perspective. Is that the proteins or is it something else? [/INST] Well, we know who hits the balls um and recruits them so that the break ah is recognized by proteins who send out a signal threw phosphor relation is typical way cells talk to other proteins and that recruits those repair factors. Those ping pong balls to the brake. So the cells actively doing this to try and help itself. But we don't know who's to blame for them not coming back. That could just be a flaw in the quote unquote design. I don't think that there's something saying, well, of you you balls proteins never go back. I just think it's hard to reset a system that's constantly changing. We have in our bodies close to a trillion DNA breaks every day and imagine that over 80 years, what damage that does to RFP genomic information. Now we know that this is well, we never know anything in biology, but we have strong evidence that this is true because we can mess with animals. We can create DNA breaks and tickle them with a few breaks. Maybe raise it by three fold over background levels of normal breakage. And if we're right, those mice should get old. Mhm. And they do. We can actually, we've we've created these breaks in a way that's tight tradable. We can like a rheostat, we can send it to 11. I drove my tesla here. I'm a big fan of spinal tap to going to 11. If we go to 11 weekend, make a mouse old in a matter of months. We prefer to go to a level of about four and it gets old in 10 months but it's definitely old. It's got all of the hallmarks of aging. It's got diseases. It looks old. It's skin is old, it's got gray hair. But importantly, we can now measure age by looking at the scratches. We can look at the evidence and we can measure it and use machine learning to give us a number And those mice are 50% older than normal. </s><s>[INST] So you can replicate the aging process in a controlled way. You can no, I mean in a way they I mean you could accelerate it in a controlled way and measure how much exactly it's aging. And that gives you step one of a two step process to when you can then figure out what, how can we reverse this? [/INST] And now we're reversing those mice. </s> |
<s>[INST] Can I ask you about some other weird particles that make up our universe? What are exxons? And what is the strong cp problem? [/INST] Okay, so uh let me start with what the strong cp problem is. First of all, we'll charge C is charged conjugation, which is the Yeah, transformation. The notional transformation if you like, that changes all particles into their anti particles. And uh the concept of seek symmetry, charge conjugation symmetry is that if you do that uh you find the same laws that would work. So the laws are symmetric. If the behavior that particles exhibit is the same as the part as the behavior you get with all their anti particles. Uh then P is parody, which is also called spatial inversion. It's basically looking at a mirror universe and saying that the laws that are obeyed in a mirror universe when you look at the mirror images obey the same laws as the as the sources of their images. There's no way of telling left from right, for instance, that the laws don't distinguish between left and right. Uh Now In the mid-20th century people discovered that both of those are not quite true that really the the equation that that the mirror universe, the universe, that's that you see in a mirror is not going to obey the same laws as the as the the universe that that that we actually exhibit. Uh and interpret you could, you would be able to tell if you did the right kind of experiments, which was the mirror and which was the real thing. Uh anyway, that </s><s>[INST] just a parody and they show that doesn't necessarily hold, [/INST] it doesn't quite hold and that that that and examining what the exceptions are turned out to be to lead to lower kinds of insight about the nature of fundamental interactions, especially properties of neutrinos and the weak interaction. It's a long story but it's a very you </s><s>[INST] just define the sea and the p the congregation, the church congregation [/INST] that I've done that I want to, what's the problem of them off? Because it's easier to talk about T. Which is time reversal symmetry. We have very good reasons to think Cpt is a an accurate symmetry of nature. It's on the same level as relativity and quantum mechanics basically. So that better be true. Uh </s><s>[INST] So our metric when you when you do conjugation, parody and time [/INST] and time and space reversal, if you do all three then you get the same physical consequences now. So but that means that CP is is equivalent to T. But what what's observed in the world is that T. Is not quite an accurate symmetry of nature either. So most phenomena of at the fundamental level. So interactions among elementary particles and the basic gravitational contraction uh If you ran them backwards in time you get the same laws. So if again going back unless this time we don't talk about a mirror but we talk about a movie, if you take a movie and then run it backwards, that's the time reversal. It's </s><s>[INST] good to think about a mirror in time. [/INST] It's like a mirror in time. If you uh if you run run the movie backwards, it would look very strange if you were looking at complicated objects and uh you know, a charlie Chaplin movie or whatever. They would look very strange if you ran it backwards in time. But at the level of basic interactions, if you were able to look at the atoms and the quarks involved, they would obey the same laws. They do a very good approximation, but not exactly. </s><s>[INST] So, you know exactly. That means you could tell, you [/INST] could tell, but you have to do very, very subtle experiments with that high energy accelerators to take a movie that looked different when you ran it backwards. Uh This was a discovery by uh uh two great physicist named Cronin and Jim Cronin and Val Fitch In the uh in the mid 1960s, previous to that, over all the centuries of development of physics with all his precise laws, they did seem to have this gratuitous property. Yeah, that they look the same. If you run the equations backwards, it's kind of an embarrassing property actually, because life isn't like that. So empirical reality does not have this imagery in any obvious way. And yet the laws did. </s><s>[INST] It's almost like the laws of physics are missing something fundamental about life if it holds that property. Right. Well, that's that's the embarrassing nature. It's [/INST] it's yeah, it's </s><s>[INST] embarrassing. Well, people, [/INST] I worked hard at what this is a problem that's thought to belong to the foundations of statistical mechanics or the foundations of thermodynamics to understand how behavior which is grossly not symmetric with respect to reversing the direction of time in large objects, how that can emerge from equations which are symmetric with respect to changing the direction of time to a very good approximation. And that's that's still an interesting endeavor. That's that's that's interested in actually, it's an exciting frontier of physics now to sort of explore the boundary between when that's true and when it's not true, when you get to smaller objects, uh and exceptions, like time crystals. Uh </s><s>[INST] I definitely have to ask about time christmas in a second here. But so the cp problem and t so there's lost all of these [/INST] are in danger of infinite regress, but we'll convert soon. </s><s>[INST] So can't possibly be turtles all the way down. We're going to get to the bottom turtle. So, [/INST] so so it became so it got to be a real. I mean, it's a really puzzling thing why the laws should have this very odd property that we don't need. And in fact, it's kind of an embarrassment in addressing empirical reality, but it seemed to be almost seemed to be exactly true for a long time. And then almost true. And in way almost true is even more disturbing than exactly true because exactly true. It could have been just a fundamental feature of the world. And, you know, at some level, you just have to take it as it is. And if it's if it's a beautiful, easily articulate herbal regularity, you could say that, okay, that's that's fine as a fundamental law of nature. But to say that approximately true, but not exactly, that's that's not that's weird. So uh and then so there was great progress in uh the late part of the 20th century uh in getting to an understanding of fundamental interactions in general that shed light on this issue. Uh it turns out that the basic principles of relativity and quantum mechanics plus the kind of high degree of symmetry that we found the so called gait symmetry that characterizes the fundamental interactions when you put all that together. It's a very, very constraining framework. And it has some indirect consequences because the possible interactions are so constrained. And one of the indirect consequences is that the possibilities for violating the symmetry between forwards and backwards in time are very limited. They're basically only two and one of them occurs and leads to a very rich theory that explains the Cronin fish experiment. And a lot of things that have been done subsequently has been used to make all kinds of successful predictions. So, that's that's turned out to be a very rich interaction. It's esoteric and the effects are only show up at accelerators and are small and so much that they might have been very important in the early universe and lead to be connected to the asymmetry between matter and anti matter in the present universe. And so, but that's a that's another digression. The the point is that uh that was fine. That was a triumph to say that there was one possible kind of interaction that would violate time reversal cemetery. And sure enough, there it is. And but the other kind doesn't occur. So we still got a problem. Why doesn't it occur? Uh but so we're close to really finally understand this profound gratuitous feature of the world that is almost but not quite symmetric under reversing the direction of time, but not quite there. And uh, to get to understand that last bit is a challenging frontier of physics today. Uh and we have a promising proposal for how it works, which is a kind of theory of evolution. So there is this possible interaction which we call a coupling and there's a numerical quantity that tells us how strong that is. And traditionally in physics we think of these kinds of numerical quantities as constants of nature that you just have to put them in right. That from experiment, they have a certain value And that that's it. And you know, who am I to question what I've gotten doing? They just, well, they seem to be just constant. Uh but in this case it's been fruitful to think and work out a theory where that the strength of interaction is actually not a constant. It's a fun, it's a field, it's a it's a fields are the fundamental ingredients of modern physics. Like there's an electron field. There's a photon field which is also called the electromagnetic field. And so every all of these particles are manifestations of different fields. And uh, there could be a field, something that depends on space and time. So a dynamical entity instead of just a constant here. And uh if you do things in a nice way, that's very symmetric, very much suggested aesthetically by the theory. Uh but the theory we do have, then you find that you get a field which as it evolves from the early universe, settles down to a value That's just right to make the laws very nearly exact invariants or symmetric with respect to reversal of time </s> |
<s>[INST] timing, [/INST] timing charm the care he takes, So he is as lovable, but he doesn't believe himself to be? And that manifests itself in areas that drive us all crazy from time to time, </s><s>[INST] which makes it for a very pleasant listening experience. Okay, so what about the two Davids? David sacks and David friedberg [/INST] David sacks is the one that I would say I have the most emotional connection with. Um he and I can go a year without talking and then we'll talk for four hours straight and then we know where we are and we have this ability to pick up and have a level of intimacy with each other. And I think that's just because I've known David for so long now, um, that I find really comforting and then Freeburg is this person who I think similar to me had a very turbulent upbringing, has fought through it to build an incredible life for himself. And I have this enormous respect for his journey. I don't particularly care about his outcomes to be honest, but I just have, I look at that guy and I think and he did it and so if I didn't do it, um, I would be glad that he did it if it makes any sense. Um, and you can see that he um feels like his entire responsibility is really around his kids and just kind of like give a better counterfactual and, and you know, sometimes I think he gets that right and wrong. Um, but he's a very special human being that way </s><s>[INST] on that show. The two of you have a very kind of like from a geopolitics perspective, I don't know, there's just a very uh effective way to think deeply about the world, the big picture of the world. [/INST] He's a very systems level thinker, </s><s>[INST] very, very, [/INST] absolutely, very </s><s>[INST] systems level. So [/INST] I'm very rooted in, you know, a broad body of knowledge, which I have a tremendous respect for. He brings all these things in sacks is incredible because he has this unbelievable understanding of things, but it has a core nucleus. So Friedberg can just basically abstract a whole bunch of systems and talk about it. Um, I tend to be more like that where I try to kind of find it to be more of a puzzle sacks is more like anchored in you know, a philosophical and historical context as the answer and he starts there, but he gets to these profound understandings of systems as well </s><s>[INST] on the podcast in life. You guys hold to your opinion pretty strong. What's uh what's the secret to being able to argue passionately with friends. So hold your position but also not murder each other, which you guys seem to come close to? [/INST] I think it's like strong opinions weakly held. So I think that's the </s><s>[INST] way it is that is that a haiku or is that, can you explain that please? [/INST] Like you know, like look today, you and I what have we, we steal man, like the two sides of three different things. Um now you could be confused and think I believe in those things, I believe that it's important to be able to intellectually traverse there whether I believe in it or not and like steel man, not like that's a really, </s><s>[INST] but we intro those things by saying let us steal man this position. Sometimes you guys skip the uh [/INST] you're right, we we edit those things out and sometimes we'll sit on either side and we'll just kind of bad things back and forth just to see what the other person thinks. </s><s>[INST] So that's how like as fans we should listen to that sometimes like so sometimes because because you hold a strong opinion sometimes like for example the cost of energy going to zero is [/INST] that </s><s>[INST] like what's the degree of certainty on that? Is this kind of like you really taking a prediction of how the world will unroll and if it does this benefit a huge amount of companies and people that will believe that idea. So you really you you like you spend a few days, a few weeks with that idea, [/INST] I've been spending two years with that idea and that idea has manifested into um many pages and pages of more and more branches of a tree. But it started with that idea. So if you think about this tree, this logical tree that I built, I would consider it more of a mosaic and at the at the base or root however you want to talk about it is this idea the incremental cost of energy goes to zero. How does it manifest? And so I talked about one traversal which is the competition of households versus utilities. But if even some of that comes to pass, we're going to see a bunch of other implications from a regulatory and technology perspective if some of those come to pass. So I've tried to think think sort of this, you know, 678 hops forward. And I have some like to use the chess analogy. I have a bunch of short lines which I think can work. Um, and I've started to test those by making investments Tens of millions over here, 200 millions over there. But it's a distribution based on how probabilistic I think this these outcomes are and how downside protected I can be and how much I will learn how many mistakes I can make, you know, etc. And then very quickly over the next two years some of those things will happen or not happen and I will rapidly re underwrite and I'll rewrite that tree and then I'll get some more data, I'll make some more investments and I'll rapidly re underwrite. So you know, in order for me to get to this tree, maybe you can ask how did I get there? It was a complete accident. The way that had happened was I have a friend of mine who works at a great organization called fortress. His name is Drew mcknight and he called me one day and he said, Hey, I'm doing the deal. Will you anchor it? We're going public and it's a rare earth mining company. And I said Drew like </s> |
<s>[INST] How do you feel about the possibility if consciousness so naturally does come with A G. I. Systems that we are just a step in the evolution. That will be just something a blimp on the record will be studied in books by the G systems centuries from now. [/INST] I mean, I think I'd probably be okay with that, especially if somehow humans are continuous with a G. I mean, I think something like this is inevitable, at the very least, humans are going to be transformed. We're gonna be augmented by technology that's already happening in all kinds of ways. We're going to be transformed by technology where our brains are going to be uploaded and computational enhanced and eventually that line between what's human and what's a, well, today, I may be kind of hard to hard to draw, How much does it matter, for example that some future being 1000 years from now that somehow descended from us actually still has biology, I think would be nice if you kind of point to its cognitive system to point to some parts that had some roots in us and chase a trace a continuous line there. That would be selfishly nice for me to think that okay, I'm connected to this thread line through the future of the world, if it turns out okay, there's a jump there. They found a better way to design cognitive systems. They designed a wholly new kind of thing. And the only line is some causal chain of designing and systems that design better systems, is that so much worse? I don't know, we're still at least part of a causal chain of uh of design. And yes, they're not humans, but still there are successes. I mean, ultimately, I think it's probably inevitable that something like that will happen. At least we were, you know, at least we were part of the process. It would be nice if they still cared enough about us to um you know, maybe to engage with our arguments. But I'm really hoping that the G. I. S. Are going to solve all the problems of philosophy. They'll come back and read all this All this crap work for the 20th and 21st century hard problem of consciousness and here is why they got it wrong. And so if that happened then I really feel like I was part of at least a intellectual process over centuries and that would be kind of cool. </s><s>[INST] I'm pretty sure they would clone or they would recreate david Chalmers and and for the fun of it sort of bring back um other [/INST] philosophy card </s><s>[INST] Carton and just put them in a room and just watch it would be a netflix of the future show where you bring philosophy from different human 100% human philosophers from previous generations, put them in a room and see them. [/INST] I am totally I'm totally up for that simulators A. G. I. S of the Future. If you're watching this podcast do that I would like to be recreated and who would who would be with </s><s>[INST] the cart would be the first who if you if you could hang out as part of such a tv show with a philosopher that's no longer with us from long ago, who would who would you choose? [/INST] They would have to be right up there. Oh actually a couple of months ago I got to have a conversation with delicate an actor who's actually a philosopher came out on stage playing delicate. I didn't know this was gonna happen and I just after I gave a talk a bit of a serial, my ideas were crap and I'll drive from him and so I made a long, we had a long argument, this was great but I would love to see what the cart would think about. Ai for example in the modern neuroscience and so I suspect not too much would surprise him but yeah but yeah William James um you know for psychologists of consciousness I think James is probably the it was probably the richest but there are a manual cars. You know, I never really understood when he was up to. If I got to actually uh talked to him about some of this hair was Princess Elizabeth who talked with the cart and who really uh you know, got the problems of how the carts, ideas of non physical mind interacting with the with the physical body couldn't really work. She's been kind of, most philosophers think she's been proved right. So maybe put me in a room with a card and Princess Elizabeth and we can all argue it out </s><s>[INST] what kind of future. So we talked about with zombies a concerning future, but what kind of future excites you? What do you think if we look forward sort of we're at the very early stages of understanding consciousness and we're now at the early stages of being able to engineer complex interesting systems that have degrees of intelligence and maybe one day we'll have degrees of consciousness maybe be able to upload brains, all those possibilities. Virtual reality. What is there a particular aspect of this future world that just excites you. [/INST] I think there are lots of different aspects. I mean frankly I wanted to hurry up and half was like yeah we've had some progress lately and ai and VR but in the grand scheme of things it's still kind of slow. The changes are not yet transformative. And you know, I'm in my fifties, I've only got so long left, I'd like I'd like to see really serious ai in my lifetime and really serious virtual world because yeah, once people I would like to be able to hang out in a virtual reality which is richer than than this reality to really get to inhabit fundamentally different kinds of space is well, I would very much like to be able to upload my mind onto a onto a computer so maybe I don't have to die. Um If this is maybe gradually replaced my neurons with silicon chips and inhabit I can feel selfishly, that would be that would be wonderful. I suspect I'm not gonna quite get there in a in my lifetime. But once that's possible then you've got the possibility of transforming your consciousness in remarkable ways, augmenting it, enhancing it. </s> |
<s>[INST] I [/INST] love them. So I love numbers also. So they have all these different categories. Like you can see how long a wall that has held a Bitcoin right? Or how many Bitcoins are in a certain wallet. And so what they've seen during the downturn writes april it kind of piqued and went down is that the whales are still buying the whales. People of 1000 more are still buying, they've said the main group of sellers is the ones who held it from 0 to 3 months. So like they don't have money, they bought it. So they thought I was going up and I was like, oh shoot, I gotta sell it, right? Whereas anyone's head out for a long time is generally still holding </s><s>[INST] on. That's interesting. That's a good indicator, right? For the whole space. Yeah. Well, let me ask you for some advice. You've been through one heck of a career, one heck of a life. What advice would you give to a young person today? [/INST] Well in wrestling, I think wrestling is really a microcosm of what your life is going to be. And that's why one of things that stress the kids is like, if we can go through this now and you figure I have a couple of days, you're struggling certain things if you can figure out this now and wrestling to me a lot better figure out now and get over this mental hump than when you're 32 you have two kids and right, and your job is not going well, it's gonna be a lot worse, can be a lot more painful then let's let's figure it out now. So a lot of these things, a lot these lessons we can learn from wrestling, whether it's persistence and perseverance or work ethic or, you know, I said, russell show up on time and they work hard, right, these things that we can learn these things at an early age, those are generally, those characteristics, characteristics will generally carry on throughout our life and those are the things they're gonna make us really successful. So, um, you know, I would say find a great coach, someone who's going to spend a lot of time and put a lot of time in you and make sure they have a lot of wisdom and steal all the wisdom that you can from them. And then though, if you can be successful at one thing, generally, whatever that recipe was that took you to be successful with that apply to everything else, right? Applied to the rest of your life, applied to getting a wife that you enjoy, apply it to living in a place you want to live doing a job you want to do right there. There's so many possibilities and you just have to be bold enough to go take those chances. </s><s>[INST] It's interesting because like early on in life is when you have much more time, people don't realize this time to learn the lessons. Like somehow later in life you get busier responsibilities and all that kind of stuff like high schools and magical time [/INST] in college. College. </s><s>[INST] Yeah, there's so much [/INST] time to learn. You can have kids yet. </s><s>[INST] I don't have kids still fills up well no, I'm on purpose and I did something that many people don't seem to be able to do. I walked away from a lot of responsibilities just by saying goodbye, okay. But like, but you know, meetings like there, everybody around me at M. I. T. Was like meetings fill the day and they have more projects and you do a great job and you become successful and then the more meetings for the day and more responsibilities as opposed to like wait a minute. Do I want to be involved in all these things? Uh and instead do I want to find one or two things to really focus on? That's what I choose. But like that becomes harder and harder and harder as you get older. [/INST] You know I mean I'm sure and and also the more success you have, you become sought after other places too. I'm sure that's happening with </s><s>[INST] you and it's hard to say keep saying no no no no it's hard. Yeah you're known for roasting people [/INST] with </s><s>[INST] with a single boom roasted lines. So any ideas maybe you want to mention malice but any ideas come to mind when you look at me [/INST] man, I do you know what uh if I was gonna boom roast someone, I would want to kind of like research their career and dissect them and figure out their biggest next to the core. And I did, I didn't have that nose with you. I figured out I got a general sense of okay, he's really successfully, super sharp. He's really interested in some really interesting things. I will have a great conversation but I had no intention to roast you. </s><s>[INST] Yeah. There you go. What about malice? You had dinner with him last night [/INST] for him. Oh man. How'd </s><s>[INST] you get to know him by the way? Uh huh. [/INST] Just twitter most magical place in the world right? I always tell me it's a great source of information if you know how to use it. Um hm. </s><s>[INST] He's insane on twitter [/INST] actually. He had to un follow him on twitter because he was too, too much, there's too much, that was too much. It fills up like I wanna be able to consume the content so if I want to see something says I can go to his page right? But it's just too much for my timeline. I want to be able to consume who I follow. So I try to not follow a lot of people cause I wanna be able to consume them. Um And he was he was he was too much. He he fights the trolls which </s> |
<s>[INST] Okay, so the building engineering, the Neocortex in itself is just creating an intelligent prediction system, modeling system, sorry, modeling system. Uh you can use it to make predictions. And um but you can also put it inside a thing that's actually acting in this world, [/INST] you have to put it inside something. It's again, think of the map analogy right? Map on its own doesn't do anything right, It's just an earth, it's just that you can learn, but it's just so we have to embed it somehow and something to do something. </s><s>[INST] So so what's your intuition here? You had, you had a conversation with Sam Harris recently that was sort of um you've had a bit of a disagreement and you're sticking on this point, you know, Elon musk Stuart, russell kinda have a worry, existential threats of ai what's your intuition? Why if we engineer increasingly intelligent neocortex type of system in the computer, why that shouldn't be a thing that we [/INST] it was used the word intuition and Sam Harris used the word intuition to and and we didn't use that intuition that were immediately stopped and said that's the crux of the problem. He's using intuition. I'm not speaking about my intuition, I'm speaking with something, I understand something, I'm going to build something. I am building something. I understand completely or at least well enough to know what it's, I'm guessing I know what this thing is going to do and I think most people who are worried they have trouble separating out, they don't have, they don't have the knowledge or the understanding about like what is intelligence, how is it manifested the brain? How's it separate from these other functions in the brain. And so they imagine there's going to be human like or animal like it's going to have, it's gonna have the same sort of drives and emotions we have but there's no reason for that, that's just because there's there's unknown if you if the unknown is like oh my God you know I don't know what this is gonna do, we have to be careful, it could be like us but really smarter. I'm saying you know it won't be like us, it would be really smart but it won't be like us at all. And um and but I'm coming from that not because I just guessing I'm not into it using intuition I'm basically like ok I understand this thing works, that's what it does explain it to you </s><s>[INST] okay? But to push back so I I also disagree with the intuitions that SAM has but but I disagree with what you just said, which you know what's a good analogy. So if you look at the twitter algorithm in the early days just recommend our systems. You can understand how recommend our systems work. What you can't understand in the early days is when you apply that recommended system at scale to thousands and millions of people how that can change societies. So the question is yes you're just saying this is how an engineering neocortex works but like when you have a very useful ah tick tock type of service that goes viral when your neocortex goes viral and then millions of people start using it cannot destroy the world. [/INST] Uh Well first of all this is back. One thing I want to say is that A i is a dangerous technology. I'm not denying that. </s><s>[INST] All technologies dangerous. [/INST] Well and ai may be particularly so. Okay, so um am I worried about it? Yeah, I'm totally worried about it. The thing where the narrow component we're talking about now is the existential risk of a right? So I want to make that distinction because I think a I can be applied poorly. It can be applied in ways that you know, people are going to understand the consequences of it. Um These are all potentially very bad things but they're not the ai system creating this existential risk on its own. And that's the only place I disagree with other people. </s><s>[INST] So I I think the existential risk things, humans are really damn good at surviving. So to kill off the human race be very very you can [/INST] even Yes, but you can even I'll go further. I don't think Ai systems are ever going to try to I don't think I systems are ever going to like say I'm gonna ignore you, I'm gonna do what I think is best. Um I don't think that's going to happen at least not in the way I'm talking about it. So you the twitter recommendation, Outrun this interesting example, Let's use computers and analogy again, right? I build a computer, it's a universal computing machine. I can't predict what people are going to use it for. They can build all kinds of things. They can they can even create computer viruses, it's, you know, all kinds of stuff. So there's some unknown about its utility, about where it's going to go. But on the other hand, I pointed out that once I build a computer it's not going to fundamentally change how it computes. Its like I use the example of a register which is a part internal part of a computer. Um you know, I say it can't just because computers don't involve, they don't replicate, they don't involve, they don't, you know, the physical manifested computer itself is not gonna there's certain things that can't do, right? So we can break into things like things that are possible to happen, we can't predict and things are just impossible to happen unless we go out of our way to make them happen. They're not going to happen unless somebody makes them happen. </s> |
<s>[INST] And so the telescopes are looking at this visual data, [/INST] what do you mean by visual? You wouldn't see the fast mean well, </s><s>[INST] but I mean it's uh it's a telescope, it's remote, it's remote. You're observing, you're what zooming in on this particular planet? I mean, what, what, what does the sensor actually look like? How many pixels are there, what does the data kind of look like. It'd be nice to kind of build up intuition of how little data we have based on which I mean if you look at like has just been reading a lot about gravitational waves and it's kind of incredible how from just very little like probably the world's most precise uh instrument. We can derive some very foundational ideas about our early universe. And in that same way, it's kind of incredible how much data, how much information you can get from. Just a few pixels. So what are we talking about here in terms of um based on which this paper saw possible signs of phosphate in the atmosphere [/INST] so far, it's been like every other molecule has a unique spectroscopic fingerprint, meaning it rotates and vibrates in special ways. I calculated how many of those ways it can rotate vibrate in 16.8 billion ways. What this means is that if you look at the spectrum of light and that light has gone through fostering gas on the other end, There should be 16.8 billion tiny marks left indentations left in that spectrum. We found one of those On Venus. One of those 16.8 billion. So now the game is can we find any of the other ones? But they're really hard to spot. They're all in terrible places in the electromagnetic spectrum. And the instruments we used to find. This one can't really find any other one. There's another one of the 16.8 billion we could find. But it would take many, many days of continuous observations and that's not really in the cards right now. </s><s>[INST] I mean how do you, there's all kinds of noise. First of all, there's all kinds of other signal. So how do you separate all of that out to pull out just this particular signature that's associated with fasting? [/INST] So the data kind of looks somewhat like a wave and a lot of that is noise and it's a baseline. And so if you can figure out the exact shape of the wave, you can cancel that shape out and you should be left where the straight line and if there's something there an absorption. So a signal. So that's what we did. We try to find out what was this baseline shape, cleaned it out and got the signal. That's part of the problem. If you do this wrong, you can create a signal But that signal is at 8.904 wave numbers and we actually have more digits than that. But I don't remember by heart and Alma in particular is a very very good telescope array of telescopes and it can focus on exactly that frequency. And in that frequency there are only two known molecules that absorb at all. So that's how we do it. We look at that exact spot where we know frosting absorbs The other molecules s. 0. 2. </s><s>[INST] If there is extraterrestrial life, whether it's on venus or on exoplanets where you look before, How does that make you feel? How should it make us feel, Should we be scared? Should we be excited? Um let's say it's not intelligent life, let's say it's microbial life. It's uh is it a threat to us? Are we a threat to it? Or is it only not only but mostly a possibility to understand something fundamental, Something beautiful about life in the in the universe. [/INST] Hard to know he would have to bring on a poet or a philosopher on the show. I I feel those things, I just don't know if those are the right things to feel. I don't certainly don't feel scared. I think it's rather silly to feel scared, definitely don't touch them. Um You know sometimes in the movies you don't go near it, Don't interfere. I think one of the things with venus is because of fosse now there is a chance that venus is inhabited. In that case we shouldn't go there. We should be very careful with messing with them, bringing our own stuff there that contaminates it. And venus has suffered enough. If there's life there, it's probably the remains of a living planet, the very last survivors of what once was potentially a thriving world. And so I don't want our first interaction with alien life to be massacre. So I definitely wouldn't want to go near out of a let's say galactic responsibility, galactic ethics and I often think of you know alien astronomers watching us and how disappointed they would be if we mess this up. So I really I really want to be very careful with anything that could be life but certainly I wouldn't be scared. Humans are plenty capable of killing one another. We don't we don't need extraterrestrial help to destroy ourselves, </s><s>[INST] scared mostly of other humans. Exactly. But these this life if there is life there it does seem just like you said it would be pretty rugged. It's like the cockroaches there or chuck Norris. I don't know it's the some kind of um it's something that survived through some very difficult conditions. [/INST] That doesn't mean it would handle us. You know, it could be like war of the worlds you come just because you're resilient in your own planet doesn't mean you can survive another even are extremophiles which are very impressive. We should all be very proud of our extremophiles. They wouldn't really make it in the Venusian clouds. So I wouldn't expect because you're tough even chuck Norris stuff that you would survive on a alien planet. </s> |
<s>[INST] But first, like I said, I will be posting old content but now in audio form For a little while I'll probably repeat this intro for reposted YouTube content like this episode and we'll try to keep it to what looks to be just over two minutes, maybe 2.30. So in the future if you want to skip the central, just jumped to the 2 30 minute mark in this episode, I talked with max tech Mark. He's a professor at MIT, a physicist who has spent much of his career studying and writing about the mysteries of our cosmological universe and now thinking and writing about the beneficial possibilities and existential risks of artificial intelligence. He's the co founder of the Future of Life Institute, Author of two books, our mathematical universe and life three point he is truly an out of the box thinker. So I really enjoyed this conversation. I hope you do as well [/INST] think there's </s><s>[INST] intelligent life out there in the universe. Let's open up with an easy question. [/INST] I have a bargain minority of you here Actually when I give public lectures, I often ask for show of hands who thinks there's intelligent life out there somewhere else? And almost everyone put their hands up and when I asked why they'll be like, oh, there's so many Galaxies out there, there's gotta be. But I'm in numbers nerd. Right? So when you look more carefully at it, it's not so clear at all. The when we talk about our universe. First of all, we don't mean all of space, we actually mean, I don't know, you can throw me in the universe if you want behind you there, it's we simply mean the spherical region of space from which light has a time to Beaches so far during the 14.8 billion years, 13.8 billion years since the Big Bang, there's more space here. But this is what we call the universe because that's all we have access to. So is there intelligent life here? That's gotten to the point of building telescopes and computers? My guess is no actually the probability of it happening on any given planet. There's some number. We don't know what it is and what we do know is that the number can't be super high because there's over a billion Earth like planets in the Milky Way Galaxy alone, many of which are billions of years older than earth. And aside from some you of all believers, you know, there isn't much evidence that and the super dry civilization has come here at all. And so that's the famous fermi paradox, right? And then if you if you work the numbers, what you find is that the if you have no clue what the probability is of getting life on a given planet. So it could be 10 to the minus 10 and minus 20 or 10 to minus two. Any power of 10 is sort of equally likely if you want to be really open minded, that translates into it being equally likely that our nearest neighbor Is 10 to the 16 m away. 10 to the 17 m away. 10 - 18. Mhm. By the time he gets Much less than than 10-16 already, we pretty much know there is nothing else that's close. And when you get beyond </s><s>[INST] that would have discovered [/INST] Us. They, yeah, they would have been discovered as long or if they're really close, we would have probably noted some engineering projects that they're doing. And if it's beyond 10-26 m that's already outside of here. So my guess is actually that there are, we are the only life in here that's gotten the point of building advanced tech, which I think is very, um, puts a lot of responsibility on our shoulders, not screw up. You know, I think people who take for granted that it's okay for us to screw up have an accidental nuclear war go extinct somehow because there's a sort of star trek like situation out there with some other life forms are going to come and bail us out and it doesn't matter what I think lulling us into a false sense of security. I think it's much more prudent to say, let's be really grateful for this amazing opportunity we've had and uh, make the best of it just in case it is down to us. So </s><s>[INST] from a physics perspective do you think intelligent life? So it's unique from a sort of statistical view of the size of the universe, but from the basic matter of the universe, how difficult is it for intelligent life to come about? The kind of advanced tech building life? It is implied in your statement that it's really difficult to create something like a human species. [/INST] Well, I think, I think what we know is that going from no life to having life that can do our level of tech? There's some sort of to going beyond that and actually settling our whole universe with life. There's some road, major roadblock there, which is some great filter as um, it's just sometimes called which which is tough to get through. It's either that that roadblock is either behind us or in front of us. I'm hoping very much that it's behind us. I'm super excited every time we get a new report from Nasa saying they failed to find any life on mars like just awesome because that suggests that the hard part, maybe maybe it was getting the first rifle zone or or some some very low level kind of stepping stone. So there we're home free because if that's true, then the future is really only limited by our own imagination would be much silkier if it turns out that this level of life is kind of a diamond dozen. But maybe there's some other problem, like as soon as a civilization gets advanced technology within 100 years, they get into some stupid fight with themselves and poof! Now, that would be a bummer. </s> |
<s>[INST] describe brain net in the future of the mind. This is an interesting concept. Do you think uh you mentioned entertainment, but what kind of effect would it have on our personal relationships. [/INST] Hopefully it will deepen it. You realize that for most of human history for over 90% of human history, We only knew maybe 20, 100 people. That's it folks. That was your tribe. That was everybody you knew in the universe. Was only maybe 50 or a 100 with the coming of towns. Of course it expanded to a few 1000 with the coming of the telephone. All of a sudden you could reach thousands of people with the telephone and now with the internet, you can reach the entire population of the planet earth. And so I think this is a normal progression </s><s>[INST] and you you think that kind of sort of connection to the rest of the world and then adding sensations like being able to share telepathically emotions and so on. That would just further deepen our connection to our fellow humans. [/INST] That's right. In fact, I disagree with many scientists on this question. Most scientists would say that technology is neutral, A double edged sword, one sword, one side of the sword can cut against people. The other side of the sword can cut against ignorance and disease. I disagree. I think technology does have a moral direction. Look at the internet, the internet spreads knowledge awareness and that creates empowerment. People act on knowledge when they begin to realize that they don't have to live that way. They don't have to suffer under a dictatorship. That there are other ways of living under freedom. Then they begin to take things, take power and that spreads democracy and democracies do not war with other democracies. I'm a scientist. I believe in data. So let's take a sheet of paper and write down every single war you had to learn since you were in elementary school, every single war. Hundreds of them. Kings, queens, emperors, dictators. All these wars were between kings, queens, emperors and dictators. Never between two major democracies. And so I think with the spread of this technology and which would accelerate with the coming of brain net, it means that, well, we will still have wars. Wars of course, is politics by other means, but there'll be less intense and less frequent. </s><s>[INST] Do you have worries of longer term existential risk from technology from ai So, I think that's a wonderful vision of a future where war is a distant memory. But now there's another agent, there's there's there's somebody else that's able to create conflict that's able to create harm. Ai systems. Do you have worry about such A. I. [/INST] Systems? Well, yes, that is an existential risk. But again, I think an existential risk not for this century. I think our grandkids are going to have to confront this question as robots gradually approach the intelligence of a dog, a cat, and finally, that of a monkey. However, I think we will digitize ourselves as well. Not only are we going to merge with our technology will also digitize our personality, our memories, our feelings. You realize you did during the middle ages, there was something called dualism. Dualism entered. The soul was separate from the body when the body died, the soul went to heaven. That's dualism. Then in the 20th century neuroscience came in and said bah humbug. Every time we look at the brain it's just neurons. That's it folks. Period end of story bunch of neurons firing. Now we're going back to dualism now we realize that we can digitize human memories, feelings, sensations and create a digital copy of ourselves and that's called the connecting project. Billions of dollars are now being spent to do not just the genome project of sequencing the genes of our body, but the connect um project which is to map the entire connections of the human brain. And even before then, already in Silicon Valley today, at this very moment you can contact Silicon Valley companies that are willing to digitize your relatives because some people want to talk to their parents. There are unresolved issues with their parents and one day yes. Berms will digitize people and you'll be able to talk to them a reasonable facsimile we live. We leave a digital trail. Our ancestors did not. Our ancestors were lucky if they had one line, just one line in a church book saying the date they were baptized and the date they died. That's it. That was their entire digital memory. I mean their entire digital existence, summarized in just a few letters of the alphabet, whole life. Now we digitize everything every time you sneeze, you digitize it, you put it on the internet. And so I think that we are going to digitize ourselves and give us digital immortality. Will not only have biologic genetic immortality of some sort, but also digital immortality. And what are we going to do with it? I think we should send it into outer space if you digitize the human brain and put it on a laser beam and shoot it to the moon, you're on the moon in one second. Shoot it to Mars, you're on Mars in 20 minutes, shoot it to Pluto, you're on Pluto in eight hours. Think about it for a moment, you can have breakfast in New York and for a morning snack, vacation on the moon. Then zap your way to MArs by noontime journey to the asteroid belt in the afternoon and they come back for dinner in new york at night, all in a day's work at the speed of light. Now, this means that you don't need booster rockets, you don't need weightlessness problems. You don't need to worry about meteorites and what's on the moon On the moon. There is a mainframe that downloads your laser beams information. And where does it download the information into an avatar. What does it avatar look like? Anything you want to think about it for a moment, you could be superman superwoman on the moon. On mars traveling throughout the universe at the speed of light downloading your personality into any vehicle you want. Now let me stick my neck out so far. Everything I've been saying is well within the laws of physics, well within the laws of physics now, let me go outside the laws of physics. Here we go. I think this already exists. I think outside the Earth there could be a superhighway, a laser highway of laser porting with billions of souls of aliens zapping their way across the galaxy. Now let me ask you a question, Are we smart enough to determine whether such a thing exists or not? No. This could exist right outside the orbit of the planet Earth and we're too stupid in our technology to even prove it or disprove it. </s> |
<s>[INST] Can we go there for a second? Uh You've talked about attractor networks? Uh and just maybe you could say what are attractive networks and more broadly what are interesting network dynamics that emerge in these or other complex systems. [/INST] You have to be willing to think in a huge number of dimensions because there's a huge number of dimensions. The behavior of a system can be thought of just the motion of a point over time, in this huge number of dimensions, right? On an attractor network is simply a network where there is a line and other lines converge on it in time. That's the essence of an attractor network, That's how you </s><s>[INST] win a highly highly dimensional space. [/INST] And the and the easiest way to get that is to do it in a high dimensional space where some of the dimensions provide the dissipation, which base which the kind of a physical system trajectories can't contract everywhere, the after contract in some places and expand in others. There is a fundamental classical theorem, was statistical mechanics, which goes under the name of Louisville's theorem, which says you can't contract everywhere, You have to, if you contract somewhere you expand somewhere else, do you? It is an interesting physical systems. You get driven systems where you have a small sub system which is the interesting part and the rest of the contraction and expansion. The physicists would say its entropy flow in the southern part of the system but but basically attract your networks are dynamics funneling down so you can't be any. So that if you start somewhere in the dynamical system, you will soon find yourself on a pretty well determined pathway which goes somewhere, You start somewhere else. You'll wind up on a different pathway. But you don't have just all possible things. You have some defined pathways which are allowed and under which you will converge, that's the way you make a stable computer and that's the way you make a stable behavior. </s><s>[INST] So in general, looking at the physics of the emergent stability and these no networks, what are some interesting characteristics that uh what are some interesting insights from studying the dynamics of such high dimensional systems? [/INST] Most dynamical systems. Most driven dynamical systems I've driven there couple somehow to an energy source. And so the dynamics keeps going because its coupling to the energy source. Most of them, it's very difficult to understand it all with the damn dynamical behavior is going to be </s><s>[INST] you have to run it. [/INST] You have to run in. There's a there's a subset of systems which has what exactly the mathematicians as its the albino function. And those systems you can understand convergent dynamics by saying you're going downhill on something or other. And that's what I found without ever knowing what the optional functions were. In the simple model I made in the early 80's was an energy functions. That you can understand how you could get this channeling on under pathways without having to follow the dynamics in an infinite detail. You started rolling a ball at the top of the mountain, it's gonna wind up in the bottom of the valley. You know, that's true. Without actually watching the ball fall roll down. </s><s>[INST] There's certain properties of the system that when you can know that [/INST] the that's right and not all systems behave that way. Most don't probably both don't. But it provides you with the metaphor for thinking about systems which are stable and who to have these attractors behave. Even if you can't find a Leopoldo function behind them or an energy function behind them. It gives you a metaphor for thought. Mm hmm. </s><s>[INST] Speaking of thought, if I had a glint in my eye with excitement and said, um you know, I'm really excited about this something called deep learning and neural networks and I would like to create an intelligence system and came to you as an advisor. What would you recommend? Uh Is it a hopeless pursuit to use neural networks? The chief thought, is it? What kind of mechanism should we explore? What kind of ideas should we explore? [/INST] Well, you look at the simple network will have one past networks. They don't support multiple hypotheses very well as I have tried to work with very simple systems which do something which you might consider to be thinking, thought has to do with the ability to do mental exploration before you take a physical action. </s><s>[INST] Almost like we were mentioning playing chess, visualizing, simulating inside your head different [/INST] outcomes. And now you could do that and feed forward network because you've pre calculated all kinds of things. But I think the way neurobiology does, it hasn't pre calculated everything. It actually has parts of a dynamical system which you're doing exploration in a way which is </s><s>[INST] there's a creative element. [/INST] Like there's there's there's there's a creative element. And in a in a simple minded neural net, you have a a constellation of instances from which you've learned. And if you are within that space, if if a new a new question is a question within this space, you can actually rely on that system pretty well to come home with a good suggestion for what to do. If on the other hand, the query comes from outside the space, you have no way of knowing how the system is going to behave, there are no limitations on what can happen. And so the artificial neural net world is always very much I have a population of examples. The test set must be drawn from this equivalent population, the test that has examples which are from a population which is completely different. There's no way that you could expect to get the answer right? </s><s>[INST] Yeah. And so what they call outside the distribution. [/INST] That's right, That's right. And so if you see a ball rolling across the street at dusk, if that wasn't in your, your training set, the idea that a child may be coming close behind that is not going to occur to the neural that </s> |
<s>[INST] So that's a computational approached to figuring something out and when when it comes together like that, that's kind of a nice clean indication that there's something this could be actually indicative of what's what's happening. [/INST] Yes, I mean so so we we need this information and you know uh three D. The three D structure gives us just a very intuitive way to look at this information and then start to ask, you know, start asking questions such as so this place of this protein that is highly mutated. Does it does it Uh huh. Is it the functional part of the protein? So does this part of the protein interact with some other proteins or maybe with some other legans, small small molecules. Right. So we will try now to functionally inform This three D structure. </s><s>[INST] So uh do you have a bunch of these mutated parts if like I don't know like how many are there in the new novel coronavirus? And compared to stars that we're talking about hundreds, thousands, like these these pink [/INST] regions. Oh no, that much less than that. And it's very interesting that if you look at that, you know. So the first thing that you you start seeing right, you know, you look at patterns, right? And the first pattern that becomes obvious is that some of the proteins in the new coronavirus are pretty much intact. All right. So they are pretty much exactly the same as SARS as the bad coronavirus where some others are heavily mutated. All right. So so it looks like that the you know, the evolution is not is not a curing you know, uniformly across the entire you know, viral genome but actually target very specific proteins. </s><s>[INST] What do you do with that? Like from the Sherlock Holmes perspective? [/INST] Well, you know, so one of the uh of the uh most interesting findings we had was the fact that the viral uh so the binding sites on the viral surfaces uh that get targeted by the non small molecules the war. Pretty much not affected at all. And so that means that the same small drugs or small small drug like compounds can be efficient for the new coronavirus. Uh </s><s>[INST] huh. Uh So this will actually maps to the drug compounds too? Like so so you're actually mapping out uh what old stuff is gonna work on this thing and then possibilities for new stuff to work by mapping out the things have mutated. [/INST] So so we essentially know which parts behave differently and which parts are likely to behave similar. And again, you know of course all our predictions need to be validated by experiments but hopefully that sort of helps us to doing it. The regions of this virus that can be promising in terms of the drug discovery, </s><s>[INST] you kind of, you kind of mentioned this already but maybe you can elaborate. So how different from the structural and functional perspective does the new coronavirus appeared to be relative to SARS. [/INST] We now are trying to understand uh the overall structural characteristics of this virus. Because I mean that's that's our next step trying to model the viral particle of a single viral particle of this virus. </s><s>[INST] So that means so you have the individual proteins like you said, you have to figure out what their interaction is. Uh yeah, this is that where this graph kind of interact home. [/INST] So so so so the interact um it is essentially a so our prediction on the potential interactions, some of them that we already deciphered from the structural knowledge but some of them that essentially are deciphered from the knowledge of the existing interactions that people previously obtained for SARS farmers or other related viruses </s><s>[INST] is they're kind of interact Tums. My pronouncing that correctly by the way. Yeah. Are those already converge towards for SARS for [/INST] So I think there is there are a couple of papers that now investigate the sort of the large scale uh set of sets of interactions between the new stars and uh it's host. And so I think that's that's an ongoing study I </s><s>[INST] think. And the success of that, the result would be an interact on Yes. And so when you say you're not trying to figure out the entire the particle, the entire single. [/INST] Right? So if you look, you know, so structure. Right. So what this wider particle looks like? Right. So as I said, it's it's you know, the surface of it is an envelope which is essentially a so called lipid bi layer uh with proteins integrated into the surface. So how social and average particle is around um 18 nanometers. Right. So this particle can have About 50 200 spike proteins. So at least we suspected. And you know, based on the micro graphs images, it's very comparable to M HV virus in mice and SARS virus. </s><s>[INST] Micro grafts are actual pictures of the actual virus. Okay, so these are models. This is [/INST] actually so they did the actual images. </s><s>[INST] Right. Where they started for the tangents. But what are these things? So when you look on the internet? The models and the pictures are and kind of in the models you have here are just gorgeous and beautiful when you actually take pictures of them with the micrografx. Like what, what do we look? [/INST] Well, they typically are not perfect. Right. So, so the most of the images that you see now is the is the sphere with those spikes, </s><s>[INST] spikes, you do [/INST] see the spikes and now, you know, the our collaborators for texas and A and M. University um Benjamin neuman, uh he actually uh in the recent paper about SARS he proposed and there's some actually evidence uh behind it that the particle is not a sphere, but it's actually as Elon gated Alex toyed like uh particles. So, so that's what we are trying to incorporate into our model and the re I mean, you know, if you look at the actual micro graphs, you see that those particles are, you know, are not symmetric, so they're the some of them. And of course, you know, it could be due to the treatment of the of the material. It could be due to the some noise in the imaging. </s> |
<s>[INST] there's plan area that have lived that long [/INST] essentially. What </s><s>[INST] does it mean physical continuity? [/INST] Because because what they do is they split in half the way they reproduces, they split in half. So so the plan area, The back, the back end grabs the Petri dish the front and takes off and they rip themselves in half. </s><s>[INST] But isn't it some sense where like you are a physical continuation? [/INST] Yes. Except that except that we go through a bottleneck of one cell, which is the egg. They do not. I mean they can, there's certain plan area </s><s>[INST] so we go through a very ruthless compression process and they don't [/INST] like an autoencoder, you know, Squashed down to one cell and then back out. These these guys just tear themselves in half. And then each and then and so the other amazing thing about them is they regenerate so you can cut them into pieces. The record is I think 276 or something like that. By Thomas Hunt Morgan. Uh and each piece re grows a perfect little worm. They know exactly every piece knows exactly what's missing, what needs to happen. Uh In fact, in fact, if you chop it in half as it grows the other half, the original, the original tissue shrinks so that when the new tiny head shows up they're proportional so it keeps it keeps perfect proportion. If you if you starve them, they shrink if you feed them again, they expand their control. Their anatomical control is just insane. </s><s>[INST] Somebody cut them into over 200 pieces. [/INST] Yeah thomas Hunt morgan did </s><s>[INST] hashtag science [/INST] amazing. Yeah. And maybe more. I mean they didn't have antibiotics back and I bet he lost some due to infection. I bet I bet it's actually more than that. You could I bet you could do more than that. </s><s>[INST] Humans can't do that. Well, [/INST] yes, I mean again. Yeah, true. Except that. </s><s>[INST] Well [/INST] that's that's the thing. Right? So, so I tell when when I talk about this, I said just remember that is as amazing as it is to grow a whole planetary in from a tiny fragment half of the human population can grow full body from one cell. Right? So, so development is really, you can look at development as a as a just an example of regeneration. Yeah. </s><s>[INST] To think we'll talk about regenerative medicine, but There's some sense that would be like that warm in like 500 years where I can just go regrow hand. [/INST] Yeah. With were given time. It takes time to grow large things. </s><s>[INST] But yeah, I [/INST] think so. I think </s><s>[INST] it probably why not accelerate. Oh, biology takes this time. I'm [/INST] not gonna say anything is impossible, but I don't know of a way to accelerate these processes. I think it's possible. I think we are going to be regenerative, but I don't know of a way to make it faster </s><s>[INST] if you just think people from a few centuries from now and be like, well they have to they used to have to wait a week for the hand to regrow. It's like when the microwave was invented you can you can toast your um what's that called when you put a cheese on a toast. Um It's delicious is all I know, I'm I'm blanking and Alright, so plan area, why are we talking about the magical plan area that they have the mystery of life? [/INST] So the reason we're talking about is not only are the immortal. Okay, not only do they regenerate every part of the body? Uh They they generally don't get cancer. Right? So which we can talk about? Why that's important. They're smart, they can learn things so you can train them. And it turns out that if you train a planetary in and then cut their heads off the tail, will regenerate a brand new brain that still remembers the original information. Do </s><s>[INST] they have a biological network going on? Yes. So their somatic cells are forming a network and that's that's what you mean by a true brain. What's the requirement for a true brain? [/INST] I like everything else. It's a continuum. But but a true brain has certain characteristics as far as the density, like a localized density of neurons that guides behavior </s><s>[INST] in the head. [/INST] Exactly, exactly. If you cut their head off the tail doesn't have that doesn't do any. It just sits there until the new brain is is, you know, until the new brain regenerates. They have all the same neurotransmitters that you and I have but here's why here's what we're talking about them in this in this context. So here's your planetary, you cut off the head, you cut off the tail of a middle fragment, That middle fragment has to make one head and one tail. How does it know how many of each to make? And where do they go? How come it doesn't switch? How come? Right. So, so we did a very simple thing and we said, okay, let's let's make the hypothesis that there's a somatic electrical network that remembers the correct pattern. And what it's doing is is recalling that memory and building to that pattern. So what we did was we used a um a way to visualize electrical activity in these cells. Right? It's a it's a it's a variant of what people used to look for electricity in the brain. And we saw that it has a that that fragment has a very, very particular electrical pattern. You can literally see it. Once, once we developed a technique, it has a very particular electrical pattern that shows you where the head and the tail goes, right, you can you can just see it. And then we said, okay, well now let's test the idea that that's a memory that actually controls where the head and the tail goes. Let's change that pattern. So basically accept a false memory. And so what you can do is you can do that in many different ways. One way is with drugs that target ion channels to say. And and so you you pick these drugs and you say, okay I'm gonna do it. So that instead of so that instead of this one head, one tail electrical pattern you have a two headed pattern. Right? You're just editing the electrical information in the in the network when you do that. Guess what the cells build. They build a two headed worm. And the coolest thing about it now. No no genetic changes. We haven't touched the genome. The genome is totally wild type. But the amazing thing about it is that when you take these two headed animals and you cut them into pieces again some of those pieces will continue to make two headed animals. So so that information that that memory that that electrical circuit not only does it hold the information for how many heads? Not only does it use that information to tell the cells what to do to regenerate but it stores it once you've reset it it keeps and we can go back we can take a two headed animal and and put it back to one headed. So now imagine. So there's a couple of interesting things here that that have implications for understanding of genomes and things like that. Imagine I take this two headed animal. Um Oh and by the way when they reproduce when they tear themselves in half you still get two headed animals. So imagine they take them and throw them in the Charles river over here. So 100 years later some scientists come along and they scoop up some samples and they go oh there's a single headed form and the two headed form, wow, a speciation event. Cool. Let's sequence the genome and see why what happened, genomes are identical. There's nothing wrong with the genome. So if you ask the question, how does so so this goes back to your very first question is where do body plants come from? Right. How does the planetary in know how many heads it's supposed to have now? It's interesting because you could say D. N. A. But what happened? What what as it turns out the D. N. A. Produces a piece of hardware that by default says one head. The way that when you turn on a calculator by default it's zero every single time. Right? When you turn it on, just zero. But it's a programmable calculator as it turns out. So once you've changed that next time it won't say zero. It'll say something else. And the same thing here. So you can make you can make one head to head and you can make no headed worms. We've done some other things along these lines. Some other really weird constructs. So so this this this this question. Right? So again it's really important. The hardware software distinction is really important because the hardware is essential because without proper hardware you're never gonna get to the right physiology of having that memory. But once you have it it doesn't fully determine what the information is going to be. You can have other information in there and it's free programmable by us by bacteria by various parasites. Probably um things like that. The other amazing thing about these plan areas think about this most animals when we get a mutation in our bodies, our Children don't inherit it. Right? So you can go on, you can run around for 50 60 years getting mutations. Your Children don't have those mutations Because we go through the egg stage plan area tear themselves in half and that's how they reproduce. So for 400 million years they keep every mutation that they've had that doesn't kill the cell that it's in. So when you look at these plan area, their bodies are what's called mixed employed meaning that every cell might have a different number of chromosomes. They look like a tumor. If you look at the genome is an incredible mess because they accumulate all this stuff and yet the their body structure is they are the best re generators on the planet. Their anatomy is rock solid, even though their genome is is all kinds of crap. So this is kind of a scandal right? That you know when we learn that, well, you know what our genomes genomes determine your body. Okay. Why is the animal with the worst genome have the best anatomical control the most cancer resistant, the most regenerative. Right? Really. We're just beginning to start to understand this relationship between the genomic lee determined hardware and and and by the way, just as a as a couple of months ago I think I now somewhat understand why this is but it's really it's really a major, you know, major puzzle </s> |
<s>[INST] So you have looked deep into these people have deep conversations and it's interesting to get a sense of stories of the way they were thinking and the way it was changed the way your own thinking about Ai has changed. So you mentioned McCarthy is um what about the years of cmu Carnegie Mellon with joe And was sure [/INST] joe was not in a I he was in algorithmic complexity. </s><s>[INST] Was there always a line between ai and computer science for example is a i it's own place of outcasts. It was that the feeling there [/INST] was a kind of outcast period for AI for instance in 1970 for the new field was hardly 10 years old. The new field of computer science was asked by the National Science Foundation I believe. But it may have been the National academies. I can't remember uh to you know tell us tell our your fellow scientists where computer science is and what it means and they wanted to leave out a I Uh huh and they only agreed to put it in because Don Knuth said hey this is important. You can't just leave that out. </s><s>[INST] Really. Don. Don [/INST] Kennedy. Yes </s><s>[INST] I talked to mrs e joseph out of all the people. [/INST] Yes. But you see an Ai person couldn't have made that argument. He wouldn't have been believed but can youth was believed. Yes. </s><s>[INST] So joe crow worked on the real [/INST] stuff, joe was working on algorithmic complexity but he would say in plain english again and again the smartest people I know are in a I really? Oh yes, no question. Anyway joe I love these guys. What happened was that I guess it was as I started to write machines who think Herb Simon and I became very close friends. He would walk past our house on Northumberland Street every day after work and I would just be putting my cover on my typewriter and I would lean out the door and say her, would you like a sherry? And her almost always would like a sherry. So he'd stop in and we talked for an hour, two hours. My journal says we talked this afternoon for three hours. </s><s>[INST] What was in his mind at the time? If in terms of the ai side of things. [/INST] Uh we didn't talk too much about ai we talked about other things in life. We we both love literature and Herb had read Proust in the original french twice all the way through. I can't read it in english and translation. So we talked about literature. We talked about languages, we talked about music because he loved music. We talked about art because he was he was actually enough of a painter that he had to give it up because he was afraid it was interfering with his research and so on. So no, it was really just chat chat, but it was very warm. So one summer I said to her, you know my students have all the really interesting conversations. I was teaching at the University of Pittsburgh then in the english department. You know, they get to talk about the meaning of life and that kind of thing. And what do I have? I have university meetings where we talk about the photocopying budget and you know, whether the course on romantic poetry should be one semester or two. So her laughed. He said, yes, I know what you mean, he said, but you know you could do something about that Dot was his wife Dot and I used to have a salon at the university of Chicago every sunday night. And we would have essentially an open house and people knew it wasn't for sure. A small talk. It was really for some the topic of uh depth, he said, but my advice would be that you choose the topic ahead of time. Fine. I said so the following we we exchanged mail over the summer that was us post in those days because you didn't have personal email and um we I decided I would organize it and there would be eight of us Allen Newell and his wife, Herb Simon and his wife Dorothy. Um there was a novelist in town a man named Mark Harris. He had just just arrived and his wife, josephine Mark was most famous then for a novel called Bang the Drum Slowly, which was about baseball And Joe and Me So eight people. And we met monthly and we we just sank our teeth into really hard topics and it was great fun. </s><s>[INST] How have your own views around artificial intelligence changed and through the process of writing machines who think and afterwards the ripple effects. [/INST] I was a little skeptical that this whole thing would work out. It didn't matter to me, it was so audacious, </s><s>[INST] the whole thing being a I [/INST] a I generally and in some ways uh yeah, it hasn't worked out the way I expected so far. That is to say there is this wonderful lot of apps, thanks to deep learning and so on. But those are algorithmic and in the part of symbolic processing there is very little yet, yes. And that's the field that lies waiting for industrial industrious graduate students. </s><s>[INST] Maybe you can tell me some figures that popped up in your life in the 80s with expert systems or there was the symbolic ai possibilities of what's you know that what most people think of as a i if you dream of the possibilities ai is really expert system and those hit a few walls and those challenges there. And I think yes, they will re emerge again with some new breakthroughs and so on. But what did that feel like? Both the possibility and the winter that followed the slow down. [/INST] You know, this whole thing about ai winter is to me a crock, </s><s>[INST] it's no winters [/INST] Because I look at the basic research that was being done in the 80s, which is supposed to be my God, it was really important. It was laying down things that nobody had thought about before. But it was basic research, you couldn't monetize it. Hence the winter. That's the winter. You know, research scientific research goes and fits and starts. It isn't this nice smooth. Oh, this follows, this follows this. No, it just doesn't work that </s> |
<s>[INST] Let me just let me just sit in the, in the silence of it. There's been so much pain on this one point, [/INST] So much suffering. </s><s>[INST] So I mean I yeah, I did a lot of algorithms on automatic synchronization of audio and video and all these kinds of things. And I know this well, uh, there's a lot of science and there's a lot of differences, but it's about and people claim it's about 100 million seconds. You can't tell the difference, but it's much more like 30 to 50 milliseconds and you can go nuts trying to see if something is in sync or not. Is it in sync or not? [/INST] We'll also </s><s>[INST] out of sync, right [/INST] brain is constantly making adjustments and so it can shift for you while you're doing that, which is probably part of the thing that's driving you crazy. Um Okay so I'll start with binding processes and then I'll just give a couple examples. Um So yes, there's this window um where your brain is essentially putting all the information together to deliver you a present moment experience that is most useful for you to navigate the world. Um So as I said and you know, I use this example of tennis in my book, so the sights and sounds are coming at us at different rates. Um It takes longer for a sensation in my hand when I hit the ball with the racket to travel to my brain than it does for the light waves to hit my retina and get processed by the brain. So all these these signals are coming in at different times. Our brains go through this process of binding to basically weave it all together so that our our conscious experience of that is of seeing hearing and feeling the ball hit the racket all at the same time. And that's obviously most useful to us binding is is mostly about timing. It can be about other things, but I was just talking to David Eagleman who was talking about a very simple experiment actually and this kind of shows how your brain is basically always interacting with the outside world and always making adjustments to make its best guess about the most useful present moment experience to deliver. So this is a very simple experiment. This from many, many years ago. Um and David David Eagleman was was involved in this research where they had participants hit a button and that button caused a flash of light. And so our brains, through binding the brain it notices, is is able to kind of calibrate the experience you have because the brain is aware that it is its own hand that is causing the light to flash, that there's this cause and effect going on. And so you have this experience of pushing the button that causes the flash of light which is true. And the the light flashes, you can start to introduce longer pauses. Um Starting with 20 milliseconds, 30 milliseconds going up to I think 100 maybe even 200 milliseconds. Where if you do it gradually since your brain is making the adjustment um you can introduce a delay, I think it's up to 200 milliseconds. If you do it gradually, you will still have the experience even though there's now a delay between when you hit the button and the light flashes you will still have the exact same experience you had initially wishes that the light flashes right? When you push the button in your experience, nothing is changing. But then so they gradually give a delay, you've acclimated to that because it was done gradually. If they then go back to the original instantaneous flash, your brain doesn't have time to make the adjustment. And you have the experience that the light flashed before you hit the button. And that is your true experience. It's not like you're confused but that is your brain didn't have time to make that adjustment. You think you're in the same environment, you're pushing the button and makes the light flash. It's kind of calibrating all the time. But then the participants are suddenly saying oh wait that was so weird. The light flashed before I hit the button. And so these </s><s>[INST] they [/INST] built a um rochambeau rock paper scissors, computer game that was unbeatable based on this glitch that you can present in in binding by training someone. If you introduce a delay slowly enough then the computer can get the information before it responds. But you still have the experience that you're both throwing out your rock or paper scissors at the same time. But in actuality the computer saw your choice before it makes its choice and it's in this window of milliseconds where you don't notice it. </s><s>[INST] So that starts to help you build up an intuition that this conscious experience is an illusion constructed by the brain after [/INST] conscious will, </s><s>[INST] conscious will. [/INST] Yeah. And and just in general that that consciousness is not the thing that we feel it is which is driving the behavior that is actually at the tail end of it. And so a lot of decision making processes and there are studies that are more controversial and I don't usually um like to cite them, although if you want to talk about them we can, they're super interesting and into intuition shattering. But there there are now studies specifically about free will to see if there are markers at the level of the brain, they can see what decision you're going to make and when you make that decision. And um I think this the neuroscience inevitably is just going to get better. And so part of the reason I'm so passionate about this, I mean there's there's a science and there's just the curiosity that that drives me of wanting to understand how the universe works. But I actually see a lot of the neuroscience presenting us with truths that are going to be difficult for us to accept. And I actually think there are really positive ways to view these truths that we're uncovering. And even though they can be initially kind of jarring and even destabilizing and creepy, I think ultimately there's actually a lot it can have a positive effect on human psychology and a whole range of things um that I and others have have experienced and that I think it's important for us to talk about because you can't hide from the truth, especially in science, right? Like it just, it will reveal itself and if this is true, I think not only for better understanding the universe and nature, which is kind of my my primary passion. Um, it's important for us to absorb these facts and realize that they don't, it doesn't necessarily take away the things from us that we fear. Um you know, I've heard people say is talked about common, common point to make a question to ask a scientist. Can you still, you know, enjoy chocolate if you're a molecular biologist and um as a molecular biologist, that would be the one who would understand how we experience </s> |
<s>[INST] You know, I but see the crucial piece for me that I've been focusing on is the social media aspect of how the it's not just about the reality of deaths. It's about the state of the collective intelligence of the human species, which is determined by our communication on social media. So we can, yeah, we can be collectively afraid. The fear can spread or it could be YOLO can spread or it could be like all different kinds of misinformation and of course during the election year, the politics influences our perception of what is true not, but you know, having real rigorous nuanced conversation about this kind of stuff is the way is the way out of this. And that's where social media really comes in because social media is drives division where people form tribes and so on. And it feels like it's honestly a technology problem. People say it's a human problem, but it just feels like I believe [/INST] he was a good technology can enable </s><s>[INST] them to be thoughtful. And [/INST] we talked earlier about um you know the magic of Silicon Valley and then maybe going too far with the facebook groups example where you know you take out all that friction. What happened was we used to have something called R Cochran reverse chronological order. That's how you consume to feed. So any kind of social feed like twitter was in reverse chronological order. The newest thing was up top and you would just work your way backwards. And so it gave this like really fresh feeling and then a guy named Dave morin and the team over at facebook realized you know there are some things that got a lot of attention two hours ago and the stuff since then has not been as important. But if you missed that there was a really good tweet where there was a really good update. Like somebody had a baby what That's how can we get the baby one at the top. And I was like, well how would we do that? How would we know that that's the important one. It's like, well let's put a like button on it and let's see how many comments there are. So if it gets a lot of likes or comments or retweets, let's show those first and then we'll kind of mix in the most recent stuff. And so when you're on twitter and then when facebook did that facebook became so addicting because facebook was on what has got the most engagement put that first. So every time you open your facebook at the dopamine hit and then what happens when you see the bar mitzvah photo or you know the enraging story about some injustice in the world, you retweeted, you write a comment, you share it on your wall and thus this addiction to the outrageous the outlandish the inspiring occurred. And it used to be like inspiring stuff, puppies or some heartwarming story and then it got dark and then people started to realize if I want to show up on the top of my friends feeds, If I say something controversial or I'm outraged, I get to the top and then that's when outrage culture came in and then that's when cancel culture came in, everybody started to realize if I try to cancel that person for being a racist or sexist or horrible human being or whatever they did, that's wrong? I get to the top of the feed and we all collectively started playing a very weird video game, which is how outraged can we all be. And to get to the top of the list. And then of course with trump, he realized, and he's like, okay, yeah, I'm just gonna make fun of a celebrity and I get more retweets, okay, I'm going to make fun of Rosie O'donnell for being overweight or something, and he just starts attacking people and people like, oh my God, what did he say? And he copied that from Howard Stern Because he was in New York and he used to be on Howard Stern and Howard Stern took over all the dialogue in the 80s and 90s because he was outrageous and then trump did that. And then social media incorporated that into the operating system. It became the actual device of social media was the Ding Ding Ding, Ding Ding, we've got something incredible for everybody salivate, like Pavlov's dog, you know, Oh my God, I could be outraged. That's what's got to be undone. And the only way for that to be undone is these things can't be billions of people where the most outrageous thing that happened in the world today in the last five minutes is now in front of you and that's why people have anxiety, they don't sleep and they doom scroll all night is because the human mind was not meant to process this much suffering pain, anger and that's why we have all this mental health issues also, you know, young girls or even adults watching other people post their private jets and their vacations and you know, YOLO adventures on their instagram to the point at which young people are now faking being on private jets to put on their instagram and creating like this crazy fomo around their instagrams. Like now we wonder why people are unhappy. Like if you think everybody's on a private jet, going to some Michelin star restaurant or whatever, the coolest thing in the world is today, like going to the Grammys, going to whatever Coachella burning man like you're like but I'm home, I'm in my house. Yeah and I'm not at burning man </s> |
<s>[INST] it's a, it's an interesting concept nuclear. It seems like people like everybody that works on nuclear power has shown how safe it is as a source of energy. And uh and yet we are seem to be, I mean based on the history based on the excellent HBO series Russian with the Chernobyl. It seems like we have our risk estimation about this particular power sources uh, drastically inaccurate. But that's that's a fascinating idea that we would use nuclear as a source for our vehicles uh, and not just in outer space. That's cool. I'm going to look into that. That's super interesting. [/INST] Well, just last year trump eased up a little bit on the regulations and Nasa and hopefully others are starting to pick up on the development. So now is a good time to look into it because there's actually some movement, </s><s>[INST] is that a hope for you to explore different energy sources that the entirety of the vehicle uses something like uh, like the entirety of the propulsion systems for all aspects of the vehicle's life travel is the same or electric. Is it possible for it to be the same? Like the kool Aid engine being used for everything [/INST] you could and you would have to do it in the same way we do different stages of rockets now where once you've used up an engine or a stage you let it go because there's really no point in holding on to it. So I wouldn't necessarily want to use the, the same engine for the whole thing, but the same technology I think would be interesting. </s><s>[INST] Okay, so it's possible, all right. But [/INST] uh, in terms </s><s>[INST] of the power source and the power source, that's really interesting. But for the current power sources and its current use cases, what's the use case for electric? Like the the Polaroid engine? Can you talk about where they're used today? [/INST] Sure. So chemical engines are still used quite a bit once you're in orbit, but that's also where you might choose instead to use an electric system and what people do with them. And this includes, you know, the ion engines and hull thrusters and orange in um, is basically any maneuvering you need to do once you're dropped off. There's even if your only goal was to just stay in your orbit and not move for the life of your mission, you need propulsion to accomplish that because the Earth's gravity field changes as you go around in orbit and pulls you out of your little box. There are other perturbations that can throw you off a bit. Um, and then you know, most people want to do things a little bit more interesting like maneuver to avoid being hit by space debris or perhaps lower their orbit to take a higher resolution image of something and then return um at the end of your mission you're supposed to responsibly get rid of your satellite whether that's um burning it up but if you're in G. O. Um you want to push it higher and to graveyard orbit um </s><s>[INST] was geo [/INST] what's so low earth orbit and then geosynchronous orbit or geo stationary orbit </s><s>[INST] and there's a graveyard. [/INST] Yeah, so those satellites are at um like 40,000 km so if they were to try to push their satellites um back down to burn up in the atmosphere, they would need, you know, even more propulsion than they've had for the whole lifetime of their mission. So instead they push them higher where it'll take, you know, a million years for it to naturally deorbit. So we're also cluttering that higher bit up as well but it's not as pressing as as leo, which is low earth orbit where more of these commercial missions are going now. </s><s>[INST] So how hard is the collision avoidance problem there? He said some debris and stuff so like how much propulsion is needed? Like how much is the life of a satellite is just like a crap trying trying to avoid, like what I [/INST] think one of the recent um you know rules of them I heard was per year some of these small satellites are doing like three collision avoidance maneuvers um So that's that's not yeah, but it's not zero um And it yeah, it takes a lot of um planning and people on the ground and you know, none of that really. I don't think right now is autonomous. </s><s>[INST] Oh that's not good. [/INST] Yeah. And then we have a lot of folks taking advantage of moore's law on cheaper spacecraft so they're launching them up without the ability to maneuver themselves and they're like, well I don't know, just don't hit me </s><s>[INST] and three times a year that could be become affordable if it's like if he gets hit maybe it won't be damaged kind of thing. That kind of logic [/INST] Affordable in that instead of launching one satellite they'll lunch you know, 20 small ones. Yeah. So if one gets taken out that's okay. But the problem is that you know, one good sized satellite getting hit Um that's like a ballistic event that turns into 10,000 pieces of debris that then are the things that go and hit the other satellites? Yeah. </s><s>[INST] Do you see a world where like in your sense in your own work and just in the space industry in general, do you see that people are moving towards bigger satellites are smaller satellites? Is there going to be a mix like what and what are we talking, what does it mean for a satellite to be big and was small. What's that [/INST] big? The space industry p Prior to, I don't know, 1990, you know, I guess the bulk of the majority of satellites were the size of a school bus and costs a couple of billion dollars. Um And now our first launches were on um satellites the size of shoe boxes that were built by high school students. So that's very different, you know, to give you the two ends of the spectrum. Um Big satellites will I think they're here to stay at least as far as I can see into the future. Um for things like broadcasting, you want to be able to broadcast to as many people as possible. Um There you also can't just go to small satellites um and say moore's law for things like optics. So if you have an aperture on your satellite, you know, that just that doesn't follow moore's law, that's that's different. So it's always going to be the size that it will be, you know, unless there's some new physics that comes out that I'm not aware of, but if you need a resolution and you're at an altitude that kind of sets your the size of your telescope. Um but because of moore's law, we are able to do a lot more with smaller packages and with that comes more affordability and opening up access to space to more and more people? </s> |
<s>[INST] you just briefly, because you used this term a lot and be nice to try to get a little more color to it, which is interception and extra reception. Uh what are we exactly talking about? So like what's included in each category? And how much overlap is there [/INST] into reception? Would be uh an awareness of anything that's within the confines or on the surface of my skin that I'm sensing </s><s>[INST] literally physiological, [/INST] physiological, like within the boundaries of my skin. And probably touch to the skin as well. Extra reception would be perception of anything that's X. Beyond the reach of my skin. So that that bottle of water um a sent um a sound although and this can change dramatically actually if you have headphones in you tend to hear things in your head if as opposed to a speaker in the room, there's actually the basis of ventriloquism. So there are beautiful experiments done by Greg reckon zone of at UC Davis looking at how auditory and visual cues are matched and we have an array of speakers and you can this will become obvious as I say it. But you know obviously the ventriloquist doesn't throw their voice, what they do is they direct your vision to a particular location and you think the sound is coming from that location. And there are beautiful experiments that Greg and his colleagues have done where they suddenly introduce an auditory visual mismatch. And it freaks people out because you can actually make it seem from a perception standpoint as if the sound arrived from the corner of the room and hit you like physically and people will recoil. And so sounds aren't getting thrown across the room, they're still coming from the defined location, an array of speakers. But this is the way the brain creates these internal representations. And again, not to I don't want to go down a rabbit hole, but um I think as much as I'm sure the listeners appreciate this but you know, everything in the brain is an abstraction, right? I mean they're they're the sensory apparatus, there are the eyes and ears and nose and skin and tasting all that are taking information and with interception. Taking information from sensors inside the body. The enteric nervous system for the gut. I've got sensory neurons that integrate my liver um etcetera. Taking all that. And the brain is abstracting that in the same way that if I took a picture of your face and I handed it to you and I'd say that's you you'd say yeah that's me. But if I were an abstract artist I'd be doing a little bit more of what the brain does. Where if I took a pad and paper, maybe I could do this because I'm a terrible artist and I could just mix it up. And let's say I would make your eyes like water bottles but I flip them upside down and I'd start assigning fruits and objects to the different features of your face. And I showed you I say lex that's you say well that's not me. And I'd say no but that's my abstraction of you. But that's what the brain does. The space time relationship of the neurons that fire that encode your face have no resemblance to your face. Right? And I think people don't really I don't know if people have fully internalized that but the day that I and I'm not sure I fully internalize that because it's weird to think about. But all neurons can do is fire in space and in time. Different neurons in different sequences perhaps with different intensities. It's not clear the action potential is all or none. Although people, neuroscientists don't like to talk about that, even though it's been published in nature a couple times, the action potential for a given neuron doesn't always have the exact same waveform people. It's in all the textbooks, but you can modify that way for </s><s>[INST] Well, I mean, there's a lot of fascinating stuff with with neuroscience about the fuzziness of all the uh of the transfer of information from neuron to neuron. I mean, we certainly touch upon it every time. We had all tried to think about the difference between artificial neural networks and biological neural networks. But can we maybe linger a little bit on this uh on the circuitry that you're getting at? So the brain is just a bunch of stuff firing and it forms abstractions that are fascinating and beautiful, like layers upon, layers upon layers of abstraction. And I think it just like when you're programming, I'm programming in python, it's uh it's all inspiring to think that underneath it all, it ends up being zeros and ones and the computer doesn't know about no stupid python or windows or Linux exit? It only knows about the zeros and ones in the same way with the brain. Is there something interesting to you are fundamental to you about the circuitry of the brain that allows for the magic that's in our mind to emerge? How much do we understand? I mean, maybe even focusing on the vision system. Is there something specific about the structure of the vision system, the circuitry of it that uh allows for the complexity of the vision system to emerge? Or is it all just a complete chaotic mess that we don't understand? [/INST] It's definitely not all a chaotic mess that we don't understand if we're talking about vision and that's not because I'm a vision scientist, ballistic division division. Well, because in the beauty of the visual system, the reason David Huebel and torn some weasel on the Nobel prize. It was because they were brilliant and forward thinking and adventurous and all that good stuff. But the reason that the visual system is such a great model for addressing these kinds of questions and other systems are hard is we can control the stimulate, we can adjust spatial frequency. How fine the grading czar, thick grading thin greetings, we can just temporal frequency, how fast things are moving. We can use cohen isolating stimuli we can use. There's so many things that you can do in a controlled way. Whereas if we're talking about cognitive encoding, like they, you know, encoding the space of concepts or something, you know, I you know, I like you if I may our am drawn to the big questions in neuroscience but I confess in part because of some good advice I got early in my career and in part because I'm not perhaps smart enough to go after the really high level stuff. I also like to address things that are tractable and I want, you know, we need to we need to address what we can stand to make some ground on at a given </s> |
<s>[INST] a unit test will catch it in some way. Yeah, in my private code, I have asserts everywhere. There's something pleasant to me, pleasurable to me about sort of the dictatorial rule of like, like, this should be true at this point. In too many times, I've made mistakes that shouldn't have been made. And I would assume I wouldn't be the kind of person that would make that mistake, but I keep making that mistake. Therefore, an assert really catches me, really helps all the time. So my code, I would say like 10 to 20% of my private code just for personal use is probably [/INST] asserts. And they're active comments. It's one of those things that in theory, they don't make any difference to the program. And if it was all operating the way you expected it would be, then they will never fire. But even if you have it right and you wrote the code right initially, then circumstances change. The world outside your program changes. And in fact, that's one of the things where I'm kind of fond in a lot of cases of static array size declarations where I went through this period where it's like, okay, now we have general collection classes, we should just make everything variable. Because I had this history of in the early days, you get doom, which had some fixed limits on it, then everybody started making crazier and crazier things, and they kept bumping up the different limits, this many lines, this many sectors. And it seemed like a good idea, well, we should just make this completely generic, it can go kind of go up to whatever. And there's cases where that's the right thing to do. But it also, the other aspect of the world changing around you is it's good to be informed when the world has changed more than you thought it would. And if you've got a continuously growing collection, you're never going to find out you might have this quadratic slowdown on something where you thought, oh, I'm only ever going to have a handful of these, but something changes, and there's a new design style, and all of a sudden, you've got 10,000 of them. So I kind of like in many cases, picking a number, some nice round power of two number, and setting it up in there and having an assert saying it's like, hey, you hit the you hit this limit, you should probably think are the choices that you've made around all of this still relevant, if somebody's using 10 times more than you thought </s><s>[INST] they were those things that in theory, they don't Yeah, this code was originally written with this kind of worldview, with this kind of set of constraints, you were you were thinking of the world in this way. If something breaks, that means you got to rethink the initial stuff. And it's nice for it to afford to afford to do that. Is there any stuff like keyboard or [/INST] or monitors or monitors? Barely pedestrian on a lot of that where I did move to triple monitors, like in the last several years ago, I had been dual monitor for a very long time. And I am. And it was one of those things where probably years later than I should have, I'm just like, well, the video cards now generally have three output ports, I should just put the third monitor up there. That's been a that's been a pure win. I've been very happy with that. But no, I don't have fancy keyboard or mouse </s><s>[INST] or anything. The key things is, the key things is an ID that has helpful debuggers has helpful [/INST] tools. So it's not the Emacs VIM route and then Diet Coke. So I did spend, you know, I spent I had one of my week long retreats where I'm like, okay, I'm going to make myself use is actually classic VI, which I know people will say you should never have done that. You should have just used them directly. But, you know, I gave it the good try. It's like, okay, I'm being in kind of classic Unix developer mode here. And I worked for a week on it. I used Anki to like teach myself the different little key combinations for things like that. And in the end, it was just like, all right, this was kind of like my Civil War reenactment phase, you know, it's like I'm </s><s>[INST] going out there doing it like they used to in the old days. And it was kind of fun. So many people [/INST] right now screaming as they're listening to this. So again, the out is this was not modern VIM, </s><s>[INST] but still, yes, I was very happy to get back to my visual studio at the end. Yeah, I'm actually, I struggle with this a lot, because there's a Kinesis keyboard and I use Emacs primarily. And I feel like I can exactly as you said, I can understand the code, I can navigate the code. There's a lot of stuff you can build within Emacs with using Lisp, you can customize a lot of things for yourself to help you introspect to code, like to help you understand the code and visualize different aspects of the code, you can even run debuggers. But it's it's work. And the world moves past you and the better and better ideas are constantly being built and that that puts a kind of I need to take the same kind of retreat as you're talking about but now I'm still fighting [/INST] the civil war I need to kind of move into the 21st century and it does seem like the world is or a large chunk of the world is moving towards visual studio code which is kind of interesting to me against the JavaScript ecosystem on the one hand and IDs are one of those things that you want to be infinitely fast you want them to just kind of immediately respond and like I mean heck I've got there's someone I know I'm an old school game dev guy that still uses visual studio six and on a modern computer everything is just absolutely instant on something like that because it was made to work on a computer that's 10,000 or 100,000 times slower so just everything happens immediately and all the modern systems just feel you know they feel so crufty when it's like oh why is this refreshing the screen and moving around and updating over here and something blinks down there and you should update this and there's you know there there are things that we've lost with that incredible flexibility but I have lots of people get tons of value from it and I am super happy that that seems to be winning over even a lot of the old vim and emacs people that they're kind of like hey visual studio codes may be you know not so bad I am that may be the final peacekeeping solution where everybody is reasonably happy with something like that. </s> |
<s>[INST] you think there was [/INST] pretty like once life </s><s>[INST] on MArs or do you think there's currently, [/INST] I'm more comfortable saying probably 3.5 billion years ago. I feel pretty confident there was life on Mars just because then it had an electromagnetic shield. It had an atmosphere has a wonderful gravity level 33 H. G. Is fantastic. You know, you're super human human. We can all slam dunk in basketball. It's going to be fun to play sports on wires. But so I think we'll find past the fossilized. Probably the evidence of past life on MArs. Currently that's again we need the next decade. The evidence is mounting for sure. We do have the organics. We're finding organics. We have water, seasonal water on MArs we used to just know about the ice caps, North and South pole. Now we have seasonal water. We do have the building blocks for life on MArs. We really need to dig down into the soil because everything on the top surface is radiated. But once we find down, will we see any any life forms? We see any bugs. I leave it open as a possibility. But I feel pretty certain that past life or fossilized life forms will find. And then we have to get to all these ocean worlds. These these beautiful moons of other planets since we know they have water. And we're looking for a simple search for life follow the water. Carbon based life. That's the only life we know there could be other life forms that we don't know about, but it's hard to search for them because we don't know. So, in our search for life in the solar system is definitely, you know, search follow the water and look for the building blocks of life. </s><s>[INST] Do you think in the next decade we might see hints of past life or even current [/INST] life? That's a pretty optimism. I'm pretty active. </s><s>[INST] Humans have to be involved or can this be robots and rovers And [/INST] Probably teams. I mean we've been at it on Mars in particular. 50 years. We've been exploring Mars for 50 years. Great data. Right. Are images of Mars today are phenomenal. Now we know how Mars lost its atmosphere are starting to know because of the Lack of the Electromagnet Shield. We know about the watermarks. So we've been studying 50 years with our robots. We still haven't found it. So I think once we have a human mission there we just accelerate things and it's always humans and our rovers and robots together. But we just have to think that 50 years we've been looking at mars mars and um taking images and doing the best science that we can. People need to realize MArs is really far away. It's really hard to get to this extreme extreme exploration. We mentioned Magellan first or all of the wonderful explorers and sailors of the past, which kind of a lot of my inspiration for exploration, mars is a different ball game. I mean eight months to get there a year and a half to get home. I mean it's really </s><s>[INST] extreme environment in all kinds of ways. But the kind of organism might be able to see hints off on MArs are kind of microorganisms perhaps [/INST] remember that humans were kind of were host, right? Were host to all of our bacteria and viruses. Right. Do </s><s>[INST] you think it's a big leap from the viruses and the bacteria to us humans. Put another way do you think on all those moons, beautiful wet moons that you mentioned. You think there's intelligent life out there? [/INST] I hope so. I mean that's that's the hope. But you know, we don't have the scientific evidence for that now. I think all the evidence we have in terms of life existing is much more compelling again because we have the building blocks of life. Now when that life turns into intelligence, that's a big unknown. </s><s>[INST] If we ever meet, do you think we would be able to find a common language? [/INST] I hope so. We haven't met yet. It's just so far. I mean, do physics display a role here? Look at all these exoplanets, 6000 exoplanets. I mean even the couple dozen earthlike planets or exoplanets that really look like habitable planets. These are very earth like they look like they have all the building blocks. I can't wait to get there. The only thing is they're 10-100 light years away. So scientifically, we know they're there, we know that their habitable they have, you know, everything going from right, you know, called the goldilocks zone. Not too hot, not too cold. Just perfect for how habitability for life. But now the reality is if there are 10 at the best 200 to thousands of light years away. So what's out there, but I just can't think that we're not the only one. So absolutely life life in the universe. Probably intelligent life as well. </s><s>[INST] Do you think there needs to be fundamental revolutions and how we the tools we use to travel through space in order for us to venture outside of our solar system? Or do you think the ways the rockets, the ideas we have now, the engineering ideas we have now will be enough to venture out? [/INST] Well that's a good question right now because the speed of light is the limit. We don't have warp speed warp drive to explore our solar system to get to MArs to explore all the planets. Then we need technology push. But technology push here is just advanced propulsion would be great if I could get humans to MArs and say you know, 3 to 4 months, not eight months. I mean have the time. 50% reduction. That's great in terms of safety and wellness of the crew, orbital mechanics. Physics rules, you know, orbital mechanics are still physics rules. We can't defy physics </s> |
<s>[INST] Yes, but this mix of denial. I mean, the I think the book is called denial of death. Is this constant denial that we're running away from it. That's uh and in fact, some would argue that the inspiration, the incredible ideas you've put out there. Your talk has been seen by millions and millions of people, right? It's just you trying to desperately fight the fact that you are biologically mortal into the your creative genius comes from the fact that you're trying to create ideas that live on long past you. Well, [/INST] that's very nice of you. I mean, I I I would like my ideas to live on beyond me because I think that is a good test that those ideas have value, have value in the lives of others. I think that's a good test that others would continue to talk about or share the ideas long after I'm gone, I think is perhaps the greatest Compliment one can get for one's own work. Uh it's very but I don't think it's my awareness of my mortality that drives me to do it. It's my desire to contribute that drives me to do it. </s><s>[INST] It's the optimal, it's the optimist vision. It's the uh the pleasure and the fulfillment you get from inspiring others. It's as pure as that is. Uh let me ask, listen, I'm Russian, I'm trying to get you get you into these dark [/INST] areas I'm enjoying. It </s><s>[INST] is the ego tied up into it somehow. So your name is extremely well known if your name wasn't attached to it. Do you think you would act differently? [/INST] I mean for years I hated that my name was attached to it. You know, I had a rule for years that I wouldn't have my, my face on the cover the front page of the website. You know, I had a fight with the publisher because I didn't want my name big on the book. I wanted it tiny in the book because I kept telling him it's not about me. It's about the ideas they want to put my name on the top of my book. I refused. None of my books have my name's on the top because I won't, I won't let them. They would like very much to put my name on the top of the book. But the idea has to be bigger than me. I'm not bigger than the idea. </s><s>[INST] That's beautifully put. Do you think ego? [/INST] But I also am aware that I have become and become recognized as the messenger. And even though I still think the message is bigger than me, I recognize that I have a responsibility as the messenger and whether I like it or not is irrelevant. I accept, I accept the responsibility. I, I'm happy to do it. </s><s>[INST] I'm not sure how to phrase this, but there's a large part of the culture right now that emphasizes all the things that nobody disagrees with, which is health, sleep, diet, relaxation, meditation, vacation are really important and there's no, you know, it's like you can't really argue against that. In fact people less sleep. That's just, I'm joking. Yes. Well that's the thing. I often, I often speak to the fact that passion and love for your, what you're doing and the two words hard work, especially in the engineering fields are more important than are more important to prioritize than sleep even though sleep is really important. Your mind should be obsessed with the hard work with the passion and so on. And then I get some pushback of course from people. What do you make sense of that? Is that just me, the crazy Russian engineer really pushing hard work probably. [/INST] Uh, what do you think that that's a short term strategy? I think if you sacrifice your health for the work at some point it catches up with you and at some point it's like it's like going, going, going and you get sick, your body will shut down for you if you refuse to take care of yourself. You know, you get sick. It's what happens sometimes, you know, more severe illness than something that just slows you down. So I think, I think taking getting sleep. I mean there've been studies on this that, you know, executives for example, who who get a full night's sleep and stop at unreasonable hour actually accomplish more are more productive than people who work and burn the midnight oil because their brains are working better because they're well rested, so you know, working hard. Yes, but one that works smart. Um, uh, I think that um, giving our minds and our bodies rest makes us more efficient. Um, I think just driving, driving, driving, driving is it's a short term short term </s><s>[INST] strategy. So, but to push back on that a little bit, the annoying thing is you're like 100% right in terms of science, right? But the thing is it's because you're 100% right, That weak part of your mind uses that fact to convince you like what? So, you know, I, I get all kinds of, my mind comes up with all kinds of excuses to try to convince me that I shouldn't be doing what I'm doing to rationalize, to rationalize. And so what I have a sense you, I think what you said about executives and leaders is absolutely right. But there's the early days, the early days of madness and passion for Sure. Then I feel like emphasizing sleep, thinking about a sleep is giving yourself a way out from the fact that those early days, especially it can be suffering [/INST] as long as it's not sustainable. You know, it's not sustainable. Sure if you're investing all that energy in something at the beginning to get it up and running, then at some point you're gonna have to slow down or your body will slow you down for you like you can choose, your body can choose. I mean, so, </s> |
<s>[INST] probably like negatively affecting your ability to calculate how much, show when you guys play, when you look at the chessboard, how much of it is calculation how much of it is intuition, how much of it has memorized openings. Um [/INST] it really depends between short form chest, so five minutes, three minutes, one minute and classical chest. </s><s>[INST] What's your favorite? I [/INST] love playing blitz now because that's most of what I do and that's actually how I got into chess streaming because I couldn't spend entire weekends or weeks playing tournaments. I would just while I was in college log on and play these long blitz or bullet sessions and it's very fast. So you don't have time to go calculate as deeply. You basically have to calculate short lines pretty quickly and a lot of it is pattern recognition and intuition. Three </s><s>[INST] minutes you said [/INST] three minutes. Yeah. </s><s>[INST] Okay, cool. And so for that it's just basically intuition a [/INST] lot of it is intuition. Yeah. </s><s>[INST] See I saw on the streams you actually keep talking while playing chess. It seems [/INST] my result, that doesn't help my results </s><s>[INST] doesn't [/INST] content not the game. Yeah, exactly, </s><s>[INST] but you can still do it because i i it feels like how can you possibly concentrate while talking? [/INST] It's because so much of it is intuition you're not while you're talking, you're thinking about that topic, but then you just come to the board and you just understand what you should be doing here and then sometimes you get in trouble because you're talking and you have now lost half of your time, you have a minute and a half your opponent has three and you're kind of at a disadvantage um but that kind of goes to show that that's how blitz chess usually works, whereas classical is very different </s><s>[INST] which of you is better at chess, let's do it this way can you um Andrea can you say what in which way is Alex stronger than you? Which way is she weaker than you? Not physically? In in terms of in terms of chess? [/INST] Um Well yes of course she is higher rated but when we do play um I think her strength against me where she really gets me is the end game, she has stronger endgame so she can and I actually have a stronger opening um But as soon as she's I'm supposed to say what is good about you not you know I'm getting there we'll see this I'm saying </s><s>[INST] because it's [/INST] related okay because if I can I can get an advantage in the beginning of the game but as soon as she starts trading pieces down like my confidence drops because I know that the end game is the hardest part of the game and the longest and that's where she ends up beating me. So her end game is her I think really what makes the difference and she </s><s>[INST] sounds like her psychological warfare is better too because if you're getting nervous [/INST] that it's harder to play against higher rated players? Same how you know magnus and former world championship champions, how that psychological edge. So I think it's always going to be different friendly because she knows statistically she should be winning something like one in four games. But she usually does better than that because she's very distracting and talks a lot. That does </s><s>[INST] help. What does it feel like to play a higher rated player? What's the experience of that in like uh playing somebody like magnus? [/INST] So it depends on how much higher rated than you they are if it's someone who's like between me and Andrea let's say it's a 200 point difference. You know they should win. But at least you still feel like you have a chance. I was playing in a title Tuesday which is this tournament chess dot com has every Tuesday and I got really lucky beat a G. M. Drew an international master and then I got paired against Hikaru Nakamura and my brain just went blank because I just know that I'm so unlikely to win that I couldn't even play the game properly when it's that much of a difference where they should be winning like 99% of the time </s><s>[INST] but that's like psychological so you're saying that's the biggest experiences like actually knowing the numbers and statistically thinking there's no way I can win but I meant like is there a suffocating feeling like positional e you feel like you're constantly under attack, you [/INST] just feel like you're slowly getting outsmarted and the worst is when you don't even know what you're doing wrong, you come out of that and you're like I thought I was doing great and I got slowly squeezed, I didn't understand what was going on and you're just kind of baffled. It's kind of like watching Alpha zero beat up stock fish and you don't really understand why it's making certain moves or how it thought of the plan, you just see it slowly getting the position better and that's what it feels like. I would add, it's kind of different for me if there's someone who's significantly higher, rated so let's say more than like three point. So you're playing magnus, what I notice is I just feel lost straight as soon as I don't know my preparation because they know so many opening lines that they're going to know the best line to beat you that you haven't studied. So then on move 10 you're like he already has a maybe plus 100.5 advantage which is really small but for someone with such a significant skill level, you know, you already lost at that point and it's like a third of the game, what </s><s>[INST] are the strengths and weaknesses of Andrea [/INST] Andrea is very good at opening preparation, </s><s>[INST] She said, she [/INST] said she likes bringing that up, I mean she's very meticulous about it where she'll really go in and learn her lines. Um And having that initial starting confidence isn't just helpful for the opening, but it helps develop your plans for the middle game. So I think she's very good at that. Um I think she's actually pretty good at tactical combinations. Um </s> |
<s>[INST] think that once we crack it, we'll be able to use assembly theory plus telescopes to find aliens. Do you have, Sarah, do you have disagreements with Lee on the number of aliens that are out there? Do actually, yeah. Well, and what they look like. So any of the things we've been talking about is they're nuanced. It's always nice to discover wisdom through nuanced disagreement. [/INST] I do actually, yeah. Well, yeah, I don't wholly disagree, but I do think I disagree. There's nuance there. You can disagree. No, it's fine. It is nuanced, right? So you made the point earlier that you think, once we discover what life is, we'll see alien life everywhere. And I think I agree on some levels in the sense that I think the physics that governs us is universal. But I don't know how far I would go to say to say that we're a likely phenomena because we don't understand all of the features of the transition at the origin of life, which we would say in assembly as you go from the no memory physics to there's like a critical transition around the assembly index where assembly and starts to increase. And that's what we call the evolution of the biosphere and brotherification of the biosphere. So there's a principle of increasing and that goes back to what I was saying at the very beginning about the physics of the possible that the universe basically gets in this mode of trying to make as much possibilities as possible. Now, how often that transition happens, that you get the kind of cascading effect that we get in our biosphere, I think we don't know. If we did, we would know the likelihood of life in the universe. And a lot of people want to say life is common, but I don't think that we can say that yet until we have the empirical data, which I think you would agree with. But then there's this other kind of thought experiment I have, which I don't like, but I did have it, which is, you know, if life emerges on one planet, and you get this real high density of things that can exist on that planet, is it sort of dominating the density of creation that the universe can actually generate? So like, if you're thinking about counting entropy, right, like the universe has a certain amount of stuff in it. And then, you know, assembly is kind of like an entropic principle, it's not entropy. But the idea is that now transformations among stuff are the actual physical histories of things now become things that you have to count as far as saying that these things exist, and we're increasing the number of things that exist. You can disagree. And if you think about that cosmologically, maybe Earth is sucking up all the life potential of the whole universe, I don't know. </s><s>[INST] How's that? Can you explain that a little bit? Why can any one geographical region suck up the creative capacity of the universe? [/INST] Well, I know it's a ridiculous thought. I don't actually agree with it, but it was just a thought experiment. </s><s>[INST] I love that you can have thoughts that you don't like and don't agree with, but you have to think through them anyway. [/INST] The human mind is fascinating. Yeah, I think these sort of like counterfactual thought experiments are really good when you're trying to build new theories, because you have to think through all the consequences. And there are people that want to try to account for, say, the degrees of freedom on our planet in cosmological inventories of talking about the entropy of the universe, and when we're thinking about cosmological arrow of time and things like that. Now, I think those are pretty superficial proposals as they stand now, but assembly would give you a way of counting it. And then the question is, if there's a certain maximal capacity of the universe's speed of generating stuff, which Lee always has this argument that assembly is about time, the universe is generating more states, really what it's generating is more assembly possibilities. And then dark energy might be one manifestation of that, that the universe is accelerating its expansion, because that makes more physical space. And what's happening on our planet is it's accelerating in the expansion of possible things that exist. And maybe the universe just has a maximal rate of what it can do to generate things. And then if there is a maximal rate, maybe only a certain number of planets can actually do that or there's a trade off about the pace of growth on certain planets versus others. </s><s>[INST] I have a million questions there, but do you have thoughts on... [/INST] Just a quick, yeah, I'll just say something very quick. It's a thought experiment. No, it's good. I think I get it. I think I get it. So, what I want to say is, when I mean aliens are everywhere, I mean memories are the prerequisite for aliens via selection and then the concentration of selection when selection becomes autonomous. So what I would love to do is to build, say, a magical telescope that was a memory... Magical. Magical one. Or a real one. Yeah. There would be a memory detector to see selection. So you could get to exoplanets and say, that exoplanet looks like there's lots of selection going on there. Maybe there's evolution and maybe there's going to be life. So what I'm just trying to say is narrow down the regions of space where you say, there's definitely evidence of memory as high assembly there, or not high assembly, because that would be life, but where it's capable of happening. And then that would also help us frame the search for aliens. I don't know how likely it is to make the transition to cells and all the other things. I think you're right, but I think that we just need to get more data. Yeah. Well, I didn't like the thought experiment because I don't like the idea that if the universe has a maximal limit on the amount it can generate per unit time that our existence is actually precluding the existence of other things. Well, I'll just say one thing. But, I think that's probably true anyway, because of the resource limitations of other things. So I don't like your thought experiment because I think it's wrong. No, I do like the thought experience. you try to say is like there is a chain of events that goes back that's manifestly co-mated with life on Earth, and you're not saying that life isn't possible elsewhere, you say that there has been these number of things, contingent things that have happened that have allowed life to merge here. That doesn't mean that life can't emerge elsewhere, but you're saying that these, the intersection of events may have, may be concentrated here, right? And I think there's, um, not exactly. It's more like, um, like, uh, you know, if you, you look at, say the causal graphs are fundamental, maybe space is an emergent property, which is consistent with some proposals in quantum gravity, but also how we talk about things in assembly theory, then the universe is causal graphs generating more structure in causal graphs, right? So this is how the universe is unfolding. And maybe there's a cap on the rate of generation, like there's only so much stuff that gets made per update of the universe. And then if there's a lot of stuff being made in a particular region that happens to look the same locally spatially, that's an after effect of the fact that the whole causal graph is updating. Like it's, uh, yeah, I don't, I don't know that. I think that that doesn't work. I don't think it works either, but I don't have a good argument in my mind about, but I do like the idea of the capacity that universe, because you've got the number of </s> |
<s>[INST] So, I have to ask you on the capitalism side, you mentioned that technology is killing capitalism or has effects that are uh unintended, well not unattended, but not what economists would predict or speak of capitalism creating. Uh I just want to talk to you about in general, the effect of even an artificial intelligence or technology, automation, taking away jobs and these kinds of things and what you think is the way to alleviate that. Whether the Andrew and presidential candidate with universal basic income UBI, I whether your thoughts there, how do we fight off the negative effects of technology [/INST] that your software guy, right, yep, a human being is a worker is an old idea, human being has a worker is a different object, right? So if you think about object oriented programming is a paradigm, a human being has a worker and a human being has a soul. We're talking about the fact that for a period of time the worker that a human being has was in a position to feed the soul that a human being has. However, we have two separate claims on the value in society. One is as a worker and the other is as a soul and the soul needs sustenance. It needs dignity, it needs meaning, it needs purpose. As long as you're means of support is not highly repetitive. I think you have a while to go before you need to start worrying. But if what you do is highly repetitive and it's not terribly generative. You are in the crosshairs of 44 loops and wild loops, and that's what computers excel at repetitive behavior. And when I say repetitive, I may me may mean things that have never happened through combinatorial possibilities, but as long as it has a looped characteristic to it, you're in trouble. We are seeing a massive push towards socialism because capitalists are slow to address the fact that a worker may not be able to make claims, a relatively undistinguished median member of our society still has, needs to reproduce, needs to, uh, to dignity. And when capitalism abandons The median individual or, you know, the bottom 10th or whatever it is going to do it's flirting with revolution. And what concerns me is that the capitalists aren't sufficiently capitalistic to understand this. You really want to court authoritarian control in our society because you can't see that people may not be able to defend themselves in the marketplace because the marginal product of their labor to too low to feed their dignity as a soul. So it's my great concern is that our free society has to do with the fact that we are self organized. I remember looking down from my office in Manhattan when Lehman Brothers collapsed and thinking, who's going to tell all these people that they need to show up at work when they don't have a financial system to incentivize them to show up at work. So my complaint is first of all, not with the socialists, but with the capitalists, which as you guys are being idiots, you're courting revolution by continuing to harp on the same old ideas that well, try harder bootstrap yourself to an extent, that works to an extent. But we are clearly headed in a place that there's nothing that ties together our need two contribute and our need to consume. And that may not be provided by capitalism because it may have been a temporary phenomena. So check out my article on anthropic capitalism and the new gimmick economy. I think people are late getting the wake up call and we would be doing a better job saving capitalism from itself because I don't want this done under authoritarian control. And the more we insist that everybody who's not thriving in our society during their reproductive years in order to have a family is failing at a personal level. I mean, what a disgusting thing that we're saying? What horrible message? Who, who the hell have we become that we've so bought into the Chicago model, um, that we can't see the humanity that we're destroying in that process. And it's I hate, I hate the thought of communism. I really do. My family has flirted with in decades past. It's a wrong bad idea. But we are going to need to figure out how to make sure that those souls are nourished and respected and capitalism better have an answer. And I'm betting on capitalism, but I gotta tell you, I'm pretty disappointed with my team. </s><s>[INST] So you're still on the capitalism team. Just, uh, there's a theme here radical capital, hyper capitalism. [/INST] And I want, I think hyper capitalism is going to have to be coupled to hyper socialism. You need to allow the most productive people to create wonders and you've got to stop bogging them down with all of these extra nice requirements. You know, Nice is dead. Good has a future. Nice doesn't have a future because nice ends up with goo legs. </s><s>[INST] Damn that's a good line. Okay last question you tweeted today a simple, quite insightful equation saying imagine that every unit F of fame you picked up as stalkers and h haters. So I imagine S and H are dependent on your path to fame perhaps a [/INST] Little bit. It's not a simple people always take these things literally when you have like 280 characters to explain </s><s>[INST] yourself. You mean that's not a mathematical, there's no law. Okay. I [/INST] put the word imagine because I still have a mathematician's desire for precision. Imagine if this were true. </s><s>[INST] But there was a beautiful way to imagine that there is a law that has those variables in it. And uh you've become quite famous these days. So how do you yourself optimized that equation with the peculiar kind of fame that you have gathered along the way? [/INST] I want to be kinder, I want to be kinder to myself. I want to be kinder to others. I want to be able to have heart compassion and these things are really important and I have a pretty spectrum kind of approach to analysis. I'm quite literal. I can go full rain man on you at any given moment. No, I can I can its faculty of autism if you like And people are gonna get angry because they want autism to be respected. But when you see me coding or you see me doing mathematics. I'm you know, I speak with speech Apnea, uh, be right down to it. You know, we have to try to integrate ourselves and those tensions between, you know, it's sort of back to us as a worker and us as a soul. Many of us are optimizing one to be at the expense of the other. And I struggle with social media and I struggle with people making threats against our families and I struggle with um just how much pain people are in. And if there's one message I would like to push out there, um irresponsible everybody, all of us, myself included with struggling struggle struggle mightily because you it's nobody else's job to do. Your struggle for you. Now with that said, if you're struggling and you're trying and you're trying to figure out how to better yourself and where you failed, Where you've let down, your family, your friends, your workers, all this kind of stuff. Give yourself a break. You know, if if if it's not working out, I have a lifelong relationship with failure and success, There's been no period of my life. We're both haven't been present in one form or another. And I I do wish to say that a lot of times people think this is glamorous. I'm about to go to a show with Sam Harris. People are gonna listen in on two guys having a conversation on stage. It's completely crazy when I'm always trying to figure out how to make sure that those people get maximum value. And uh that's why I'm doing this podcast. You know, just give yourself a break. You owe us. You always your struggle. You don't owe your family or your co workers or your lovers or your family members success. Um, as long as you're in there and you're picking yourself up, recognize that this this new situation with the economy that doesn't have the juice to sustain our institutions has caused the people who have risen to the top of those institutions to get quite brutal and cruel. Everybody is lying at the moment. Nobody is really a truth teller. Um try to keep your humanity about you try to recognize that if you're failing, if things aren't where you want them to be and you're struggling and you're trying to figure out what you're doing wrong, which you can do. It's not necessarily all your fault. We are in a global situation. I have not met the people who are honest, kind good, successful. Nobody that I've met This chick is checking all the boxes. Nobody is getting all 10s. So I just think that's an important message that doesn't get pushed out enough either. People want to hold society responsible for their failures, which is not reasonable. You have to struggle. You have to try Or they want to say you're 100% responsible for your failures, which is total nonsense, </s> |
<s>[INST] uh Blood Lands. But by timothy Snyder, probably the most upsetting book I've ever read if you want to because it's not just Stalin or Hitler Stalin and Hitler, it's the worst, the worst hits, the opposite of the best hits. It's really, really, really well written, really difficult. Um I read uh Solzhenitsyn's um gulag archipelago and what else? Red Famine, which is um um and Applebaum, does that, does that hurt? Yeah. Anyway, so those are truly upsetting and that's and those are a lot of times the results of um hiding the truth versus pursuing the truth. So, truth in the short term might hurt, but it did ultimately set this free. I [/INST] believe that. And I also think um whatever the truth is, we have to find a way to maintain civil society and love and all the things that are important to us, </s><s>[INST] if you can jump around a little bit can I just ask you on a personal note because you said you've suffered from depression and there's a lot of people that see guidance on this topic because it's such a difficult one. How were you able to when it was when it has struck you? How were you able to overcome it? [/INST] Yeah. I mean that's just maybe too long an answer. So I've experienced it in different forms. So it was my I would say my depression has almost always mostly taken the form of anxiety. I didn't realize how anxious I was I think until I was an adult so I was always very functional. Um I think you know all the positive sides of suffering in that way. Um I think I'm a little O. C. D. As until </s><s>[INST] this whole conversation is hilarious because we're both suffered to some level of anxiety [/INST] psychology is just laid out in front of us here </s><s>[INST] a giant mess [/INST] of human, </s><s>[INST] great [/INST] just trying trying to organize, </s><s>[INST] hold on like the Tom waits song. [/INST] Um But then I I suffered from postpartum depression after both of my daughters after both pregnancies. Um That was a very different experience from anything I've ever experienced. But clearly I had a predisposition towards you know suffering from something like that. Um Anyway what it really wasn't until I fully recovered from the second experience of postpartum depression that I realized That I had been suffering on some level my whole life and I think I always knew I I thought of myself as a very sensitive person, an empathic person. Um I mean I've been in therapy for 10 years. Um I knew I had a lot of anxiety, I would never have denied that I had a lot of anxiety. I just didn't realize it crossed over into a disorder really until I was an adult and ended up taking um PROzac, I took an ss ri for postpartum and it was fascinating to me, I ended up interviewing my psychiatrist because I was so fascinated in the whole thing. Once I was on the other side of it, just what I had been through, how different I felt during that period of time and then how quickly the medication made me feel like myself again, I had come out the other side of the experience of postpartum um and was going to start tapering off the medication and in this window where I no longer had postpartum depression um and hadn't yet gone off the S. S. R. I. I realized that life was not only a lot easier than when I had postpartum, but it was easier than it had ever been, and it took taking all of that anxiety away to recognize how much I had been grappling with it my entire life. And it first started coming in the form of realizations like, oh is this how other </s><s>[INST] people, is [/INST] this how other people feel? Is this how that like the things that I just always thought of myself, I'm really sensitive, I I'm an introvert, I need a lot of time to myself, and all of these things that I felt like. Um I mean, it's always very high functioning. Um and in some ways, you know, I was a professional dancer and I think that was the type of therapy for me, there was the obsessing over um The training and dancing nine hours a day and all of that, I now look back on and see how much that was um therapeutic for me. Um and that I was kind of treating something. But yeah, it was so it was just this experience of um treating an anxiety disorder that caused me to realize that I had one, </s><s>[INST] I didn't know I could [/INST] feel the way I felt after taking PROzac and I became very interested in. I mean, I was already working with neuroscientists, I was always already interested in consciousness in the brain, and um it just, you know, this kind of rattled other intuitions for me in terms of how our child, how our childhoods shape who we become. Um because I had been convinced my father was, my father </s><s>[INST] was I was [/INST] just gonna say again, But I think, I mean, so he he was not diagnosed, I think he was born, I think he had borderline personality disorder and was emotionally abusive and I thought that all of the ways I experienced the world and all of my anxiety and my sensitivities. Um, I thought almost all of that, if not all of that was because of these experiences I had growing up and trauma that I experienced as a child and obviously those things play a part. But what I realized um after going through postpartum and then the thing that was extremely informative to me was having my own Children, um, because they were basically living my dream childhood. They had none of the things that I thought were the cause of the psychological suffering that I experienced. Um, there was none of that and they have a lot of the same, um, they struggle with a lot of the same anxiety and panic attacks and what I realized was how much were kind of born into the world with these things that we, that we struggle with and with our strength and with all of that. And of course then if you have an abusive childhood, if you're, if you're someone </s> |
<s>[INST] And uh to one side note, first of all, it's an exceptionally popular video game. The side effect is that there's a lot of incredible human experts at that video game. So the benchmark that you're trying to reach is very high and the other, can you talk about the approach that was used initially and throughout training these agents to play this [/INST] game? And so they personally used is self play. And so you have two agents, they don't know anything, They battle each other, they discover something a little bit good and now they both know it and they just get better and better and better without bound. And that's a really powerful idea, right? Uh that we then went from the one versus one version of the game and scaled up to five versus five, right? So you think about kind of like with basketball where you have this like team sport, you need to do all this coordination. Um and we were able to push the same idea, the same self play too, To really get to the professional level at the full five versus 5 version of the game. And uh and and the things I think are really interesting here is that these agents in some ways they're almost like an insect like intelligence right? Where, you know, they have a lot in common with how an insect is trained, right? Insect kind of lives in this environment for a very long time or you know that the ancestors of this insect has been around for a long time and had a lot of experience that gets baked into into into this agent. And you know it's not really smart in the sense of a human rights not able to go and learn calculus but is able to navigate its environment extremely well. It's able to handle unexpected things in the environment that's never seen before pretty well. Um And we see the same sort of thing with our Dota bots right? They're able to within this game they're able to play against humans which is something that never existed in its evolutionary environment totally different play styles from humans versus robots. And yet it's able to handle extremely well. And that's something I think was very surprising to us. Was something that doesn't really emerge from what we've seen with P. P. O. At smaller scale. Right? The kind of scale we're running this stuff out was uh you know I could take 100,000 CPU cores running with like 100 GPU s. It's probably about you know like you know something like hundreds of years of experience going into this body every single real day. And so that scale is massive and we start to see very different kinds of behaviors out of the algorithms that we all know and love </s><s>[INST] Daughter. You mentioned beat the world expert one v. 1 and then uh you didn't weren't able to win five U. Five this year at [/INST] the best players in the </s><s>[INST] world. So what's what's the comeback story? What's first of all talk through that exceptionally exciting event and what's what's the following months and this year looks like? [/INST] Yeah. Yeah. So well one thing that's interesting is that you know, we lose all the time because we so the Dota team at open ai we played the boat against better players than our system all the time. Or at least we used to write like you know, the first time we lost publicly was we went up on stage at the international and we played against some of the best teams in the world. Um And we ended up losing both games but We gave them a run for their money, right? That both games were kind of 30 minutes, 25 minutes and they went back and forth, back and forth back and forth. And so I think that really shows that were at the professional level. Um and kind of looking at those games, we think that the coin could have gone a different direction and we could have could have had some wins. And that was actually very encouraging for us. Um and you know, it's interesting because the international was at a fixed time, right? So we we knew exactly what day we were going to be playing and we pushed as far as we could as fast as we could. Two weeks later we had a boat that had an 80% win rate versus the one that played at T. I. Um so the March of progress, you know, you should think of as a snapshot rather than as an end state. Um and so in fact we'll we'll be announcing our our finals pretty soon. I actually think that will announce our final match prior to this podcast being released. So there should be uh we'll be playing, we'll be playing against the World champions. And you know, for us, it's really less about like the way we think about what's upcoming is the final milestone, the file competitive milestone for the project, right. That our goal in all of this isn't really about beating humans at Dota. Our goal is to push the state of the art in reinforcement learning and we've done that right? And we've actually learned a lot from our system and that we have uh you know, I think a lot of exciting next steps that we want to take. And so, you know, kind of the final showcase of what we built, we're going to do this match, but for us it's not really the success or failure um to see, you know, do do do do we have the coin flip, go in our direction are against. </s><s>[INST] Where do you see the field of deep learning heading in the next few years. What do you see the work and reinforcement learning perhaps heading and uh [/INST] more specifically </s> |
<s>[INST] money or attention or fame. None of that. That's not the primary prey. [/INST] Well, that's what's so appealing to me To me in particular about him, just like and how he built. Like I read a biography of him and just like the way that he constructed his life and like we're able to hyper focus in meeting after meeting and drill down and also hire all the right people who execute each one of his tasks discreetly to his perfection is amazing. Like that's actually the mark of a good leader, but I mean if you think about his career, the reason he's a renegade because probably he was told to like put it in an index fund or whatever whenever he made his 29 million and from Paypal, I don't know how much you made and then just go along that and he's like no, you know succeeds spectacularly. So you have to have somebody who is willing to come in and buck that system. So for for for now I think our politics are generally frozen. I think that that model is going to be most generally appealing to the mean person, but somebody will come along and we'll change everything. </s><s>[INST] I'm just surprised. It's not more of them on that topic. It's now 20, what is it? 21? Yes, let's let's make some predictions. You can be wrong about what major political people are you Thinking will run in 2024 including Trump Junior Senior or Ivanka, I don't know any trump trump. Ah And who do you think wins? [/INST] I think Joe Biden will run again in 2024 and I think he will run against someone with the last name Trump. I do not know whether that is Trump or Trump jr but I think one of those people will probably be the GOP nominee in 2024. Who </s><s>[INST] is it? Some prominent political figure. Was that Romney. Somebody like that said that trump will win the primary few runs again. [/INST] Of course that's not even a question trump is the single most popular figure in the republican party by orders of magnitude. I mean probably more honestly, there was a actually, I can tell you because I saw the data which is that pre january 6th, it was like 54% of republicans wanted him to run again. Then it went down eight points after january 6th two days later and then after impeachment and went right back up to 54%. So the exact same number is in february at post impeachment vote as it was after november. Now look yeah, again surveys bullshit etcetera. But like that's all the data we have that striking pointing to If Trump runs, he will be the nominee and he will be he will be the 2024 nominee. I just don't know if he wants to. It really depends like </s><s>[INST] do you think he wins after the trump vaccine heals all of us? Do you think trump wins? [/INST] It depends on how popular culture functions over the next four years. And I can tell you that they are because I don't think biden has that much to do with it because again, trump is not a manifestation of an affirmative policy action. It is a defensive bulwark wall against cultural liberalism at its best. So it's like this is why it doesn't matter what biden does. If there are more riots, if there is a more sense of persecution amongst people who are more lean towards Conservative or like, hey, I don't know about that. That's crazy, then he very well could win. Let's, okay, let's say joe biden doesn't run and they put up like Kamala Harris. I think I think he would beat her. And I don't think there's a question that trump would be Kamala Harris in 2024 </s><s>[INST] and you don't think anybody else, I don't know how the process works. You don't think anybody else on the Democratic side can take the, [/INST] Well, how could you run against the sitting vice president? You know, it's like if Joe Joe Biden is a 98% approval rating in the Democratic Party, if he says she is my air, I think enough people will listen to him in a competitive primary or a non competitive primary. And then there's all these things about how primary systems themselves are rigged. The DNC could make it known that they'll blacklist anybody who does try and primary, Kamala Harris. Um and look, I mean, progressives aren't necessarily all that popular amongst actual democrats. Like we found that out. Um during the election, there's an entire constituency which loves joe biden and joe biden level politics and so if he tells them to vote for Kamala. I think, I think she would probably get it. But again, there's a lot of game theory obviously happening. </s><s>[INST] But I think you're talking about Everything you're saying is correct about Mediocre candidates, it feels like if there's somebody like a really strong, I don't want to use this term incorrectly, but populist, somebody that speaks to the 80% that's is able to provide bold, eloquently described solutions that are popular. I think that breaks through all of this [/INST] nonsense. How do they break through the primary system? Because the problem is the primary system is not populism, its primary. So it's like, </s><s>[INST] but you don't think they can tweet, they're way too [/INST] well, you have to be willing to win a GOP primary. You basically have to be whoever wins the GOP primary, in my opinion will be the person most hated by the left. One of the people things that people forget is, you know, who came in second to trump Ted Cruz and the reason why, because Ted Cruz was the second most hated guy by liberals in America a second to trump, they have nothing in policy in common. But </s> |
<s>[INST] See that's in part of what I refer to when I say love because I feel like if like political leaders and people in conflict that the small scale and large scale were able to tune into that longing to seek in each other, that basic longing for human connection. A lot of problems could be solved. But of course um is difficult because it's a it's a game of chicken. Because if you open yourself up to reveal that longing for connection with others, people can hurt you. [/INST] Well I would go a step further and I would say that taking the connection away, punishing uh penalizing people by removing the connection is a powerful tool. And that's what we see. That's why we send people to jail. That's why we put economic sanctions on countries. That's why we ground our Children and send them to their rooms. We are penalizing them whether we know it or not. We're using punitive damage by taking away that basic human connection that longing for for community. </s><s>[INST] What was your recruitment process and training process and things you could speak to uh [/INST] in </s><s>[INST] in the C. I. A. [/INST] As I was leaving the Air Force, All that was on my mind, I don't know what you were like at 27, but I was a total dipshit at 27. I'm not much better now at 42. But yeah, but I was like I just wanted to be anything other than a military officer. So I was actually in the process of applying to the Peace Corps through this thing called the internet, which was still fairly rudimentary in 2007. I had a computer lab that we went to and had 10 computers in it. You have to log in and log out and slow internet and everything else. But anyways I was filling out an online application to go work in the U. S. Peace Corps. I wanted to grow my hair out, I wanted to stop wearing shoes that were shiny. I wanted to meet a hippie chick and have hippie babies in the wild teaching Nigerian Children how to read. So that was the path I was going down. And as I filled in all of my details there came this page that popped up and it was this blinking red page and it said stop here, you may qualify for other government positions if you're willing to put your application on hold for 72 hours that gives us a chance to reach out to you. So again, 27 year old dipshit. I was like sure I'll put myself on hold if I might qualify for other government opportunities. And then about a day later I got a phone call from an almost unlisted number. It just said 703. Which was very strange to see on my flip phone at the time. Just 1703 area code. Uh And I picked it up. And and it was it was a person from northern Virginia asking me if I would be telling me that I was qualified for a position in national security And if I would be interested, they'll pay for my ticket and fly me up to langley Virginia. They didn't say C. I. A. They said Langley, I put one on one together and I </s><s>[INST] was like, maybe this is yeah, [/INST] like this, how cool is this? Or maybe this is all make believe and this is totally fake. So either way, it doesn't hurt me at all to say yes. They already have my phone number. So, yes, Yes, Yes. And then I remember thinking there's no way that happened and this isn't real. And then a day later I got a Fedex or overnight delivery of an airplane ticket and a hotel reservation and a rental car reservation. And then I just kept doing the next thing which I found out later on is a form of control. You just do the next thing that they tell you to do. And then before I knew it, I was interviewing in a nondescript building with with a person who only told me their first name for a position with the National Clandestine Service. So </s><s>[INST] you never really got a chance to think about it because there's a small steps along the way and it kind of just leads you uh and your maybe your personality is such that [/INST] that's an adventure. </s><s>[INST] It's an adventure. And you don't because it's one step at a time. You don't necessarily see the negative consequences of the adventure you don't think about. And you know that you're just stepping stepping into the adventure. [/INST] And it's easy, there's no work involved. Somebody else is doing all the work, telling me where to be. And when it's a lot like basic training in the military, anybody who's ever been through basic training will tell you they hated the first few days. And then by the end it was really comforting because you just did what you were told. They told you when to eat, they made the decision of what to eat. And then you just you marched when they told you to march, shine your shoes. When they told you to shine your shoes. Human beings love being told what to do. </s><s>[INST] What about the training process um For for for become becoming a covert C. I. A. Agent. [/INST] Yeah. So uh so the interview process is </s><s>[INST] the interview process to this. Was that how rigorous was that? [/INST] It was? It was very rigorous. That was where it became difficult. Everything up to the first interview was easy. But there's three interviews. Uh And some people are lucky enough to have four or five interviews if something goes wrong or something goes awry with the first few interviews. Uh And again this might be dated from what I went through. But uh but during the interview process is when they start they do your psychological evaluations, they do your uh they do um personality assessments, they do skills assessments, they'll start sending you back to your wherever you're living with assignments. Not not intel assignments but actual like homework assignments, write an essay about three parts of the world that you think will be most impacted in the next 3 to 5 years or you know, prioritize the top three strategic priorities for the United States and you know, put it into 250 words or 2500 words and whatever else double spaced in this font yada yada yada. Like super specific stuff. It's kind of stressful but it's just like going back to college again. So you go through all of those acts and then you submit this stuff to some P. O. Box that doesn't have anybody is ever gonna respond to you. And then you hope you just send it into the ether and you hope that you hope that you sent it right. You hope that you wrote right well enough. You hope that your assessment was right, whatever else it might be. And then eventually get another phone call that says, hey we received your package. You've been moved to the next level of interview and now we need to go to this other nondescript building in this other nondescript city. And then you start meeting, you start uh you start sitting in waiting rooms with other groups of people who are at the same phase of interview with you which were some of the coolest experiences that I remember. Still one of my best friends to this day who I don't get to talk to because he's still undercover is a guy I met during the interview process is and I was like oh we met I saw what he was wearing, he saw what I was wearing, I was </s> |
<s>[INST] typical for you to say something is easy. I got it. It's not as challenging as the other ones. Okay, well [/INST] it's okay. Maybe it's obvious how to solve it. The third one's the hardest. Where do we get a lot of people don't even think about the third one and you can see it is different from the second one. So the 2nd 1 is dynamic. The second one is like say there's a an obvious example is like a car stopped at a red light, right? You can't have that car in your map because you don't know whether that car is going to be there or not. So you have to detect that car in real time and then you have to, you know, do the appropriate action, right? Also that car is not a fixed object. That car may move and you have to predict with that car will do, right? So this is the dynamic problem. Yeah. So you have to deal with this. Um, this involves again, like you're going to need models of other people's behavior. </s><s>[INST] Do you, are you including in that I don't want to step on the third one. But are you including in that your influence and people [/INST] That's the 3rd 1. Okay. That's the best we call it. The counterfactual </s><s>[INST] Yeah. That I just talked to judea pro who's obsessed with counterfactual. [/INST] Oh yeah. Um so the static and the dynamic our our approach right now for lateral will scale completely to the static and dynamic. The counterfactual. The only way I have to do it yet. The the thing that I want to do once we have all of these cars is I want to do reinforcement learning on the world. I'm always going to turn the exploiter up to max. I'm not going to have them explore what the only real way to get at the counterfactual is to do reinforcement learning because the other agents are humans. </s><s>[INST] So that's fascinating that you break it down like that. I agree completely [/INST] my life thinking about </s><s>[INST] this beautiful. Uh, so and part of it because you're slightly insane because um [/INST] not my life just the last four years. </s><s>[INST] No, no, you have like uh some some non 0% of your brain has a madman in it, which that's a really good future but there's a safety component to it that I think when there's sort of with counterfactual and so on, that would just freak people out. How do you even start to think about just in general? I mean you've you've had some friction with nitze and so on. I am frankly exhausted by safety engineers, the the prioritization on safety over innovation to a degree where kills, in my view kills safety in the long term. So the counterfactual thing, they just just actually exploring this world of how do you interact with dynamic objects and so on. How do you think about safety? You can [/INST] do reinforcement learning without ever exploring and I said that like, so you can think about your in like reinforcement learning is usually called a temperature parameter um and your temperature parameter is how often you deviate from the arg max, I could always set that to zero and still learn and I feel that you'd always want that set to zero on your actual system. </s><s>[INST] Got you. But the problem is you first don't know very much and so you're going to make mistakes. So the learning the exploration happens through mistakes. [/INST] Yeah, but okay, so the consequences of a mistake open pilot and autopilot are making mistakes left and right. We have we have we have 700 daily active users, 1000 weekly active users. Open pilot makes 10s of thousands of mistakes a week. These mistakes have zero consequences. These mistakes are oh it I wanted to take this exit and it went straight so I'm just going to carefully touch </s><s>[INST] the wheel. Humans catch [/INST] the humans catch them and the human disengagement is labeling that reinforcement learning in a completely consequence free way. </s><s>[INST] So uh driver monitoring is the way you ensure they keep they keep paying attention. How is your messaging? Say I gave you a billion dollars? You would be scaling and now [/INST] oh if I could I couldn't scale with any amount of money was raised money. If I could if I had a way to scale you're not No, I'm not focused on, I don't know how to do like I guess I could sell it to more people but I want to make the system better, but I don't know him. </s><s>[INST] But what's the messaging here? Uh I got a chance to talk to you on and he he basically said that the human factor doesn't matter. Mhm. You know the human doesn't matter because the system will perform that there would be sort of uh sorry to use the term but like a singular like a point where it gets just much better. And so the human, it won't really matter. But it seems like that human catching the system when it gets into trouble is like the thing which will make something like reinforcement learning work. So how do you, how do you think messaging for Tesla for you should change for the industry in general should change. [/INST] I think my message is pretty clear at least like our messaging wasn't that clear in the beginning and I do kind of fault myself for that. We are proud right now to be a level two system. We are proud to be leveled to if we talk about level for it's not with the current hardware, it's not going to be just a magical OTA upgrade. It's going to be new hardware. It's going to be very carefully thought out right now. We are proud to be leveled to and we have a rigorous safety model. Um I mean not like like okay, rigorous, who knows what that means. But we at least have a safety model and we make it explicit is in safety dot MD and open pilot and it says seriously though safety dot MD. </s> |
<s>[INST] Canada left him. What is it about politicians and governments not willing to do their job? Well, not willing to do a big part of the job, which is like you send people to war, these are heroes and then you should spend most of the time repaying the debts to those, right? [/INST] Well, what is it about, why can't we? Because we're disposable numbers and we hire them out of high school when they're stupid enough to not understand what they're going to get themselves into and then we blame it on themselves </s><s>[INST] for making that decisions by volunteering. Yeah, but I mean, this still doesn't make sense to me. No, it's a comp out. I mean, Trudeau, I feel like he is a good human being that wants to do good for, I mean, I want to believe that leaders want to do good by the heroes of this world. And it doesn't, like, I don't understand the system of delusion you have to live in to not understand who the heroes are. [/INST] Like, I refuse to believe Trudeau is somehow a bad person. Is this to me? No, it's a comp out. </s><s>[INST] You haven't met him though, have you? I don't, actually, I'm speaking about Trudeau without knowing, but I mean, in general, think that way about leaders. I just think they surround themselves by people [/INST] who delude them, who like- </s><s>[INST] They're yes people. Yes, people that lead them into a kind of reality that becomes detached from actual reality. And so they misunderstand the priorities of this world. They think maybe some kind of special interest, they focus on that versus like the humans. If you look back, was there a way we could have done something better in Afghanistan, assuming we do the invasion? So is it ultimately about taking care of the veterans, like investing more money in the education of women and liberating people who are suffering injustice in those parts of the world? Like, what's the better way to do it? And one other aspect is on the US side, paid over $6 trillion for the wars in the Middle East since 9-11. So the financial side as well. [/INST] Is there something you can comment on to the things we could have done better? That's a loaded question because you're talking to someone who had no hand in what happened other than do this and do that. So I can go from my perspective, which is, there was probably plenty of things that we could have been doing better. I think there was a lack of leadership from the get-go. I think the preparation that the Canadian military gave me was nowhere sufficient for a deployment of that level. Mind you, things happen. They didn't realize things would happen, but yet they happen. With little to absolutely no cultural idea was what I was walking into. Like when the male and the family grab the back of my vest because my hair was tucked in and he thought I was a man going into a room with a bunch of women, I couldn't understand why he was attacking me. There was no real breakdown of this is what you're going into. This is the culture. This is why they do what they do. This is how they do it. This is how you should handle a situation like that. There was none of that. And something I speak about frequently and I think it's important to acknowledge is when you're doing any of that training, we are giving none of our soldiers proper mental health training. tools in that fucking toolbox or ways or things to look for on their buddy. Because we've created a system and a problem where if you say that you're ill or that you're struggling with PTSD, you're done. No one's going to say that, they're going to keep struggling with it. And that's when you get loose cannons, it's when you get problems happen. So you get fractures start to happen in leadership. And that's being seen and has been seen now for a while. So in terms of what we could have done, say for a better way to go into the country, a better way to help the country, I can't speak to that as much as I wish I could. Because I don't know that I would have all the answers. </s><s>[INST] What about withdrawal? [/INST] Oh my God. Oh my God. That's a cover. </s><s>[INST] Do you think more gradual, do you think it's better to maintain? And presence there for indefinitely? [/INST] I don't know about that, but I do know we have bases. I say weeks, I served with them. Americans have bases in Japan, Americans have bases in Korea, Americans have bases in Germany. There's reasons there's bases everywhere. There's a smart, there's an intellectual way to look at this. You want to be able to have eyes and ears. Can't have eyes and ears when you do things like you just did. The way that we pulled out of that country, the way that I hate saying American British cause it puts a blame on them, I say we because I'm a soldier, I'm a NATO soldier. The way we pulled out of that country, my five-year-old could have done it better. He could've said, Mommy, why are we not keeping that Bhagram base? Mommy, why are we not keeping that base? Just until we get everyone out that we need. Why are we going to a civilian airport that we don't control, that we don't understand? Mommy, why are we doing this when there's only one road to it? My five-year-old would have that conversation with me. It was so poorly done. It was so poorly executed and no fault of the soldiers on the ground on their own. My God, I can tell you there's operators. I just call him A. I don't say who he is because he's told me not to. There's a guy named A and there's another guy named R and there's a few other named D and these guys, they gave everything to try to pull my family when no one else would pull my family for me. They just got me on the phone because I don't know who the fuck you're talking to. I don't know how many people you're trying to get a hold of here, but you've got everyone looking for your family. It's because I've gone to everyone I know. I've done stuff on Instagram. I've got a contact. The contact called me. I called them. I was handling this family. And when they call you and say, we can't go back to the gate. My three-year-old just got beat up by the Taliban and they say, what do we do now? I'm in Vancouver. Why am I being left to deal with this? Why is the civilian and ex-military population being left to deal with this? Why was this not thought out? We knew this was coming. </s> |
<s>[INST] Yes. But the thing is just like you said, we understand so little about human intelligence, the human mind and especially consciousness that the this giant mystery this darkness that we don't understand how how how it feels like to be something to be a conscious being that because of that we're not able to really even reason about free will or not because there might be some magic that comes from consciousness. You know, the consciousness might be the thing that makes us different from a car that breaks down. There might be something uh, totally fascinating, totally undiscovered yet that will make us realize that free will is actually real and somehow fundamental to the human experience. So it's sometimes I think we forget when we talk about free will and physics and all seeming to be predetermined. We forget how little we actually understand about the world. And I think in that mystery that could be totally new ideas that are yet to be discovered and will make us realize that it's not just an illusion. There is something that is like at the core of how the universe works. Some people believe that consciousness is uh, fundamental property of the universe. Like it's it's one of the forces of physics. Like consciousness permeates everything. It's in everything like this table is conscious, but it's not as conscious as us and where this little peak of consciousness and if that's true, and if we get to understand that maybe there's something there's an extra bonus we get in terms of free will once you become one of those entities that are super conscious. So I tend to be sort of humbled by the mystery of it. [/INST] Do you believe one day with, with the technology that keep improving, we will make robot that will be able to be somehow </s><s>[INST] conscious. Absolutely. I that's been my dream. That's been I hope I hope to do just that. First of all, I believe that all people are capable and want to be good to each other. And I think love is a is a really powerful thing that connects us and can create better and better worlds, sort of like uh create better and better societies that uh improve both the technology, the quality of life and just the basics of human experience. And I think creating ai systems that are conscious that our human like can enable us to be better to each other. Like they can, it's almost like adding more and more kindness to the world through the systems we interact with will inspire us to be better and better to each other in terms of them being conscious. I think that is uh an absolute requirement that entities will interact with, communicate some element of consciousness to us. Like that's how we connect to each other. The reason we, you and I connect is that we believe that each of us are conscious and to me what consciousness means is the ability to hurt ability to to suffer to to struggle in this world. Because just like you said, uh with with without the struggle, you don't have the love. You don't have the pleasure. And ultimately consciousness is an entity's ability to struggle to suffer and from that arises the pleasure and us together. Being able to appreciate sort of uh appreciate the highs and experience together the lows? That's how we form the deep connections. I personally think we can create that in robots. And I personally believe it's a lot easier than uh than we think. [/INST] Does it make you afraid sometimes about the fact that one day, hey hi like intelligence, artificial intelligence could be, you know, could hurt us or you know like because we're, you know, because of Hollywood of course, you know the movies we watch. But it seems like when I hear sometimes Elon musk talking, you know like </s><s>[INST] yeah, so Ellen talks about with ai we're summoning the demon. He is very concerned and I talked to him about it quite a bit. He's very concerned about all the different ways Ai could hurt us humans. I tend to believe that there's a lot more ways in which a I can make our lives better and can make life awesome for humans. I think humans are the ones that can do a lot of evil things. So I'm less worried about a I am more worried about humans. [/INST] If I look at what humans have done on on the course of history, you know, for for example, in regards the planet to the scale of the universe, I think what I'm afraid is that we have more of a destructive force than a beneficial forces. So if a high take that in consideration in order to protect us against herself, he could hurt us. And I don't know if you if you understand what, like what do you think about that? The does it makes you afraid? Sometimes not because of a high, but because of what humans are doing that hey, I could do to us to prevent us of hurting each ourself, </s> |
<s>[INST] way to get it to do what we're interested in doing. There does seem to be a difference between a horse and a large language model or something that could be called AGI connected to the internet. So let me just ask you about big philosophical question about the threats of these things. There's a lot of people like Eliezer Adkowski who worry about the existential risks of AI systems. Is that something that you worry about? You know, sometimes when you're building [/INST] an incredible system like Wolfram Alpha, you can kind of get lost in it. </s><s>[INST] Oh, I try and think a little bit about the implications of what one's doing. You know, it's like the Manhattan Project kind of situation where you're like, it's some of the most incredible physics and engineering being done. But it's like, huh, [/INST] where's this going to go? I think some of these arguments about kind of, you know, they'll always be a smarter AI, they'll always be, you know, and eventually the AIs will get smarter than us. And then all sorts of terrible things will happen. To me, some of those arguments remind me of kind of the ontological arguments for the distance of God and things like this. They're kind of arguments that are based on some particular model, fairly simple, model, often of kind of, there is always a greater this, that and the other. And what tends to happen in the sort of reality of how these things develop is that it's more complicated than you expect that the kind of simple logical argument that says, oh, eventually there'll be a super intelligence, and then it will do this and that turns out not to really be the story. It turns out to be a more complicated story. So, for example, here's an example of an issue. Is there an apex intelligence? Just like there might be an apex predator in some ecosystem, is there going to be an apex intelligence? The most intelligent thing that there could possibly be? Right? I think the answer is no. In fact, we already know this and it's a kind of a back to the whole computational irreducibility story, there's kind of a question of, even if you have sort of a Turing machine and you have a Turing machine that runs as long as possible before it halts. You say, is this the machine, is this the Apex machine that does that? There will always be a machine that can go longer, and as you go out to the infinite collection of possible Turing machines, you'll never have reached the end, so to speak. never, it's kind of like the same question of whether there'll always be another invention. Will you always be able to invent another thing? The answer is, yes, there's an infinite </s><s>[INST] tower of possible inventions. That's one definition of apex, which I also think might be true. Is there a species that's the apex intelligence right now on earth? So it's not trivial to say that humans are that. [/INST] Yeah, it's not trivial. I agree. It's a, you know, I think one of the things that I've long been curious about kind of other intelligences, so to speak, I mean, I, you know, I view intelligence is like computation and it's kind of a, you know, you're sort of, you have the set of rules, you deduce what happens. I have tended to think now that there's this sort of specialization of computation that is sort of a consciousness-like thing that has to do with these, you know, computational boundedness, single thread of experience, these kinds of things that are the specialization of computation that corresponds to a somewhat human like experience of the world. Now the question is, so, so that's, you know, there may be other intelligences, like, you know, the, you know, the aphorism, you know, the weather has a mind of its own. It's a different kind of intelligence that can compute all kinds of things that are hard for us to compute, but it is not well-aligned with us, with the way that we think about things. It doesn't think the way we think about things. In this idea of different intelligences, every different mind, every different human mind is a different intelligence that thinks about things in different ways. In terms of the formalism of our physics project, we talk about this idea of a ruled space, The space of all possible sort of rule systems and different minds are in a sense that different points in real space, human minds, ones that have grown up with the same kind of culture and ideas and things like this might be pretty close. In real space, pretty easy for them to communicate, pretty easy to translate, pretty easy to move from one place in real space that corresponds to one mind to another place in real space that corresponds to another sort of nearby mind. When we deal with more distant things in ruellial space like the pet cat or something, the pet cat has some aspects that are shared with us. The emotional responses of the cat are somewhat similar to ours, but the cat is further away in ruellial space than people are. So then the question is, can we identify, can we make a translation from our thought processes to the thought processes of a cat or something like this? What will happen when we get there? I think it's the case that many animals, dogs for example, they have elaborate factory systems, they have the smell architecture of the world, so to speak, in a way that we don't. If you were sort of talking to the dog and you could communicate in a language, the dog will say, well, this is a flowing, smelling, this, that, and the other thing, concepts that we just don't have any idea about. Now what's interesting about that is one day we will have chemical sensors that do a really pretty good job. We'll have artificial noses that work pretty well, and we might have our augmented reality systems show us the same map that the dog could see and things like this similar to what happens in the dog's brain. Eventually, we will have kind of expanded in ruellial space to the point where we will have those same sensory experiences that dogs have. We will have internalized what it means to have the smell landscape or whatever. And so then we will have kind of colonized that part of ruellial space until we haven't aren't gone. Some things that animals and so on do, we successfully understand, others we do not. And the question of what representation, how do we convert things that animals think about to things that we can think about, that's not a trivial thing. And I've long been curious. I had a very bizarre project at one point of trying to make an iPad game that a cat could win against its owner. I see it, Dave. It feels like there's a deep philosophical goal there, though. Yes. Yes. I was curious if pets can work in Minecraft or something and can construct things. What will they construct? And will what they construct be something where we look at it and we say, oh yeah, I recognize that. Or will it be something that looks to us like something that's out there in the computational universe that one of my cellular automata might have produced? Where we say, oh yeah, I can kind of see it operates according to </s> |
<s>[INST] in tournament [/INST] tournaments. Yeah, he was crushing tournaments, he was going against the norm in terms of what is like, you know, this is what you should do as a poker player in the spot. He wasn't doing that. He was doing what he thought was best and he was doing things outside the norm that again in a vacuum, you could look at that and you go that's incorrect that he should not do that is a clear cut mistake even, you know, the solvers or the computers or game theory would say this is wrong, what he's doing, But it's not wrong if he's doing it in a way that he's exploiting other players tendencies. So for example, with him say he's playing far too aggressively, Okay, that's not good unless your opponents are playing way to passively. So if your opponents are playing passively, the answer is to be more aggressive with them. And that's I think one of the biggest advantages he had was he was willing to do that. </s><s>[INST] So bet big big pots bluffing. [/INST] So in a spot where somebody would make it 1000, he's making it 22,000. What what is this? This makes no sense. </s><s>[INST] And then people kind of know he has nothing, but they're too afraid to uh call him on it [/INST] Well. And then sometimes what happens is this is where the leveling comes in. You're like, man, this guy's crazy, he's bluffing like nuts. Then he bets the 22,000 and you say I'm taking my stand, I call and then he shows you like, you know, four of a kind or something like that. So he gets people out of their comfort zone and I really enjoy watching him play. He's probably my favorite player to watch today. </s><s>[INST] Watching a guy like that? What aspect of his play have you been able to incorporate into your own? Like what do you learn from that? Because you're constantly learning constantly adjusting? [/INST] Well, no, and I love it. And as I said, so I think a lot of players sort of come to the same conclusions about this is how you play the spot, but he doesn't. And I love watching and thinking in terms of like why he's doing this. And one specific thing, for example, is He's willing to really go for it. So in a spot where let's say he bets 2000, he knows he'll get you call 2000, right? But he wants it all, he wants it all. So he says, you know what, I'll give up the 2000 that's guaranteed and I'll bet 50 </s><s>[INST] 1000 [/INST] Maybe if you call that now, you know, so listen, you lose the 2007 08 times, but if I get called for the 50 just once, you know, I'm profiting from that. And it also sets the uh, you know, the template for you to really sort of be a player that people are afraid to play against. He he knocked me out in a tournament very early on in a huge event and he had he was so far ahead, he was one step ahead of my thought process in hand and he did something that makes no sense whatsoever. I looked it up on the computer, huge mistake if you will, but not a mistake because he was taking advantage of my tendency, </s><s>[INST] remember [/INST] remember the whole thing I remember like yesterday, </s><s>[INST] can you take it? Like for example hand that really demonstrates it. [/INST] So I'll explain the hand here. So I'm on the button and I have a sneaking very good hand and I raise and he calls from the big blind. The flop is 975. So I have nothing really here. He checks. I check behind the turn card is an ace. He checks. I bet half the pot. There were 6000 in there. I bet 3000. Okay, Now this is not a typical thing you see people do, but he raised me to 36,000 massive raise bigger than the size of the pot </s><s>[INST] nine [/INST] 75 turning. </s><s>[INST] Well [/INST] He could have a straight, you could have 33 of a kind, he could have, you know, aces up, you could have a whole bunch of hands. So he check raises me big to 36,000. I call the bet. So now there's something like 75,000. The river is a five. So the board pairs okay. He thinks for a while and he bets all of it, which is three times the pot. He bets 225,000. There's only 75,000, right? And in theory he should never ever have a hand that can do that, right? So he confused me. And I was like, okay, well this guy's aggressive, he likes to blow up and all this kind of stuff. So I made the call with the ace king and he turned over 68. So we had a straight. But here's the thing in theory that river card is bad for him when I called the turn. I have a lot of the time three of a kind two pair that just made a full house. So he was risking that. And the reason he did it was because he thought I would perceive him to be bluffing a lot. So he just went for it and it worked. He was able to double up right away and knocked me out of the tournament like an hour in I think specific. I think it was, it came down to this, it's as simple as this. He was cognizant of his image as being a wild aggressive lover, right? And he was fully taking advantage of me knowing that my tendency in these spots is to be curious and I want to call and I want to see it. So he was fully taking advantage of the fact that he thought I would call too often because otherwise his play makes no sense. A small bet. A medium sized bet. Those makes sense. But the bet that he made, in theory is indefensible. It's just like clearly a mistake. But that's why poker is so fascinating because he makes this planet wasn't a mistake, It was above the rim. That's what it </s> |
<s>[INST] kind of a difficult topic and let's see if we can talk about it, figure it out. But so with the help of a I more and more we can create versions of ourselves versions of reality that are in some ways more beautiful than actual reality. Uh you know, and uh some of the creative effort there is as part of doing this, creating this illusion. So of course this is inevitable. But how do you think we should adjust this human beings to live in this digital world? That's partly artificial. That's better then? Uh the world that we lived in 100 years ago when you didn't have instagram and facebook versions of ourselves. And the unlike [/INST] this is sort of showing off better versions of ourselves. </s><s>[INST] We're using the tooling of modifying the images or even with artificial intelligence ideas of deep fakes and creating adjusted or fake versions of ourselves in reality. [/INST] I I think it's an interesting question, well sort of historical bent on this. So I actually wonder if 18th century aristocrats who commissioned famous painters to paint portraits of them had portraits that were slightly nicer than they actually looked in fact well played. So human desire to put your best foot forward has always been true. I think it's interesting you you sort of framed it in two ways. One is if we can imagine alternate realities and visualize them. Is that a good or bad thing in the old days, you do it with storytelling and words and poetry, which still resides sometimes on websites. But uh, you know, we've become a very visual culture in particular In the 19th century were very much a text-based culture. People would read long tracks. Political speeches were very long. Nowadays, everything is very kind of quick and visual and snappy. I think it depends on how harmless your intent. It's a lot of it's about intent. So if you have a somewhat flattering photo that you pick out of the photos that you have in your inbox to say this is what I look like. It's, it's probably fine. If someone's going to judge you by how you look, then they'll decide soon enough when they meet you whether the reality, you know, um, I think where it can be harmful as if people hold themselves up to an impossible standard, which they then feel bad about themselves for not meeting. I think that's definitely can be an issue. And uh, but I think the ability to imagine and visualize an alternate reality which sometimes sometimes of which he then go off and built later, um, can be a wonderful thing. Two people can imagine architectural styles, which they then, you know, have a startup, make a fortune and then build a house that looks like their favorite video game. Is that a terrible thing. I think uh I used to worry about exploration actually that part of the joy of going to the moon when I was a tiny child. I remember it and grainy black and white was to know what it would look like when you got there. And I think now we have such good graphics for knowing for visualizing the experience before it happens, that I slightly worried that it may take the edge off actually wanting to go, you know what I mean? Because we've seen it on tv, we kind of, you know, by the time we finally get to mars we go, yeah, it's mars if that's what it looks like, but then, you know, the outer exploration. I mean, I think Pluto was a fantastic recent discovery where nobody had any idea what it looked like and it was just breathtaking in a varied and beautiful. So I think expanding the ability of the human toolkit to imagine and communicate on balance is a good thing. I think there are abuses. We definitely take them seriously and try to discourage them. Um, I think there's a parallel side where the public needs to know what's possible through events like this, right? So that you don't believe everything you read in print anymore and it may over time become true of images as well or you need multiple sets of evidence to really believe something rather than a single media asset. So I think it's a constantly evolving thing that's been true forever. There's a famous story about anne of Cleves and Henry the Eighth where luckily for an they didn't get married right. Um So all they got married and what's the story? So whole bind went and painted a picture and then Henry the eighth wasn't pleased and you know, history doesn't record whether and was pleased, but I think she was pleased not to be married more than a day or something. So, I mean this has gone on for a long time, but I think it's just part of the magnification of human capability. </s><s>[INST] You've kind of built up an amazing research environment here. Research culture research lab and you've written at The Secret to a thriving research labs interns. Can you unpack that a [/INST] little bit? Absolutely. So a couple of reasons, as you see looking at my personal history, there are certain ideas you bond with at a certain stage of your career and you tend to keep revisiting them through time. Um if you're lucky, you pick one that doesn't just get solved in the next five years and then you're sort of out of luck. So I think a constant influx of new people brings new ideas with it from the point of view of industrial research, because a big part of what we do is really taking those ideas to the point where they can ship us very robust features. Mhm. You end up investing a lot in a particular idea and if you're not careful, people can get too conservative in what they choose to do next, knowing that the product teams will want it. And in turns let you explore the more fanciful or unproven ideas in a relatively lightweight way, ideally leading to new publications for the internet and for the researcher. And it gives you then a portfolio from which to draw, which idea am I going to then try to take all the way through to being robust in the next year or two to ship. So it sort of becomes part of the funnel. It's also a great way for us to identify future full time researchers. Many of our greatest researchers were former interns. It builds a bridge to university department so we can get to know and build an enduring relationship with the professors and we often do academic give funds to as well as an acknowledgement of the value of the interns ad and their own collaborations. So it's sort of a virtuous cycle and then the long term legacy of a great research lab hopefully will be not only the people who stay, but the ones who move through and then go off and carry that same model to other companies. And so we believe strongly in industrial research and how it can complement academia and we hope that this model will continue to propagate and be invested in by other companies, which makes it harder for us to recruit of course, but you know, that's a sign of success and a rising tide lifts all ships in that sense. </s> |
<s>[INST] So even though the origins are humble just like it is for a genetic code for a i even though the origins are humble the the uh the result of it being deployed into the world is infinitely complex and that's and yet there's not it's not yet able to be cognizant of all the other layers in uh of it's, you know, it's not uh it's not able to think about space and time. It's not able to think about the hardware on which it runs, the electricity on which it runs yet. [/INST] So, so if you look at humans, we basically have the same cognitive architecture as monkeys as the great apes. It's just a ton more of it. If you look at GPT three vs. GPT two, again, it's the same architecture, just more of it and yet it's able to do so much more. So if you start thinking about what's the future of that GPT four and 55 do you really need fundamentally different architecture? Or do you just need a ton more hardware? And we do have a ton more hardware. Like these systems are nowhere near what humans have between our ears. So, you know, there's something to be said about stay tuned for emergent behavior. We keep thinking that general intelligence might just be forever away. But it could simply be that we just need a ton more hardware and that humans are just not that different from the great apes, except for just a ton more of it. </s><s>[INST] It's interesting that in the Ai community, maybe it is a human centric fear. But the notion that gpt 10 will be, we'll achieve general intelligence is something that people shy away from, that. There has to be something totally different and new added to this. And yet it's not seriously considered that um this very simple thing, this very simple architecture when scaled might be the thing that achieves super intelligence [/INST] and people think the same way about humanity and human consciousness, they're like, oh consciousness might be quantum or it might be, you know, some some non physical thing and it's like or it could just be a lot more of the same hardware that now is sufficiently capable of self awareness just because it has insurance to do it. So maybe the consciousness that is so elusive is an emergent behavior of you basically string together all these cognitive capabilities that come from running from seeing for reacting from predicting the movement of the fly as you're catching it through the air, all of these things are just like great look up tables and quoted in a giant neural network, I'm over simplifying of course the complexity and diversity of the different types of excitatory inhibitory in europe. The wave forms that shine through the, you know, the connections across all these different layers. The amalgamation of signals to the brain is enormously complex, I mean of course, but again, it's a small number of primitives encoded by a tiny number of genes which are self organized and shaped by their environment, babies that are growing up today are listening to language from conception basically. As soon as the auditory apparatus forms, it's already getting shaped to the types of signals that are out in the real world today. So it's not just like, oh have an Egyptian be born and then ship them over. It's like no that that Egyptian would be listening in to the complex of the world and then getting born and sort of seeing just how much more complex the world is. So it's a combination of the underlying hardware which if you think about as a geneticist in my view, the hardware gives you an upper bound of cognitive capabilities. But it's the environment that makes those capabilities shine and reach their maximum. So we are a combination of nature and nurture. The nature is our genes and our cognitive apparatus and the nurture is the richness of the environment that makes that cognitive apparatus reach its potential. And we are so far from reaching their full potential so far. I think that kids being born 100 years from now, they'll be looking at us now and saying what primitive educational systems they had. I can't believe people, we're not wired into this virtual reality from birth as we are now because like they are clearly inferior And so and so forth, basically think that our environment will continue exploding and our cognitive capabilities. It's not like we're only using 10% of the brain. That's ridiculous. Of course we're using 100% of the brain but it's still constrained by how complex our environment is. </s><s>[INST] So the hardware will remain the same. But the software in a quickly advancing environment. The software will make a huge difference in the nature of like the human experience. The human condition is fascinating to think that humans will look very different 100 years from now just because the environment changed, even though we're still the same great apes, the descendant of apes at the core of this is kind of a notion of ideas that I don't know if you're there's a lot of people that's including you eloquently about this topic, but Richard Dawkins talks about the notion of memes and let's say this notion of ideas, you know, multiplying selecting in the minds of humans. Do you ever think from about ideas from the from that perspective ideas as organisms themselves, They are breeding in the minds of humans. [/INST] I love the concept of needs. I love the concept of this horizontal transfer of ideas and sort of permeating through through our layer of interconnected neural networks. So you can think of sort of the cognitive space that has now connected all of humanity where we are now one giant information and idea sharing network well beyond what was thought to be ever capable. When the concept of a meme was created by Richard Dawkins. So, but I want to take that concept just a you know, into another twist which is the horizontal transfer of humans with fellowships and the fact that as people apply to MIT from around the world. There's a selection that happens not just for their ideas but also for the cognitive hardware that came up with those ideas. So we don't just ship ideas around anymore. They don't evolve in a vacuum. The ideas themselves influence the distribution of cognitive systems are humans and brains around the planet. </s> |
<s>[INST] You've talked about this a little bit in the past, you kind of left the movie making business at an interesting time. Perhaps you saw the changes. Um, there's been a lot of excitement with netflix with tv. So the roll of film in society has changed. So what do you think is the future of movies versus tv? Like if you were as a business person, as a creator, as a consumer, as a technologist are thinking about the next 10, 20 years, what do you think is going to be the the godfather? The great Pieces that move us as a society in the next 10, 20 years is going to be TV? Is it going to be a movie? Is it going to be a tiktok uh clips? What is it? [/INST] Well, so I, and I, and I think the other category that I would add to that, that will be the next great medium is truly immersive, virtual reality, uh in which new storytellers will emerge, especially when you can go into VR and there's enough computing power to sustain it and to allow it to be social and, you know, for you to have different paths to go down. That'll be, I think the next realm of what storytelling and experience will will look like, </s><s>[INST] Do you think video game kind of world, or is it more movies, or is it more social network, or is it all of it kind of blending reality and gaming and movies? [/INST] Yeah, I thought uh if you saw Ready Player One, which I loved the book and and and the movie was cool too. But you know, that that's one version of it, right? Where you go in now, everybody's talking about the metaverse and all that, but you go into a world that's fully rendered as yourself and you interact with that world. The other side of it is to go in somewhere between being a passive observer, but being able to move around your point of view and experiences, which I think is interesting. And then I think another, another adventure, so to speak, I could think of is a blend of video games, so there's a mission, right? There's a there's obstacles, there's everything and you move through it, but it's immersive and it tells a story at the same time and that's why I think you're going to see new, amazing storytellers that we don't know yet that understand how to innovate and how to make you feel something in, in that environment. Um, And to your earlier point, you know, I saw probably around 2,015 when netflix decided to be bold, put out house of cards, put out all the episodes, leave you in charge of the pace at which you would view them uh which I thought was was great. </s><s>[INST] That's a gutsy move. [/INST] Yes, it was. And I can't tell you around Hollywood, anybody that says that everybody thought it was a great ideas not being truthful because everybody I talked to said this is they're idiots right there. What do they know about moviemaking and tv and what I saw happening was if you look at what netflix pulled off and they realized that there isn't really a moat around the studios, you really could make stuff. Um, and really good stuff. And so they started to create their own content that pulled in amazon, which pulled in google through youtube and, and then you had hulu, then you had Disney deciding that they're going to have Disney Plus and the next thing, you know, you have some of the, the biggest companies with the largest balance sheets on the planet being in the creative business, that's, you know, if you're an independent, that's, that's bringing a knife to a gunfight to be sure. And so, you know, I thought that was interesting. The other thing that it used to be, that movies were, where the big things happened and television was sort of, it is a small screen different experience. And you had something like game of Thrones come out which was not only in the same epic level visually and storytelling wise, but had the budget to be able to do it. And now I think you're you're seeing um you're seeing uh all kinds of different storytelling taking place and also like that, it's you're not pigeonholed into a time, like you've got two hours to tell the story, you can do a three part mini series, a five part mini series, you can do television, that's you know, all kinds of different format that I think is, it allows creators to do a lot more interesting things. </s> |
<s>[INST] Yeah, but the cynical take is that on the surface they're changing their opinion, but that there's a boys club. boys means the elite, that under in the smoke-filled rooms in secret, they actually have their own agenda and they're following that agenda and they're just saying anything publicly to placate the public based on whatever the new trends are. So here's- The Seneca take a pull-up. [/INST] Yeah, I understand. So here's- The Seneca take a pull-up. Yeah, I understand. Somebody asked me this question and it flipped, I 180'd completely. I was a Bernie Sanders supporter in 2016 and my single issue voting thing was lobbying. I thought that lobbying, the government was corrupt, they weren't responding to the people. It was completely destroyed my faith in government everything. And I had one question posed to me by a conservative that used to come on my stream and show me. And he said, and then he asked me, can you think of any popular opinion that the American public has that the government is unresponsive to? Is there some big piece of legislation or policy or whatever that people want that the government isn't doing? When they asked me that, I couldn't think of a single good answer. And I'm like, oh, geez. There's good hamps. There's drugs. There's not. There's not legal cheese. </s><s>[INST] There's good hamps. There's a draw-ization. Legalization of drug? Hold on a second. Yeah, go for it. [/INST] Oh, shoot. You're doing the Joe Rogan thing. You're pushing back because I brought up weed. </s><s>[INST] Go ahead. I'm sorry. I have become meme. I don't even... I don't want to interrupt your book. No, you're good. Because there's memes upon memes upon memes. I can go with here. But, no. I don't feel bad because people bring up... OK, there's no issues. [/INST] There's no issues that the government is not representing of the public. No, you're good. So here's the issue. So somebody will bring up like, well, what about the legalization of drugs? OK, the first issue people have is, one, they look at national polling. Yeah, very few things are decided on a national level. So that's the first huge mistake. Arguably, a lot of BLM made mistakes in this arena where they're saying like, why isn't the government doing anything about policing? Federal government can't do anything about policing. That's going to be your... of your state government. It's not like your local city government. The people that elect like your chief of police, your police commissioner, that's coming from your mayor. So you've got people looking, one, at the wrong parts of the government to even figure out the solution to the problem. Two, oftentimes for polling, the questions are vague enough that you can poll very high, but when you get into the weeds on things, no pun intended, you start to realize like, oh shoot, this is more complicated than I thought. I don't know the numbers in particular for legalization of marijuana, but this is what I'm gonna guess is the case. If you poll and you say, should we legalize marijuana, that number might poll at 65, 70%. But that's including people that are in favor of medical marijuana. If you were to poll, should we legalize, should we decriminalize recreational use of marijuana? That number might drop to 52%. And then, if you poll, should we completely legalize? Not just decriminalize, but completely legalize recreational use of marijuana? That number might drop to 40%. There's all these different ways you can poll around issues where people are like, oh, no, we broadly agree on this topic, but when you really figure out well, Do you, do we really agree? Or is there just like broad consensus around a thing that's never gonna show up like in a piece of legislation? A really good example, one example I do know is socialized healthcare. I think if you poll, there was a time a few years ago where if you poll America, do you think every American citizen should have access to like free healthcare? I think that answer, that poll like 74% yes. But when you asked, should the government be the sole provider of healthcare, it dropped to like 26%, dropped 50 points. And you could see it was both asking questions about single payer, but the way that it was asked was so different that even if you all, it looks like there's consensus, there's not nearly as much consensus as people think around certain ideas. </s><s>[INST] Yeah, go, we aren't getting any flip-flops. You're right, you're right, you're right. That polls, the way you ask the polls really matters. But when you ask, should the government be in charge of a thing, that also biases the answer, right, like because there's such a negative experience with government creating a.gov site that runs the thing. But sometimes. Sometimes, I think if you dig in, if you have a one-hour conversation with each individual citizen. Sure. I think you will understand that yes, there is support for socialized medicine. Like it's not. The argument has to be made though, yeah. [/INST] What do you mean, the argument has to be made. Like if you just ask a conservative, like what about a single payer, they're gonna tell you no. You might be able to build up to an argument for it, but you're gonna have to make the case for it. </s><s>[INST] No, but I thought we were talking about the feeling deep inside your mind and heart. Does the government represent that? Oh. So it's not like some shallow surface layer public opinion. Does the government effectively represent what the people want? Not a shallow survey, but deeply what they want. I'm not actually that familiar with the debates over healthcare, but let's maybe look at an easier one. [/INST] Maybe you'll say it's harder, but war. War is a really good example where the government was very responsive, </s> |
<s>[INST] Well, certainly. Right. But but what what is it? Is it definite ble So it's getting to the point that you're [/INST] there is no function. It's the function function. It's that we trade it for material goods and the And we have to agree. Right. We all impose on whatever it is. Salt barley, little shelves, big rocks in the ocean that can't move Bitcoin pieces of plastic mortgages which are basically a promise of something in the future, nothing more. Right? All of these things, we impose value on them and we all agree that we can exchange them for material goods. </s><s>[INST] Yeah. And uh that's by the way, you're attributing some of that to Darwin [/INST] that he thought, no I'm no I'm saying </s><s>[INST] that because it's a brilliant view of what a species is the function. [/INST] You know what I'm saying? Is that what Darwin, Darwin really talked about variation in. Um so if you read for example the biologist ernst Mayr who was an evolutionary biologist and then when he retired became a historian and philosopher of biology and his suggestion is that Darwin, Darwin did talk about variation. He vanquished what's called essential Ism the idea that there's a single set of features that define any species. And um out of that grew um really discussions of the functions, you know, like some of the functional features that species have, like they can reproduce off so they can have offspring. The individuals of a species can have offspring. It turns out that's not a perfect uh you know, that's not a perfect criterion to use, but it's a functional criterion, right? So what I'm saying is that in cognitive science people came up with the idea they discovered the idea of conceptual categories or ad hoc concepts, these concepts that can change based on the function they're serving, right? And uh that it's their Darwin, it's in Darwin and it's also in the philosophy of social reality. You can, the way that philosophers talk about social reality just look around you. I mean, we impose we're treating a bunch of things as similar, which are physically different. And sometimes we take things that are physically the same and we treat them as separate categories. </s><s>[INST] But it feels like the number of variables involved in that kind of categorization is nearly infinite. And that's [/INST] no, I don't think so, because there is a physical constraint, right? Like you and I could agree that we can fly in real life, but we can't that's a physical that's a physical constraint that we can't break you. And I could agree that we could walk through the walls, but we can't we could agree that we could eat glass, </s><s>[INST] but there's a lot of [/INST] constraints. We could agree that the virus doesn't exist and we don't have to wear masks. Yeah, But, you know, physical reality still holds the trump card, right? But still there's a lot of uh upon completely unintended, but there you go. That's a predicting brain for you. Uh but but there's a tremendous amount of leeway. Yeah, that's the point. So what I'm saying is that emotions are like money, </s><s>[INST] basically, [/INST] they're they're like money, they're like, countries they're like um kings and queens and presidents they're like everything that we construct, that we impose meaning on. We take these physical signals and we give them meanings that they don't otherwise have by their physical nature and because we agree they have that function. But </s><s>[INST] the beautiful thing maybe unlike money, I love this similarity is it's not obvious to me that this kind of emergent agreement should happen with emotion because their experiences are so different for each of us humans. And yet we kind of converge [/INST] well, in a culture we converge, but not across cultures, there are huge, huge differences. There are huge differences in what, what concepts exist, what their, what they look like. Um, so what I would say is that, I feel like what, what we're doing with our young Children as we, as their brains become wired to their physical and their social environment, right? Is that we are curating for them. We are bootstrapping into their brains a set of emotion, uh concepts that's partly what they're learning. And we curate those for infants just the way we curate for them. What is a dog? What is a cat? What is a truck? We sometimes explicitly label and we sometimes just use mental words, um, when, you know, your kid is, you know, throwing cheerios on the floor instead of eating them or your kid is crying when, you know, she won't put herself to sleep or whatever. You know, we use mental words and um, a word is this words with for infants words are these really special things that they help infants learn abstract categories. There's a huge literature showing that Children can take things that don't look infants like infants. Really young infants pre verbal infants can take if you label, if I say to you and you're an infant, okay, so I say lex lexi, this is a bling and I put it down and the bling makes a squeaky noise and then I say let's </s><s>[INST] see [/INST] this is a bling and I put it down and it makes a squeaky noise and then I say lexy this is a bling, you As young as four months old will expect this to make a noise a squeaking noise. And if you don't if it doesn't you'll be surprised because it violated your expectation, right? I'm building for you an internal model of a bling. Okay, infants can do this really, really at a young age and so there's no reason to believe that they couldn't learn emotion categories and concepts in the same way and in in one and what happens when you go to a new culture, when you go to a new culture, you have to do what's called emotion acculturation. So my colleague Mesquita in Belgium studies emotion acculturation. She studies how, when people move from one culture to another? How do they learn the emotion concepts of that culture. How do they learn to make sense of their own internal sensations and also the movements, You know the raise of an eyebrow, the tilt of a head. How do they learn to make sense of cues from other people using concepts they don't have but have to make on the fly. </s> |
<s>[INST] there a good definition of good guys in the sense the ones who were white, there's layers to this. You're like modern day Shakespeare, Is there a danger in thinking um uh Adolf, Hitler was probably pretty confident that he led a group of good guys. [/INST] Listen, if Hitler did anything wrong wise and in jail, Czech friend thought </s><s>[INST] of that joke [/INST] and he actually, he says that his accent, he goes, Hitler's so bad Wise and he in the jail, </s><s>[INST] that's a good point, he's probably still alive, Right? [/INST] And look yeah, hopefully, </s><s>[INST] Oh boy, two of the three people listening to this are very upset right now? Uh what were you even talking about? How do you, how do you know the what is good? [/INST] there's lots of standards of good, but if you're, for me to be a good guy is if you want to leave the world a little bit better than you found it. That to me is the definition of a good guy. And I think there are many people that that that's not their motivation. </s><s>[INST] And also it's about your motivation. [/INST] Well, it's also about if your motivation is at all um correlated to reality, no one thinks we're the bad guys. That that's correct. But are you taking steps to check your motivations? And and also takes a certain amount of humility? Because if you're going to start interfering with people's lives, you really better be sure. You know what you're talking about </s><s>[INST] the control of others. If you do have centralized control and you kind of, you become a leader of a group, you better know you better do so humbly and cautiously. And [/INST] I also have uh steam valves, right? So if in case things go wrong, let's have, I'm sure this is a lot happening with ai would ever work with computers like, okay, if something goes wrong here, how do we do have a work around to make sure it doesn't cause everything to </s><s>[INST] collapse. Yeah, the going wrong thing. I mean the whole the feedback mechanism like uh I wonder if people in congress, I think that things are really wrong. [/INST] It's working for them. </s><s>[INST] Are you? Are you sure? Because I'm not sure because I I'd like to believe that the people that at least when they got into politics actually wanted some of it is ego, but some of it is like wanting to be the kind of person that builds a better world. [/INST] Sure. I also think it's a it's diverse something. We're going to have different motivations than others, </s><s>[INST] but like once you're in the system and trying to build a better world, how do you know that it's not working? Like how do you take the basic feedback mechanisms and like and actually productively change? I mean that's what it means to be a good guys. Like something is wrong here. And that's that's why I like the law mosque. I think from first principles like way Okay, let's ask the big question. Like can this be one, is this working at all? Like the way we're solving this particular problem of government, Is this working at all? And then like stepping away and saying like as opposed to modifying this bill or that bill or like this little strategy like increased attacks by this much or decrease attacks by this much? Like why do we have a democracy at all or why do we have any kind of representative democracy? Shouldn't it be a pure democracy or why do we have states? Uh like representation states and federal government so on. Why do we have this kind of separation of powers? Is this different? Why don't why don't we have term limits or not? Like big things like how do you actually make that happen? And is that what it means to be a good guy? It's like taking big revolutionary steps as opposed to incremental steps? Well, [/INST] I, I don't know that you can be a politician to be a good guy, to be analyst and let me give you a counter example of someone who you can tell it's not being a good guy. Uh Joe biden said he was, he regards the Iraq war is a mistake. Okay, You and I have made mistakes in our lives. I'm sure none of our mistakes have caused tens of thousands of people to die. Um if let's suppose something for yourself, if it's fair, okay, I'll take that. I don't build the kill bots. Um if I were a chef, let's take it out of politics and in my restaurant somehow accidentally someone ate something, they died a I would feel horrible. But more importantly, I would be like, we need to look through the system and figure out how it got to the point where someone lost their life because that can never happen again and we need to figure out step by step, it's, there's, I'm not a gun person, but there's like this checklist of like if you're holding a gun, there's five things to do and even if you get too wrong, you're gonna be, it's like assume every gun is loaded, only pointed at something that you want to kill and there's like three other things and it's like to make sure that nothing goes wrong. So if I made, if I'm not chef and I would have to not only feel guilt but take preventative action to make sure this has no possibility of happening again. If you look at the staff, he's putting in, it's the same warmongers that would have advised him to get into the Iraq war on the first time. That is, to me is not a good guy. That to me is someone who does not feel remorse for their responsibility in killing. Not only many americans, but some of us think that you know dead Iraqis isn't necessarily ideal either. </s> |
<s>[INST] So what about the would be interesting to get your perspective, because I know Travis perspective is the just the number of injuries like what do you make of the perseverance through all the injuries he had to overcome specifically. Like you just observing this creature that you've coached. I mean, he seems to not see the injuries as, as a problem. He just, like, just like you said, head through the wall, it's like, what, like when I, when we were talking about injuries, he kind of, he doesn't even see the injuries themselves is the problem because he thinks that the injuries, you know, you heal back stronger, I forget the exact quote, but he said, like, my body is is now less injury prone than most of anyone else, because I've already broken, I've broken everything and it's just grown back stronger. [/INST] Because I asked him </s><s>[INST] something like, do you regret sort of pushing your body to all those places that resulted in those injuries? He was his response was like, no, I'm stronger now, so I don't know if that's justification, but that certainly describes the mindset that, yeah, head through the wall, that doesn't uh it's almost not dramatic, like, look, I got this injury, it's so I'm so, like, brave and special for overcoming this injury. He's just he's just that's part of the job and he gets the job done, but like, that job, it's a [/INST] lot of injuries, one of those talks, I gave Travis and that team at that at that particular tournament was at the very beginning of the camp, after the tournament, I said to them. Listen, my, my vision, I shared my vision with them. I said my vision is You know in seven years because that was 2005. I said in seven years I want to have a U. S. Team that steps on the mat that is ready to kick ass. And in order to get there, all of you guys can be a part of this team and part of this process, but in order to get there, you guys have to be the first ones to practice, you have to be the last ones to leave because we have to work harder than the rest of the world because we're up against all odds. I said I am sick of America being a laughingstock of judo and being the first round, easy match warm up for everybody else. I said if you get injured, you're not going to be on the side with with you know with an ice bag on taking off rounds and then get back on the mat the next day and tell me you're okay if you can train the next day, you can train today. So there's no injury. The only time you leave in this dojo is if the ambulance has to take you out of here, you know? And I do think subliminally Travis bought into that message and heard that message then said if I'm going to be a champion, that's the way I'm going to do it and he did and he he embodied it, he lived it man, there were many times in europe where I said, dude, just tape it up, we'll go off to the side, just take the day off, like take the rest of the day off your beat up, you can't do it, it's not, I'm going to tape it up on the table, you know, you don't need to right now. And he said no, since I'm doing it, you know, the ambulance isn't taking me out and just just my wrist, it's just my ankle, it's just my, just my ankle, you know? </s><s>[INST] Yeah, I love it. Yeah. What about the the other really big thing is you comment on a little bit is the weight cut. So early in his career he was 81 kg And that was presumably not so difficult, but later in his career he is 81 Kg and it's becoming more and more difficult. So that's, that's the other thing with him is uh so I've known a lot of really, really tough people at the highest level is broken by the way cut. Like that can break the toughest minds and it doesn't seem to have broken him and he delivered on it often on like insane weight cuts. So um just as a coach, what do you think about his particularly his mind and the challenge of the weight cut, [/INST] it was part of his process, it was part of his way of getting ready for battle </s><s>[INST] suffering. [/INST] Yeah, it really was. And if I'm gonna suffer this much, then I'm gonna make my opponents pay for all the suffering that I went through to get here. That was his mindset. Um, later on in his career, you're right. Like a lot of times Travis, he would never step on a scale until he got to the tournament. You know, and even when he gets to the tournament, he'd weigh like 90 kg. He'd show up at the tournament, nine kg over and like you have to, but I never, it was just an expectation of making weight. Not making weight was never an option for any of our athletes. And Travis knew it. And he said as a professional, my job is to make weight. If I don't make weight, he, you know, he, he was never going to allow that to happen and he was never going to allow us to come to him and say, hey, I told you because he losing wasn't an option. Making a weight was not making weight was not an option for him ever either. But you know, a lot of times he wouldn't even, he'd be nine kg over on the plane going over to the tournament and have to make wait three days later. You know, he didn't break 86 kg until the day before the tournament. He had five kg over the day before. That was his way. But he would do three workouts, you know, wake up in the morning workout, then he'd eat, then he worked out in the afternoon, then heated again. Then he'd he'd work out again at night and then he'd reward himself. He had worked out three times today. He'd go have a, you know, a mountain dew, you know, or chocolate bark, You know, and then his next morning he's back up to 87 and he would never touch wait, I'll tell the warning of Lance that that's a When he he wasn't on wait for more than like five minutes, </s> |
<s>[INST] that, like just like you said, the one is close and yet distant in terms of our understanding of it. Do you, are you also captivated by the fact that this very conversation is happening because two brains are communicating the, I guess what I mean is the subjective nature of the experience, if you can take a small attention into the mystical of it, the consciousness or or when you're saying you're captivated by the idea of the brain, are you talking about specifically the mechanism of cognition? Or are you also just like at least for me it's almost like paralyzing the beauty and the mystery of the fact that it creates the entirety of the experience, not just the reasoning capability, but the experience. Well, [/INST] I definitely resonate with that latter thought and I I often find discussions of artificial intelligence to be um disappointingly narrow. Uh you know, speaking of someone who has always had an interest in in, in art right, it was </s><s>[INST] just going to go there because it sounds like somebody who has an interest [/INST] in art. Yeah, I mean there there there there are many layers to, you know, to full bore human experience and um in some ways it's not enough to say, oh well don't worry, you know, we're talking about cognition but we'll add emotion, you know, there's there's there's an incredible scope to what humans go through in every moment and ah and yes, so it's that's part of what fascinates me is that is that our brains are producing that. Ah but at the same time it's so mysterious to us how, you know, like we literally our brains are literally in our heads producing this </s><s>[INST] experience and yet [/INST] there and yet there's there it's so mysterious to us and so in the scientific challenge of getting at the the actual explanation for that is so overwhelming. That's just I don't know that certain people have fixations on particular questions and that's always that's just always been mine. </s><s>[INST] Yeah, I would say the poetry that is fascinating and I'm really interested in natural language as well. And when you look at artificial intelligence community, it always saddens me how much when you try to create a benchmark for the community together around how much of the magic of language is lost when you create that benchmark, that there's something, would we talk about, experience the music of the language, the wit, the something that makes a rich experience. Something that would be required to pass. The spirit of the turing test is lost in these benchmarks. And I wonder how to get it back in because it's very difficult the moment you try to do, like real good rigorous science, you lose some of that magic when you try to study cognition in a rigorous scientific way, it feels like you're losing some of the magic the seeing cognition in a mechanistic way that ai at this stage in our history. [/INST] Okay, I I agree with you. But at the same time 11 thing that I found really exciting about that first wave of deep learning models in cognition was um there was the fact that the people who were building these models were focused on the richness and complexity of human cognition. So an early debate in cognitive science which I sort of witnessed as a grad student was about something that sounds very dry which is the formation of the past tense. But there were these two camps, one said well the mind encode certain rules and uh it also has a list of exceptions because of course you know the rule is addie D but that's not always what you do. So you have to have a list of exceptions. And um and then there were the connection ists who you know evolved into the deep learning people who said well you know if you look carefully at the data, if you look at actually look at corporate language corpora, it turns out to be very rich because yes there there are there's a you know, they're most verbs that and you know you just tack on E. D. And then there are exceptions but they're also there's also there are there are rules that you know there's the exceptions aren't just random, they there are certain clues to which which which verbs should be exceptional. And then there are some exceptions to the exceptions. And there was a word that was uh kind of deployed in order to capture this, which was a quasi regular. In other words, there are rules, but it's it's messy and there there's there's structure even among the exceptions. And and it would be, yeah, you could try to write down, we could try to write down um the structure and some sort of closed form, but really the right way to understand how the brain is handling all this. And by the way, producing all of this is to build a deep neural network and trained it on this data and see how it ends up representing all of this richness. So the way that deep learning was deployed in cognitive psychology was that was the spirit of it. It was about that richness. Um and that's something that I always found very, very compelling still do. </s><s>[INST] Is it is there something especially interesting and profound to you in terms of our current deep learning neural network, artificial neural network approaches and the whatever we do understand about the biological neural networks in our brain is there there's some there's quite a few differences are some of them to you either interesting or perhaps profound in terms of ah in terms of the gap we might want to try to close in trying to create a human level intelligence, [/INST] what I would say here is something that a lot of people are saying, which is that all right, one seeming limitation of the systems that we're building now is that they lack the kind of flexibility, um the readiness to sort of turn on a dime when the when the context calls for it, that is so characteristic of human behavior. </s> |
<s>[INST] as we walk along this conversation, you keep opening doors, I don't want to walk into and that I will, but I want to try to stay kind of focused. You mentioned Daniel Dennett, let's lay it out since he sticks to his story upon unintended and then you stick to yours, what is your story? What is your theory of consciousness versus his? Can you clarify his position? [/INST] So, my view, I defend the view known as panpsychism, which is the view that consciousness is a fundamental and ubiquitous feature of the physical world? So it doesn't literally mean that everything is conscious despite the meaning of the word Pan everything psyche mind. So literally that means everything has mind, but the typical commitment of the pants I kissed is that the fundamental building blocks of reality. Maybe fundamental particles like electrons and quarks have incredibly simple forms of experience and that the very complex experience of the human or animal brain is somehow rooted in or derived from this much more simple consciousness at the level of fundamental physics. So </s><s>[INST] I [/INST] mean that's that's a theory that I would justify on the grounds that it can account for this datum of consciousness that it that we are immediately aware of in our experience in a way that I don't think other theories. Can you ask me to contrast that to Daniel Dennett? I think he would just say there is no such data then it says the data for science of consciousness is what he calls hetero phenomenology, which is specifically defined as what we can access from the third person perspective, including what people say. But crucially we're not treating what they say, we're not relying on their testimony as evidence for some un observable realm of feelings and experiences were just treating their what they say as a datum of public observation experiments that we can account for in terms of underlying mechanism. </s><s>[INST] I feel like there's a deeper view of what consciousness is. So, you have a very clear and we'll talk quite a bit about panpsychism, we have a clear view of what, [/INST] you </s><s>[INST] know, almost like a physics view of consciousness. He I think has a kind of view that consciousness is almost the side effect of this massively parallel computation system going on in our our brain that the that the brain is has a model of the world and it's taking in perceptions and it's constantly weaving multiple stories about that world that's integrating the new perceptions and the multiple stories are somehow it's like a google doc collaborative editing and that collaborative editing is the actual experience of what we think of his consciousness. Somehow, the editing is consciousness of this of this story. I mean, that that's that's a theory of consciousness, isn't it? The narrative theory of consciousness or the multiple versions editing, collaborative editing of a narrative theory of consciousness. [/INST] Yeah, he calls it the multiple drafts model. Incidentally, there's a very interesting paper just come out by very good philosopher Luc roll offs defending a pan sickest version of Den. It's multiple drafts model um </s><s>[INST] like a deeper turtle that that was just the [/INST] difference being that this is luke rollovers view, All of the drafts are conscious. So I guess I guess um I guess for </s><s>[INST] Dennett, [/INST] there's sort of no fact of the matter about which of these drafts is the correct one. Um, on rollovers view maybe there's no fact of the matter about which of these drafts is my consciousness, but nonetheless, all the draft correspond to some consciousness and I mean it just seems kind of funny, I guess, I think he calls it dignity and panpsychism. But this but luke is a one of the most rigorous and serious philosophers alive at the moment, I think. And I hate having luke roll offs in an audience if I'm giving a talk because he always cuts straight to the the weakness in your position that you hadn't thought of. And so it's nice, you know, panpsychism sometimes associated with fluffy thinking, but you know, contemporary panpsychism have come out of this tradition, we call analytic philosophy which is rooted in, you know, detailed rigorous argumentation and and it has defended it in that manner. </s><s>[INST] Yeah, those analytic philosophers are sticklers for terminology, it's very fun, very fun group to talk, talk ship with overseas. [/INST] Well, I mean it gets boring if it's if you just start and end defining words right? I think starting with defining words, it's good actually. The philosopher Derek Parfitt said when, when he first was thinking about philosophy, he went to a talk in analytic philosophy and he went to a talk in continental philosophy and he decided that the problem with the continental philosophy, if it was really on rigorous, really imprecise the problem with the analytic philosophy is it was just not about anything important. And he thought there was more chance of working within analytic philosophy and asking some more meaningful and more profound questions than there was in working continental philosophy and making it more rigorous now they're both horrific stereotypes and you know, I don't want to get nasty emails from either of these groups, but that that there's something there's something to what he was saying, </s> |
<s>[INST] the attacks that work on the invitation model. In some cases at least transfer to the original model. That's incredible and terrifying. Okay, that's amazing work. [/INST] And that shows that again, real world systems actually can be easily. 4th, and in our previous work, we also show this type of black box attacks can be effective cloud vision Api is as well. </s><s>[INST] So that's for natural language and provision. Let's let's talk about another space that people have some concern about which is autonomous driving as sort of security concerns. Um, that's another real world system. So do you have, should people be worried about adversary in which you're learning attacks in the context of autonomous vehicles that use like Tesla autopilot for example, that uses vision as a primary sensor for perceiving the world and navigating in that world. What do you think from your stop sign work in the physical world? Um, should people be worried, how hard is that attack? [/INST] So actually there has already been like, there have always been um, like research showing that for example, actually, even with Tesla, like if you put a few stickers on the road, it can actually once arrange in certain ways. It can fool the that's </s><s>[INST] right, but I don't think it's actually been, I'm not, I might not be familiar, but I don't think it's been done on physical worlds, physical roads yet, meaning I think is with the projector in front of the Tesla. So it's a, it's a physical, so as you're on the other side of the side of the sensor, but you're not instill the physical world. The question is whether it's possible to orchestrate attacks that work in the actual physical like end to end attacks? Like not just a demonstration of the concept but thinking, is it possible on the highway to control Tesla? I [/INST] Think there are two separate questions. One is the feasibility of the attack and I'm 100% confident that's the attack is possible. And there are several questions whether, you know, someone will actually go, you know, deploy that attack. I hope people do not do that. two separate </s><s>[INST] questions. So the question on the word feasibility. So to clarify feasibility means it's possible. It doesn't say how hard it is because, uh, to implement it. So sort of the barrier, like how, how much of a heist it has to be like how many people have to be involved, What is the probability of success that kind of stuff? And couple with how many evil people that are in the world that would attempt such an attack? Right that. But the, to my question is, is it sort of, uh, you know, uh, I talked to you on musk and ask the same question. He says it's not a problem. It's very difficult to do in the real world that, you know, this won't be a problem. He dismissed it as a problem for adversarial attacks on the Tesla, of course he happens to be involved with the company. So he has to say that. But I mean, let me linger and a little longer do you? Yeah. So you know, where does your confidence that it's feasible come from? And what's your intuition, how people should be worried and how we might be, how people should defend against it. How Tesla, how Waymo, how other autonomous vehicle companies should defend against sensor based attacks on whether on lighter or on vision or so on. [/INST] And also even for light actually there has been research shown like no, no, no, no. But </s><s>[INST] it's really important to pause. There's really nice demonstrations that is possible to do. But there's so many pieces that it's kind of like um it's it's kind of in the lab now it's in the physical world, meaning it's in physical space, the attacks. But it's very like you have to control a lot of things to pull it off. Uh it's like the difference between opening a safe when you have it and you have unlimited time and you can work on it versus like breaking into like the crown stealing the crown jewels or whatever. Right? [/INST] So one way to look at this in terms of how real these attacks can be one way to look at it. Is that actually you don't even need any sophisticated attacks. Already. We've seen, you know many real world example of the incidents were showing that the vehicle was making the wrong decision, </s><s>[INST] decision without [/INST] attacks. Right. Right. Right. That's one way to demonstrate. And this is also like so far with many talks about work in this adversary setting. Um showing that today's learning system, they are so vulnerable to the ever serial setting but at the same time actually we also know that even in natural settings these learning systems they don't generalize well and hence they can really misbehave. And there's certain situations like what we have seen and hence I think using that as an example you can show these issues can be </s><s>[INST] real, they can be real. But so there's two cases. One is something, it's like perturbations can make the system misbehave versus make the system do one specific thing that the attacker wants. As you said, the targeted that seems that seems to be very difficult. Like a extra level of difficult step in the, in the real world. But from the perspective of the passenger of the car, you know, I don't think it matters either way whether it's yeah, it's misbehavior or a targeted attack. Okay. And [/INST] also that's why I was also saying earlier, like one defense is this multi model defense and more of these consistent checks and so on. So in the future. I think also it's important that for this autonomous vehicles, they they have a lot of different sensors and they should be combining all these sensory readings to arrive at the decision and the interpretation of the world and so on and the more of these sensory inputs they use and the better they combined the sensory inputs. The hydrate is going to be attacked and hence I think that is a very important direction for us to move towards? </s> |
<s>[INST] Oh, that's fascinating. Yeah. Okay, this is true. I mean, yeah, there's some basics. I'm focusing too much on the silver lining because the people that were like matt lab code. Yeah. Single variable, single letter variable names to those kinds of things [/INST] and its accessibility. Right. It's I want my my code to be open source, but it is it's good to have anyone can download it. But is it really open source if it's written so cryptically so poorly that no one can really use it to its full functionality. Have I really published my work and that weighs on on me, I feel guilty for my own inadequacies as a programmer, you can only do so much. So, I've already learned quantum chemistry in astrophysics so, you know. </s><s>[INST] Uh yeah, I mean, there's there's uh, there's all kinds of ways to contribute to the world. One of them is publication, but publishing code is a fascinating way to contribute to the world. Even if it's very small, very basic element, uh great code. I guess. I was also kind of criticizing the software engineering process versus like, which is a good thing to do is code that's readable almost like without documentation is readable. It's understandable. The variable names, the structure, all those kinds of things [/INST] and that's the dream. </s><s>[INST] That's the dream dumb question. What do you uh, [/INST] tell me you don't question. I want to hear it. </s><s>[INST] Okay. Uh, I mean, okay, this is the question about beauty. It's way too general. It's very impossible to, it's like asking, what's your favorite band? What's your favorite music band? [/INST] I thought you meant wavelength band. I was like, I definitely have a favorite with my friends. Absolutely. </s><s>[INST] Well, it's hard to narrow down. Huh? Okay. What what do you is the most beautiful idea in science? [/INST] It's a dumb question. Do you try that question again? Probably. </s><s>[INST] I have a really good question to ask. You [/INST] don't over sell it. I have a question to ask. You </s><s>[INST] know? Yeah. What what do you is the most beautiful idea in uh, science? Something you just find inspiring or just maybe the reason you got into science or the reason you think science is cool. [/INST] My favorite thing about science is kind of the connection between the scales. So when I was little and I wanted to know about space, I really felt that it would make me feel powerful to be able to predict the heaven, something so much larger than myself that felt really powerful. It was almost a selfish desire. And that's what I wanted. There was some control to being able to know exactly what the sky would do. And then as I got older and I got more into astronomy and I didn't just want to know how the stars moved. I wanted to know how the planets around them moved and, and then as I got deeper into that field, I really didn't care that much about the planet, I want to know about the atmospheres around the planets and then the molecules in those atmospheres and what am I mean? So I ended up shrinking my scale until it was literally the quantum scale. And now all my work, the majority of my work is on this insane quantum scale. And yet I'm using these literal tiny tiny tools to try and answer the greatest questions that we've ever been able to ask and crossing of scales from the quantum to the astronomical. That's so cool, isn't </s><s>[INST] it? It spends the entirety, the tiny and the huge. That's that's the cool thing about I guess being a quantum master chemist is you're using the tools of the tiny to look at the heavenly bodies, the giant stuff [/INST] and the potential life out there. That this is the thing that connects us that you can't escape the rules of the quantum world and how universal they themselves are despite being probabilistic and that makes me feel really pleased to be in science. But in a really humbling way, it's no longer this thirst for power. Um I I feel less special. The more work I do less exceptional, the more work I do. I feel like humans and the earth and our place in the universe is less and less exceptional. And yet I feel so much less lonely. And so it's been a really good tradeoff that I've lost power, but I've gained company, </s><s>[INST] wow, that's a beautiful answer. I don't think there's a better way to actually end it. You're right. I asked a mediocre question and you came through, uh you made the question good by brilliant [/INST] dancing. </s><s>[INST] Um you're the Michael Jordan and I'm the who's the dentist? Right. I'll be the Dennis Rodman. This is, [/INST] I don't know enough about basketball. I mean literally you've reached the peak of my basketball knowledge because I know those people are basketball. That's it. Pros I believe. But only because I watched space jam I think </s><s>[INST] are there books or movies in your life long ago or recently? Do you have any time for books and movies had an impact on you? What ideas did you take away? [/INST] I absolutely have time for books and movies. I try as best I can to not work very hard. Um I mostly fail. I should point out, but I think I'm a better scientist when I don't work evenings and weekends. I'm, if I get four good hours in a day, I often don't, I often get eight crappy hours, you know, emails, meetings, bad code, data processing. But if I can get four high quality scientific hours, I just stopped working for the day because I know it's diminishing returns after that. So I have a lot of time. I try to make as much time as I can. </s> |
<s>[INST] I think the great mask debate of the 2020 will probably be the thing that ultimately leads to World War III. I've been very surprised how tense, how much like division this one little arguably silly thing has led to. I think a lot of people sort of project their like, it's almost like not wearing a mask is a statement of sovereignty, of freedom, of like saying, fuck you to the man, the government, the centralized power, or the dishonesty or the in the scientific community, all those kinds of things. And then wearing a mask is a sort of kind of signaling of various kind of social aspects, I don't know, I'm not paying attention to it, I actually tuned out, I was part of a group of scientists that were looking into like, do masks work, this very interesting question. To me, it was an interesting question. I sort of roll in to ask that very interesting question. Because I think it is an interesting scientific question. But then I quickly realized that just as I was doing this like scientific exploration of this very interesting question about viral particles, like what kind of things. Like from a scientific perspective, how do we prevent the spread of a pandemic? Forget COVID, any pandemic, super deadly or not deadly. Like there's tools, there's testing, there's masks, there's all these tools, how well do they work? And then I realized, you know, in April or so, it became a tool of politics, a tool of philosophy. And that's when I sort of pulled out. So it's fascinating. I think it's a canvas on which people project their emotions and I guess until I got caught up in that kind of, so there's nothing fundamental [/INST] I suppose to their disagreement. Yeah, not that I'm aware of, but he is, you know, he's written some about in his books, the problem with centralization. I mean, a lot of his writing addresses that and he actually points to, I think Switzerland is the best government in the world because it's decentralized. So there's that, I don't think he has any, I'm not to speak for him, but I don't think he's voiced any specific critiques </s><s>[INST] on Bitcoin per se, could be wrong about that. It's just maybe his flavorful language [/INST] and the way he likes to communicate. And the other theory is that maybe he's playing 4D chess and having a Twitter boating accident, you know? So I don't believe in Bitcoin, I've sold all my Bitcoin. Oh, I see. Sorry, the boating accident in Bitcoin is this, I guess it's proverbial by this point where it's the way you lose your Bitcoin. So if someone comes after you and says, hey, you know, whether it's a government or an individual is coming after you saying, give me all your Bitcoin or pay these taxes in Bitcoin, you go, oh, I had a boating accident and lost them all. Lost them all. </s><s>[INST] So, yeah. But back to the fundamental nature of space-time, let me ask you, because we're kind of left it, I'd like to go back to this idea of, that you said that everything we think, say, or do occurs within the bounds of space-time. So first of all, maybe you can comment on, what do you mean in this context about space-time, but also about the nature of truth, like how much of all of this is knowable? How much of this is accessible to us humans? How much uncertainty, like we're talking about, is there in the world? Are you, do you fall in a place where we can reason deeply about this world and it's knowable, or is it mostly chaos and we're just holding on [/INST] for dear life? Yeah, I think, I said that all action occurs within the bounds of space-time. The other thing, everything we say, do, or make, the other thing is that everything we say, do, or make starts out as an idea. So there's this concept of universal Darwinism, which basically applies Darwinian principles, but outside of the biological sphere. So we could say that this kind of gets into Richard Dawkins memetics, </s><s>[INST] that even ideas are competing, reproducing, recombining. That idea is so powerful, by the way. I don't think it's been understood fully. I think in the 21st century, in the digital world, from my perspective in artificial intelligence, there's yet to be some profound things to be discovered [/INST] about this whole construct, full construct. Agreed completely. It's been called an acid actually, and then it just strips away all of the non-informational components of something and just strips it down to its bare bones. </s><s>[INST] I have a quote in there somewhere about that, but... I'm sorry, which is called an acid, the ideas? The universal Darwinism. Darwinism applied broadly outside of the ideas? [/INST] By broadly, and I will condition all of this with saying that a lot, most of my thinking is shaped by a book I wrote recently called The Case Against Reality, which introduced me to this concept, but it tied into Darwinism that I've used more broadly in the past, looking at things like money and economics. So the book, The Case Against Reality, by Donald Hoffman, he has a quote in the book that describes universal Darwinism. It says, quote, universal Darwinism can, without risk of refuting itself, address our key question. Does natural selection favor true perceptions? If the answer happens to be no, then it hasn't shot itself in the foot. The uncanny power of universal Darwinism has been likened by the philosopher Dan Dennett to a universal acid. And Dan Dennett says, quote, there is no denying at this point that Darwin's idea is a universal solvent capable of cutting right to the heart of everything in sight. The question is, what does it leave behind? I have tried to show that once it passes through everything, we are left with stronger, sounder versions of our most important ideas. Some of the traditional details perish and some of these are losses to be regretted, but good riddance to the rest of them. What remains is more than enough to build on, unquote. So the way I would interpret that is that life itself, I've come to view life as information propagating through flesh and that we are, I guess DNA is a quadratic code. I think it's four letters, maybe versus a binary, zeros and ones. And we are ideas, we are strategies competing with each other. And nature is that which selects, it's what selects the winning ideas, the ones that are most fit </s> |
<s>[INST] Yeah. And now we're talking about the narrative, it's very matter. Looking back what is the most beautiful or surprising idea in deep learning or ai in general that you've come across, you've seen this field explode uh and grow in interesting ways. Just what what cool ideas like, like we made you sit back and go small, big or small. [/INST] Well, the one that I've been thinking about recently the most probably is the </s><s>[INST] the [/INST] transformer architecture. Um so basically uh neural networks have a lot of architectures that were trendy have come and gone for different sensory modalities, like for vision, audio, text, you would process them with different looking neural nuts. And recently we've seen these this convergence towards one architecture, the transformer and you can feed it video or you can feed it, you know, images or speech or text and it just gobbles it up and it's kind of like a Bit of a general purpose computer that is also trainable and very efficient to run in our hardware. And so this paper came out in 20 </s><s>[INST] 16, I [/INST] want to say </s><s>[INST] attention is all you need [/INST] attention is all you need, You </s><s>[INST] criticized the paper title in retrospect that it wasn't um it didn't foresee the bigness of the impact that it was going to [/INST] have. I'm not sure if the authors were aware of the impact that that paper would go on to have probably weren't, but I think they were aware of some of the motivations and design decisions behind the transformer and they chose not to, I think expand on it in that way in the paper. And so I think they had an idea that there was more um, than just the surface of just like, oh we're just doing translation and here's a better architecture. You're not just doing translation. This is like a really cool differentiable optimism, efficient computer that you've proposed and maybe they didn't have all of that foresight. But I think it's really interesting, </s><s>[INST] isn't it funny, Sorry to interrupt that that title is memorable that they went for such a profound idea. They went with the I don't think anyone used that kind of title before. Right, [/INST] attention is all you need. It's like a meme or something. </s><s>[INST] It's not funny that one. Like maybe [/INST] if </s><s>[INST] it was a more serious title didn't have the impact. [/INST] Honestly, I yeah, there is an element of me that honestly agrees with you and prefers it this way. </s><s>[INST] If [/INST] it was two grand it would over promise and under deliver potentially. So you want to just mean your way to greatness </s><s>[INST] that should be a t shirt. So you, you tweeted the transformers, a magnificent neural network architecture because it is a general purpose differential computer. It is simultaneously expressive in the forward pass optimize Herbal via back propagation and gradient descent and efficient. High parallelism compute graph. Can you discuss some of those details, expressive. Optimize herbal, efficient, yeah. From memory or or in general whatever comes to your [/INST] heart. You want to have a general purpose computer that you can train on arbitrary problems like say the task of next word prediction or detecting if there's a cat in the image or something like that and you want to train this computer so you want to set its its weight. And I think there's a number of design criteria that sort of overlap in the transformer simultaneously. That made it very successful. And I think the authors were kind of deliberately trying to make this really powerful architecture and so basically it's very powerful in the forward past because it's able to express um very general computation as sort of something that looks like message passing, You have notes and they all store vectors and these nodes get to basically look at each other and it's each other's vectors and they get to communicate and basically notes, get to broadcast. Hey, I'm looking for sir, certain things and then other nodes get to broadcast. Hey, these are the things I have those are the keys in the values. </s><s>[INST] So it's not just attention. [/INST] Exactly, transformer is much more than just the attention component that's got many pieces architectural that went into it. The residual connection of the way it's arranged, there's a multi layer perceptron and there the way it's stacked and so on. Um but basically there's a message passing scheme where nodes get to look at each other, decide what's interesting and then update each other. And so I think the, when you get to the details of it, I think it's a very expressive function so it can express lots of different types of algorithms and forward pass. Not only that, but the way it's designed with the residual connections. Later normalization is the softmax attention and everything. It's also optimize double this is a really big deal because there's lots of computers that are powerful that you can't optimize or they are not easy to optimize using the techniques that we have, which is back propagation ingredients sent. These are first order methods, very simple optimizers really. And so um you also needed to be optimized. Herbal. Um and then lastly, you wanted to run efficiently in our hardware. Our hardware is a massive throughput machine, like gps, they prefer lots of parallelism. So you don't want to do lots of sequential operations. You want to do a lot of operations serially. And the transformer is designed with that in mind as well. And so it's designed for our hardware and is designed to both be very expressive in the forward pass but also very optimized able in the backward pass. </s><s>[INST] And you said that uh the residual connections support a kind of ability to learn short algorithms fast and first and then gradually extend uh longer during training. Yeah. What's what's the idea of learning short algorithms? [/INST] Right. Think of it as So basically a transformer is a series of blocks, right? And these blocks have attention and a little multi layer perception. And so you you go off into a block and you come back to this residual pathway and then you go off and you come back and then you have a number of layers arranged sequentially. And so the way to look at it I think is because of the residual pathway in the backward pass, the gradients sort of flow along it uninterrupted because addition distributes the gradient equally to all of its branches. So the gradient from the supervision at the top just floats directly to the first layer and the all the residual connections are arranged so that in the beginning of during initialization they contribute nothing to the residual pathway. Um so what it kind of looks like is imagine the transformer is kind of like a python function like a death. And um you get to do various kinds of like lines of code. Um so you have 100 layers deep transformer, typically they would be much shorter, say 20 so 20 lines of code and you can do something in them. And so think of during the optimization, basically what it looks like is first you optimize the first line of code and then the second line of code can kick in and the third line of code can kick in and I kind of feel like because of the residual pathway and the dynamics of the optimization, You can sort of learn a very short algorithm that gets the approximate answer, but then the other layers can sort of kick in and start to create a contribution and at the end of it you're you're optimizing over an algorithm that is 20 lines of code Except these lines of code are very complex because it's an entire block of a transformer. You can do a lot in there. What's really interesting is that this transformer architecture actually has been remarkably resilient, basically a transformer that came out in 2016 as the transformer you would use today except you reshuffle from the layer norms, the normalization have been reshuffled to a pre norm um formulation and so it's been remarkably stable, but there's a lot of bells and whistles that people have attached on it and try to improve it. I do think that basically it's a, it's a big step in simultaneously optimizing for lots of properties of a desirable neural network architecture and I think people have been trying to change it, but it's proven remarkably resilient. Um but I do think that there should be even better architecture is potentially, </s> |
<s>[INST] And then the rest is what you're saying, it's uh maybe you mean it's not technically sophisticated but it's social engineering sophisticated, [/INST] very sophisticated on that end. Very sophisticated. </s><s>[INST] It's a fascinating study of [/INST] that establishment of trust and then using that trust to defraud that victim. That is something </s><s>[INST] I wish obviously all of these folks are really good at hiding. I wish you could [/INST] tell </s><s>[INST] their stories in a way, which is why you're fascinating as you're able to tell these stories now because it is studying human nature [/INST] by </s><s>[INST] exploiting it. But you get to understand like our weak points are um our hope, our desire to trust others also sort of the the weak points and the failures of of digital systems and at scale humans have to connect. It's fascinating is um this is a weird question um asking for a friend uh is spearfishing itself illegal? What's the legality here? [/INST] Oh it's all illegal. Absolutely, it is. But </s><s>[INST] absolutely. So here here's what, Okay, let me let me construct an example. So if if my friend were to spearfish like a ceo right and get their information and after they get controls they have their twitter account, they tweet something loving and positive? What's the crime? [/INST] Unauthorized access of advice? </s><s>[INST] What will be the punishment? Do you think [/INST] that becomes questionable. So, so no monetary loss or was there a monetary loss? Probably not. Alright, so you have to figure out who the victim is before charges are pressed now the crime would be another unauthorized access. All right. But no real victim on that unless you know the person whose account you took over takes, you know, exception to that. Um no monetary loss. But </s><s>[INST] there's not really standard like fines [/INST] is going to happen. </s><s>[INST] Right? I mean, that's that's kind of interesting because it's, it's, so when I got the ransom where um, when I get with the zero day attacks on the Qnap Mass, you know, they basically say the criminal is Qnap the company for having so many security vulnerabilities there. Uh like you are the victim of Q naps incompetence. That's the way they kind of phrase it. [/INST] And see, I don't agree with that. I don't agree with that at all. So solar winds, let's let's so I've got 100 and 30 page class action lawsuit printed out at the house. I've been going through it. That catalogs how solar winds lied for years about their vulnerabilities and they lied to investors. The people who came in, the auditors would, who they would hire would, you know, they would uh not pay attention to them when they said, you know, you've got these issues? They would say go away. I should like that for years until solar winds, you know the attacks become apparent. Um my view on that is that the only person responsible for the crime are the criminals who did the attacking the actual criminals? Not not solar winds now, does that mean the solar winds isn't isn't all funked up? They are and there needs to be some accounting in place. But the the the only individual, the only people responsible for crime are the criminals and that's either online in the physical world, what have you can be being an idiot is not being, being, being criminally negligent is and I think that that solar winds is certainly responsible not not responsible, they're culpable for what happened. </s><s>[INST] Can you actually tell folks about solar winds? What is it? Um what what what was what are some interesting things that you're aware of? The [/INST] solar winds was very it was it provided a background, a backbone of security for hundreds thousands of different companies. If you looked at, a lot of security companies were using solar winds that would that would allow you to get a snapshot of the entire system that they were working on. So what happens is is you get a Russian group that comes in and they basically they hack into solar winds and get access to it and it allows them to view every single thing. I mean every single thing about every single client that solar winds had at that point. So entire snapshots of all the I. P. That were that was going on all the emails, all the communications, every single secret that was going on with those companies. If the company had software like Microsoft it allowed them to look at the source code of everything that was going on. I mean it's just a complete and total nightmare alright. And something that you are not going to to recover from. You're not I mean it's done at that point. Um you know, there's not been a lot of news lately about it but the fact of the matter is is that's the type of attack that's a catastrophic attack. </s><s>[INST] So there's a huge amount of information that was red saved elsewhere probably. And so now there's people sitting on information. [/INST] Absolute. So think about one of the attack vectors is has been Microsoft Outlook 365 things like that. This allowed the Attackers to look at the source codes of that. So they have the source code now. So they go through it line by line or other vulnerabilities, let's find new vulnerabilities new zero days and you know, I said zero days aren't common but this opens up an entire new threat surface all of a sudden. So it's a it's a completely catastrophic attack. Once all the chips are down, everything is tallied up, people are gonna be like yeah we're done, we're </s><s>[INST] done alright. This whole computer [/INST] thing, we tried it we were walking away. That's </s><s>[INST] terrifying because so you're saying that there's not been obvious big negative impact from that yet, but like [/INST] there's been a lot of negative impact. But we're just starting, </s> |
<s>[INST] I think it's similar to science to interrupt, but there's a, there's times when people think like with gestures you can control devices, like basically gesture based control of things. It feels similar to me because like it's so compelling. That was just like, like tom cruise, I can control stuff with my hands. But like when you get down to, it's like, well why don't you just have a touch screen or why don't you just have like a physical keyboard and mouse? It's uh, so that chat was always, yeah, it was perplexing to me. I, I still feel augmented reality, Even virtual realities in that ballpark in terms of being a compelling interface. I think there's going to be incredible rich applications just how you're thinking about it, but they won't just be the interface to everything. It'll be its own thing that will create uh, like amazing magical experience in its own right. [/INST] Absolutely. Which is I think kind of the right thing to go about. Like what's the magical experience with that. Um, but that interface specifically, </s><s>[INST] how did you discover that [/INST] for? Um, I just thought, okay, we'll have this attack, we can build any child, but we want, we have the most at that point, the most sophisticated tech that other companies have. Um, I mean startups, obviously not, probably not bigger ones, but still because we've been working on it for a while. So I thought, okay, we can build, build any conversation. So let's just create a scale from 1 to 10 And one would be conversations that you pay to not have and 10 would be conversation would pay to have. And I mean obviously we want to build conversations, people would pay to, you know, to actually have and so for the whole, you know, for a few weeks, me and the team, we're putting all the conversations we were having during the day on the scale and very quickly, um, you know, we figured out that all the conversation that we would pay to never have were um, conversation we were trying to cancel Comcast or talk to customer support or make a reservation or just talk about logistics with a friend when we're trying to figure out where someone is and where to go or um, all sorts of, you know, setting up, um, scheduling meetings. That was just a conversation. We definitely didn't want to have um, basically everything task oriented. It was a one because if there was just one button for me to just not even a button, if I could just think and there was some magic BC that would just immediately transform that into an actual, you know, um, into action that would be perfect. But the conversation there was just this boring, not useful and dull and very also very inefficient thing because it was so many back and forth stuff and as soon as we look to the conversation that we would pay to have those were the ones that, Well first of all therapists because we actually paid to have those conversations and would also try to put like dollar amounts. So you know if I was calling calm because I would pay $5 to not have this one hour talk on the phone, I would actually pay straight up like money, hard money. Uh but it just takes a long time. It takes a really long time. But as soon as we started talking about conversation that we would pay for. Those were therapists, all sorts of therapists, coaches, old friend. Um someone I haven't seen for a long time. Stranger on a train. Weirdly stranger stranger in a line for coffee and ice back and forth with that person was like a good five, solid 56, Maybe not a 10, maybe I won't pay money but at least I won't pay money to not have one. That was pretty good. Some intellectual conversations for sure. But more importantly the one thing that really was um I was making those very important and very valuable for us. Um We're the conversation where we could the where we could be pretty emotional. Yes, some of them were about being witty and about intellectual, being intellectually stimulated but those were interestingly more rare. Uh and most of the ones that we thought were very available were the ones where we could be vulnerable and interestingly when we could talk more </s><s>[INST] so we, like I [/INST] could put me on the team, so we're talking about it like, you know, a lot of these conversations, like a therapist. I mean it was mostly me talking or like an old friend and I was like opening up and crying and it was again me talking. Um, and so that was interesting because I was like, well maybe it's hard to build a chatbot that can talk to you very well and in a witty way, but maybe it's easier to build the chat about that could listen. So that was, that was kind of the first, the first notch in this direction. And then when my, when my friend died, we just built, you know, at that point we were kind of still struggling to find the right application and I just felt very strong that all the child boss would build so far just meaningless and this whole grind, the startup grind and how do we get to, you know, the next fundraising and you know, how can I talk? You know, talking to other founders and what's your investors and how are you doing? Are you killing it? Because we're killing it. I just felt that this is just, </s><s>[INST] uh, intellectually for me. It's exhausting having encounter those folks. [/INST] It just felt very, um, very much a waste of time. </s> |
<s>[INST] So that's telepathic communications from human to human is fascinating. You're reminding us that there's there's uh to biological systems on both ends of that communication. And the one the easier I guess maybe half as difficult thing to do in the hope with neuralink is that we can communicate with an Ai system. So where one side of that is a little bit more controllable. But but even just that is exceptionally difficult. Let's talk [/INST] about two to neuronal systems talking to each other. Suppose that GPT four tells GPT three. Hey, give me all your knowledge. It's ready. I have 10 times more hardware. I'm ready. Just beat me. What's GPT three gonna do? He's gonna say, oh, here's my 10 billion parameters. No, no way. The simplest way and perhaps the fastest way for GPT three to transfer all this knowledge to its older body. That has a lot more hardware is to regenerate every single possible human sentence that it can possibly create. </s><s>[INST] Keep talking, [/INST] keep talking and just re encode it all together. So maybe what language does is exactly that it's taking one generative cognitive model. It's running it forward to emit utterances. That kind of makes sense in my cognitive frame and it's re encoding them into yours through the parsing of that same language. </s><s>[INST] And I think the conversation might might actually be the most efficient way to do it. And not just talking but interactive. So talking back and forth asking questions [/INST] interrupting. So GPT 34 will constantly be interrupted. Just annoying </s><s>[INST] and annoying. And yeah, [/INST] but that is also that as we're interrupting each other, there's all kinds of misinterpretations that happen that as basically when my students speak I will often know that I'm misunderstanding what they're saying and I'll be like hold that thought for a second. Let me tell you what I think I understood which I know is different than what you said. Then I'll say that and then someone else in the same zoom meeting will basically say, well, you know, here's another way to think about what you just said. And then by the third iteration, we're somewhere completely different. That if we could actually communicate with full, you know, neural network parameters back and forth of that knowledge and idea encoding would be far inferior because the re encoding with our own as we said last time, emotional baggage and cognitive baggage from our unique experiences through our shared experiences, distinctive codings in the context of all our unique experiences is leading to so much more um diversity of perspectives. And again, going back to this whole concept of this entire network of all of human cognitive systems connected to each other and sort of how ideas and memes permeate through that. That's what really creates a whole new level of human experience through this reasoning layer and this computational layer that </s><s>[INST] obviously lives on top of our cognitive layer. So you're one of these aforementioned cognitive systems mortal but thoughtful and you're connected to a bunch like you said students, your wife, your kids. What do you, in your brief time here on earth? This is the meaning of life episode. So what do you hope this world will remember you as, what do you hope your legacy will be? [/INST] I don't think of legacy as much as uh, maybe most people think it's kind of funny. I'm consciously living the present. Many students tell me, you know, give us some career advice. I'm like, I'm the wrong person. I've never made a career plan. I still have to make one. I am. It's funny to be both experiencing the past and the present in the future, but also consciously living in the present. And just, you know, there's a conscious decision we can make, do not worry about all that, which again goes back to the under lucky one kind of thing of living in the present and being happy, winning and being happy losing. And um there's a certain freedom that comes with that. But again, a certain sort of I don't know if temerity of living for the present. But if you if you step back from all of that where basically my my current modus operandi this is live for the present, make you know, every day, the best you can make and just make the local blip of local maximum of the universe of the awesomeness of the planet and the town and the family that we live in, both academic family and, and you know, biological family, um, make it a little more awesome by being generous to your friends, being generous to the people around you, being kind to your enemies. And uh, you know, just showing level around you can't be upset at people if you truly love them, if somebody yells at you and insult you every time you say the slightest thing. And yet when you see them, you just see them with love. It's a beautiful feeling. It's like, you know, I'm feeling exactly like I, when I look at my three year old who's like screaming even though I love her and I want her good, she's still screaming and saying, no, no, no, no, no. And I'm like, I love you, genuinely love you, but I can, I can sort of kind of see that your brain is kind of stuck in that little motive, anger and you know, there's plenty of people out there who don't like me and I see them with love as a child that is stuck in a cognitive state that they're eventually going to snap out of or maybe not. And that's okay. So there's that aspect of sort of experiencing life with the best intentions and you know, I love when I'm wrong, I had a friend who was like one of the smartest people I've ever met, who would basically say, oh, I love it when I'm wrong because it makes me feel human and it's so beautiful. I mean she's really one of the smartest people I've ever met and she was like, such a good feeling and I love being wrong. But there's, you know, there's something about self improvement, there's something about how do I not make the most mistakes but attempt the most rights and do the fewest wrongs, but with the full knowledge that this will happen, that's one aspect. So, so so through this life in the present, what's really funny is, and that's something that I've experienced more and more, really thanks to you. And through this podcast is this enormous number of people who will basically comments, wow, I've been following this guy for so many years now, or wow, this guy has inspired so many of us in computation, biology and so and so forth, and like, I don't know any of that, but I'm only discovering this now through these sort of sharing our emotional states and our cognitive states with a water audience. We're suddenly, I'm sort of realizing that wow, maybe I've had a legacy. Like basically I've trained generations of students from MIT And I've put all of my courses freely online since 2001. So basically all of my video recordings of my lectures have been online since 2001. So countless generations of people from across the world well meet me at a conference and say like it was at this conference where somebody heard my voice, it's like, I know this voice, I've been listening to your lectures and it's just such a beautiful thing where like, we're sharing widely and who knows which students will get where from whatever they catch out of these lectures, even if what they catch, it's just inspiration and passion and drive. So there's this intangible legacy code that every one of us has through the people we touch. One of my friends from undergrad basically told me, oh, my mom remembers you vividly from when she came to camp from, like, I didn't even meet her, she's like, no, but she she sort of saw you interacting with people and said, wow, he's exuding this positive energy. And there's there's that aspect of sort of just motivating people with your kindness, with your passion, with your generosity and with your, you know, just selflessness of of, you know, just, just just give, it doesn't matter where it goes. I've been to conferences were basically before, you know, I'll ask them a question and then they'll come back or like, there was a company where I ask somebody a question, they said, oh in fact, this entire project was inspired by your question three years ago at the same conference, Like, wow, </s> |
<s>[INST] you know, I'm sorry to interrupt you, but uh you just made me realize that that's exactly right. So the visualizations you're creating while you're sometimes talking about abstractions, the actual visualization is a explicit low level example. Yes. So there's an actual, like in the code, you have to say what the what the vector is, what's the direction of the arrow? What's the magnitude? Uh Yeah, so that's you're going, the visualization itself is actually going to the bottom [/INST] I think, and I think that's very important. I also think about this a lot in writing scripts where even before you get to the visuals um the first instinct is to I don't know why. I just always do. I say the abstract thing. I say the general definition, the powerful thing and then I fill it in with examples later. Always it will be more compelling and easier to understand when you flip that. And instead you let someone's brain do the uh pattern recognition, you just show them a bunch of examples. The brain is going to feel a certain similarity between them. Then by the time you bring in the definition or by the time you bring in the formula, it's articulating a thing that's already in the brain that was built off of looking at a bunch of examples with a certain kind of similarity and what the formula does is articulate what that kind of similarity is rather than being a a high cognitive load set of symbols that needs to be populated with examples later on assuming someone still with you. </s><s>[INST] What is the most beautiful or awe inspiring idea you've come across in mathematics? [/INST] I don't know man, </s><s>[INST] maybe it's an idea of exploring your videos. Maybe not what like just gave you pause. [/INST] It's the most beautiful idea. </s><s>[INST] Small or big. So [/INST] I think often the things that are most beautiful are the ones that you have like a little bit of understanding of but certainly not an entire understanding. It's a little bit of that mystery. That is what makes it beautiful. </s><s>[INST] Almost the moment of the discovery for you personally, almost just that leap of uh ha moment. [/INST] So something that really caught my eye, I remember when I was little there were these like um I think the series was called wooden books or something. These tiny little books that would have just a very short description of something on the left and then a picture on the right. I don't know who they're meant for, but maybe it's like loosely Children or something like that, but it can't just be Children because of some of the things that was describing on the last page of one of them. Some were tiny in there was this little formula that on the left hand had a sum over all of the natural numbers. It's like 1/1 to the S plus 1/2 to the X. Plus 1/3 to the S. On and onto an infinity. Then on the other side had a product over all of the primes. And it was a certain thing had to do with all the primes. And like any good young math enthusiast, I'd probably been indoctrinated with how chaotic and confusing the primes are which they are. And seeing this equation where on one side you have something that's as understandable as you could possibly get the counting numbers. And on the other side is all the prime numbers. It was like </s><s>[INST] this. Whoa! [/INST] They're related like this. There's a there's a simple description that includes like all the primes getting wrapped together like this. This is like the oiler product for zeta function. As I later found out the equation itself essentially encodes the fundamental theorem of arithmetic that every number can be expressed as a unique set of primes to me still, there's I mean I certainly don't understand this equation or this function all that well, the more I learned about it, the prettier it is the idea that you can this is sort of what gets you representations of primes. Not in terms of crimes themselves, but in terms of another set of numbers that are like the non trivial zeros of the zeta function. And again, I'm very kind of in over my head in a lot of ways as I try to get to understand it. But the more I do it always leaves enough mystery that it remains very beautiful to me. So </s><s>[INST] whenever there's a little bit of mystery just outside of the understanding that and by the way, the process of learning more about it, how does that come about? Just your own thought? Or are you reading reading [/INST] or is the </s><s>[INST] process of visualization itself revealing more to you [/INST] visuals help? I mean in one time when I was just trying to understand like analytic continuation and playing around with visualizing complex functions. This is what led to a video about this function. It's titled something like visualizing the Riemann zeta function. It's one that came about because I was programming and tried to see what a certain thing looked like. And then I looked at it and like, whoa, that's elucidating. And then I decided to make a video about it. Um but I mean, you try to get your hands on as much reading as you can. You you know, in this case, I think if anyone wants to start to understand it, if they have like a a math background of something like they studied some in college or something like that. Um like the Princeton Companion to Math has a really good article on analytic number theory. And that itself has a whole bunch of references and you know, anything has more references and it gives you this like tree to start flying through. And like, you know, you try to understand, try to understand things visually as I go, that's not always possible. But it's very helpful when it does you recognize when there's common themes like in this case cousins of the fourier </s> |
<s>[INST] states. Yeah, we can come back to it. Well, let me ask real quick, like why does, uh, different like local pockets of the universe start remembering stuff? How does memory emerge exactly? So at the origin of the universe, uh, it was very forgetful. That's when the physicists were happiest. There's low, low memory objects, um, which is like ultra low memory objects, which is what the definition of stuff. Okay. And so how does memory emerge? How does, which is [/INST] this, how does this, the temporal stickiness of objects emerge? Um, I, I, I'm going to take a very chemo centric point of view because I can't imagine any other way of doing it. You, you could think of other ways maybe. Um, um, but I would say heterogeneity in matter is where the memory, so you must have enough different ways of rearranging matter for there to be a memory. So what that means is if you've got particles colliding in a box, let's just take a, um, some in some elements in a box, those elements can combine in a combinatorial set of ways. So there's a combinatorial explosion of the number of molecules or minerals or solid objects, bonds being made because there's such a large number, the population of different objects that are possible. This goes back to assembly theory, where assembly theory, there's four types of universes. Right. So you've got basically, um, and this is what one was up earlier where one universe where you've just got everything is possible so you can take all the atoms and combine them and make everything. Then you've got basically what is the assembly combinatorial where you basically have to accrue information in steps. Then you've got, um, assembly observed. right? And then you've got the object assembly going back. So what I'm trying to say is like, if you can take atoms and make bonds, let's say you take a nitrogen atom and add it to a carbon atom, you find an amino acid, then you add another carbon atom on in a particular configuration, then another one, all different molecules, they all represent different histories. So I would say for me right now, the most simple route into life seems to be through recording memories and chemistry. But that doesn't mean there can't be other ways and can't be other emergent effects. But I think if you can make bonds and lots of different bonds, and they those molecules can have a causal effect on the future. So imagine a box of atoms, and then then you combine those atoms in some way. So you make molecule A from a load of atoms, and then molecule A can go back to the box and influence the box, then you make a prime or a B or ABC, and that process keeps going. And that's where the memories come from is that heterogeneity in the universe from bonding. I don't know if that makes any sense. </s><s>[INST] And it's it's beginning to flourish at the chemistry level. Yeah. [/INST] So the physicists have no no, like not enough. Yeah. I mean, </s><s>[INST] they're like, they're like desperately begging for more freedom. Yeah. And heterogeneous components to play with. Yeah. Yeah. Well, the physicists will claim. Yeah, that's exactly it. [/INST] What do you think about that? That's exactly it. I mentioned already, I think it's significant that whatever physics governs life emerges actually in chemistry, it's not relevant at the subatomic scale, or even at the atomic scale, it's in, well, atomic scale, because chemistry, but like when you get into this, this combinatorial diversity that you get from combining things on the periodic table, that's when selection actually matters, or the fact that some things can exist and others can't exist actually starts to matter. So I think of it like, you don't, you don't study gravity inside the atomic nucleus, you study it in terms of large scale structure of the universe, or black holes, or things like that. And whatever we're talking about, as physics of information or physics of assembly becomes relevant at a certain scale of reality. And, and the transition that you're talking about, I would think of is just when you get a sufficient density, in terms of the assembly space of like the relationship of the overlap and, and the assembly space, which is like a feature of common memory, there is this transition to assembly dominated physics, whatever that is. Oh, like, we're trying to map out exactly what that transition looks like, we're pretty sure, you know, if some of its features, but we haven't done </s><s>[INST] all of that. Do you think if you were there in the early university would have been able to predict the emergence of chemistry and biology? And I asked that because at this stage, as humans, do you think we can possibly predict the length of memory that's that might be able to be formed later on in this pocket of the universe? Like how, how complex is a, what is [/INST] the ceiling of assembly? I think as much time as you have in the past is how much you can predict in the future, because that is actually physical in the system. And you have to have enough time for features of that structure to exist. Wait, let </s><s>[INST] me push back and then wake up. What isn't that? Isn't there somewhere in the universe that's like a shortest path that's been the stretches all the way to the beginning? Yeah, that's [/INST] building some giant monster. Yeah, right. Maybe. Yeah. Yeah. Yeah. So the universe has as much memory as the largest </s> |
<s>[INST] Before this COVID thing, you and Brad had some good conversations. I would say we're friends. What's your, what do you admire most about Brad outside of all the criticism we've had [/INST] about this COVID topic? Well, I think Brett is very smart and he's a very ethical person who wants good things for the world. I mean, I have no reason to doubt that. So the fact that we're on, you know, we're crosswise on this issue is not, does not mean that I think he's a bad person. I mean, the thing that worried me about what he was doing, and this was true of Joe and this was true of Elon, this is true of many other people, is that once you're messaging at scale to a vast audience, you incur a certain kind of responsibility not to get people killed. And I do, I did worry that yeah, people were, people were making decisions on the basis of the information that was getting shared there. And that, that's why I was, and that I think fairly circumspect. I just said, okay, give me the center of the fairway expert opinion at this time point and at this time point and at this time point, and then I'm out, right? I don't have any more to say about this. I'm not an expert on COVID. I'm not an expert on the safety of mRNA vaccines. If something, if something changes, so as to become newsworthy, then maybe I'll do a podcast. So I mean, I just did a podcast on the lab leak, right? I was never skeptical of the lab leak hypothesis. Brett was very early on saying this is a lab leak, right? At a point where my only position was who cares if it's a lab leak, right? Like this, the thing we have to get straight is </s><s>[INST] what do we do given the nature of this pandemic, the lab leak, right? But also we should say that you've actually stated that it is a possibility. Oh, yeah. You just said it doesn't quite matter. [/INST] It did. Oh, yeah. You just said it doesn't quite matter. The time to figure that out. Now, I've actually, I have had my podcast guest on this topic changed my view of this because one of the guests, Alina Chan, made the point that, no, actually the best time to figure out the origin of this is immediately, right? Because in the event you lose touch with the evidence and I hadn't really been thinking about that. If you come back after a year, there are certain facts you might not be able to get in hand, but I've always felt that it didn't matter for two reasons. One is we had the genome of the virus and we could design, we're very quickly designed immediately designing vaccines against that genome. And that's what we had to do. And then we had to figure out how to vaccinate and to mitigate and to develop treatments and all of that. So the origin story didn't matter. Generically speaking, either origin story was politically inflammatory and made the Chinese look bad, right? And the Chinese response to this looked bad, whatever the origin story, right? They're not cooperating. They're letting, they're stopping their domestic flights, but letting their international flights go. I mean, it's just, they were bad actors and they should be treated as such regardless of the origin, right? And I would argue that the wet market origin is even more politically invidious than the lab leak origin. Because lab leak, to my eye, the lab leak could happen to anyone, right? We're all running, all these advanced countries are running these dangerous labs. That's a practice that we should be worried about in general. We know lab leaks are a problem. There've been multiple lab leaks of even worse things that haven't gotten out of hand in this way, but worse pathogens. Why do you think? We're wise to be worried about this. And on some level, it could happen to anyone, right? The wet market makes them look like barbarians living in another century. You got to clean up those wet markets. What are you doing putting a bat on top of a pangolin on top of a duck? It's like, get your shit together. So if anything, the wet market makes them look worse in my view. Now, I'm sure that what they actually did to conceal a lab leak, </s><s>[INST] if it was a lab leak, all of that's going to look odious. Do you think we'll ever get to the bottom of that? I mean, one of the big negative, I would say failures of Anthony Fauci and so on is to be transparent and clear and just a good communicator about gain and functional research. The dangers of that, why it's a useful way of research, but it's also dangerous. Just being transparent about that as opposed to just coming off really shady. Of course, the conspiracy theorists and the [/INST] politicians are not helping, but this just created a giant mess. Yeah, no, I would agree. So that exchange with Fauci and Rand Paul, they went viral. Yeah, I would agree that Fauci looked like he was taking refuge in very lawyered language and not giving a straightforward account of what we do and why we do it. So yeah, I think it looked shady. It played shady and it probably was shady. I don't know how personally entangled he is with any of this, but yeah, the gain of function research is something that I think we're wise to be worried about. And insofar as I judge myself adequate to have an opinion on this, I think it should be banned. Probably a podcast I'll do, if you or somebody else doesn't do it in the meantime, I would like a virologist to defend it against a virologist who would criticize it. Forget about the gain of function research. I don't even understand virus hunting at this point. It's like, I don't even know why you need to go into a cave to find this next virus that could be circulating among bats that may jump zoonotically to us. Why do that when we can sequence in a day and make vaccines in a weekend? What kind of head start do you think you're getting? </s> |
<s>[INST] That's just not factored into his math. That's just not factored into his math. Well, I think that's where I've had a few sort of offline conversation with him. I think he said, because I mentioned I'm talking to the two of you, he said that that's probably where he disagrees about sort of the level of threat that global warming causes. [/INST] Well, Steve Kuhnen is another one. He's a brilliant guy. He lived right close to me in Hudson Valley. He was in the Obama administration energy department. It's K-O-O-N-I-N. He wrote a bestseller that came out recently on skepticism about climate. There are other smart people who somehow feel we can literally adapt our way forward without any constraint on the gases changing the climate. I've spent enough time on this. I think I'm a pretty level-headed reporter when it comes to this issue. I think having some sense that we can adapt our way into the world we're building through relentless climate change with no new normal, remember, more gas accumulating in the air every year. These are not static moments. That's a good thing to do doesn't strike me as smart. I'd probably say that I think it's more sort of a, at least the thing that I take away from Alex is the fact, as you point out, that we need to recognize that fossil fuels is basically the backbone of our society today. We get 80% of our energy from fossil fuels today. Still, as we did 50 years ago, 40 years ago. People have no sense of this. They have the idea because you see so many wind turbines and solar panels and everybody's talking about that this is huge, big things. The reality is, remember, only about a fifth of all energy use is electricity. The rest is processes and heating, industrial processes and so on. Actually, solar and wind right now produces 1% of energy from wind and 0.8% from solar. This is not a huge thing. It's a fairly tiny bit. And growing explosively, but it's from this- Yes, it's absolutely growing. Actually, it's growing slower than what nuclear was growing in the 70s and 80s, which I thought was a fun point, not by a little amount, by like two or three times. We're still talking about something which is somewhat boutique, at least. When you then look out into the future, and I think this is the interesting part of it, when you look out into the future, if you look at the Biden administration's own estimate of what will happen by 2050 today- Still, as we did 50 years ago, 40 years ago, explosively, but from this- Yes, it's up. If all countries do all the stuff that they promised and everything, we will be at 70% fossil fuels by 2050, globally. This is just- Yes, it's a better world. I think it's good that we're now down to 70 instead of 80, but it is still a world that's fundamentally dependent on fossil fuels for almost everything that we really like about the world. And forgetting that, and I think we are doing that in the sense, as you also mentioned, that people say no fossil fuels, and we're in all development organizations, we're now telling the poor countries, you can't get any funding for anything that has to do with fossil fuels. We have literally reduced our investment in oil and gas by more than half since 2014, and much of this is because of climate concerns. This has real world consequences. This is why energy prices have gone up. It's not the only reason, COVID also certainly the war in Ukraine, but this is an underlying systemic reason why fossil fuel costs will go up dramatically. Now, a lot of Greens will tend to say, well, that's great, because we want fossil fuels to be expensive. We want people to be forced over to renewables, but that's very easy to say if you're rich. It's the kind of thing that New Yorkers will say when you go to rich, well-meaning green New Yorkers and say, yes, gasoline should cost $20 a gallon. Well, you don't have a car, you just ride the Metro. It's very easy for you to say that, but lots of people both in the rich world, but in poor parts of the US, but all around the world, their lives are basically dependent on fossil fuels. And so the idea that we're going to get people off by making it so expensive that it becomes impossible for them to live good lives is almost morally reprehensible. And I think Alex has the right point there. We need to get people to realize we're not going to get off fossil fuels anytime soon, so we need reasonably affordable fossil fuels for most of the world. And that's of course why we need to focus so much more on the innovation so that we can get to the point where we no longer need fossil fuels as soon as possible. But to say to everyone, look, we're going to make fossil fuels expensive way before we have the solution is just terrible. And so much is on the rich countries of the world. I did a conversation recently with Johann Rockstrom, who's a famed sustainability scientist in Stockholm, actually Potsdam now. And he's come up with the idea of planetary boundaries. There's lots of things he has said that as a journalist, I'm still looking into about that. Planetary boundaries? Yeah, that there are limits to what Earth can absorb in human. Our use of water, phosphorus, our carbon dioxide loading in the atmosphere, there are these boundaries if we cross them, we're in a hot zone, a danger zone. He's an interesting thinker. But on this point, last year at the Glasgow Climate Talks, he gave a very important talk about the equity thing here that you... He basically laid out a landscape saying the rich nations of the world need to greatly ramp up their reduction of emissions or what they're going to pay poor countries to do to allow poor countries, some of which have fossil resources like in Africa, a danger zone, to have the carbon space, to own whatever space or time is left to be able to develop their fossil fuels as a fundamental right. Because also they're starting from this little baseline. Ghana hasn't contributed squat to the global warming problem in terms of emissions. Ghana has natural gas. And right now, this month, environmental groups are outside the World Bank, actually tonight, saying this was on their list of dirty projects. World Bank should stop financing Ghana's right to get gas out of the ground, to develop its economy, get its people less poor, make them more productive, innovative parts of humanity. To me, that's really reprehensible. One of the other projects on their list as a World Bank gotcha, like how dare they give money, was for a new fertilizer factory in Bangladesh that is designed to get three times as much fertilizer from the same amount of natural gas as the old plants that are now dormant. This is in a time when we're facing high energy prices, high gas prices, high food prices, and food insecurity is spreading rapidly when a country like Bangladesh has millions of rice farmers who need urea, tablets to put in their rice fields, and to say how dare they finance that because there's a fossil fuel involved is immoral. </s> |
<s>[INST] Your support means a lot and inspires me to keep this series going. And now here's my conversation with kevin scott. [/INST] You've </s><s>[INST] described yourself as a kid in a candy store Microsoft because of all the interesting projects that are going on. Can you try to do the impossible task and give a brief world wind view of all the spaces that Microsoft is working in both research and product [/INST] if you include research, it becomes even even more difficult. So, so I think broadly speaking, Microsoft's product portfolio includes everything from, you know, big cloud business, Like a big set of sass services. We have, you know, sort of the original or like some of what are among the original productivity uh software products that everybody used. We have an operating system business, we have a hardware business where we make everything from uh computer mice and headphones to high end high end personal computers and laptops. We have a fairly broad ranging research group where like we have people doing everything from economics research. So like there's this really uh really smart young economist Glenn while who like my group works with a lot who's doing this research on these things called Radical markets. Uh like he's written an entire entire technical book about about this whole notion of a radical market. So like the research group sort of spans from that human computer interaction to artificial intelligence. And we have we have ghetto, we have linked in, we have search advertising and news business and like probably a bunch of stuff that I'm embarrassingly not recounting in this gaming </s><s>[INST] to Xbox and so on [/INST] gaming for sure. Like I was I was having a super fun conversation this morning with with Phil spencer. So when I was in college there was this game that Lucas arts made called day of the tentacle that my friends and I played forever and like we're you know, doing some interesting collaboration now with the folks who made uh day of the tentacle and I was like completely nerd ng out with tim schafer, like the guy who wrote a day of the tentacle this morning, just a complete Fanboy, which you know, sort of it like happens a lot, like, you know, Microsoft has been doing so much stuff, it's such breath for such a long period of time that you know, like being cto like most of the time, my job is very, very serious and sometimes like I get to get caught up in like how amazing it is to be able to have the conversations that I have with the people, I I get to have them with, </s><s>[INST] you had to reach back into the sentimental and what's the, the radical markets and economics. So [/INST] the idea with radical markets is like, can you come up with new market based mechanisms to uh you know, I think we have this, we're having this debate right now, like does capitalism work like free markets work? Can the incentive structures that are built into these systems produce outcomes that are creating sort of equitably distributed benefits for every member of society, you know, and I think it's a reasonable, reasonable set of questions to be asking. And so what glenn and so like, you know, one motor thought there, like if you have doubts that the, that the markets are actually working, you can sort of like tip towards like, okay let's let's become more socialist and uh you know like have central planning and you know governments or some other central organization is like making a bunch of decisions about how you know, sort of work gets done and uh you know like where the you know where the investments and where the outputs of those investments get distributed. Glenn's notion is like lean more into uh like the market based mechanism. So like for instance, uh you know, this is one of the more radical ideas like suppose that you had a radical pricing mechanism for assets like real estate where you were, you can be a bit out of your position and in in your home. Uh you know, for instance, so like if somebody came along and said, you know, like I've I can find higher economic utility for this piece of real estate that you're running your business in. Uh like then uh like you either have to, you know, sort of bid to sort of stay or like the thing that's got the higher economic utility uh you know, sort of takes over the asset, which would make it very difficult to have the same sort of rent seeking behaviors that you've got right now because uh like if you did speculative bidding, like you would very quickly like lose a whole lot of money. And so like the prices of the assets would be sort of like very closely index to uh like the value that they could produce And like because like you have this sort of real time mechanism that would force you to sort of mark the value of the asset to the market, then it could be taxed appropriately. Like you couldn't sort of sit on this thing and say like this house is only worth $10,000 when like everything around it is worth 10 million, </s><s>[INST] let's finish. So it's an incentive structure that where the prices match the value much better. So [/INST] Glenn does a much, much better job than I do and selling and I probably picked the world's worst example, you know? And and and but like it's it's intentionally provocative, you know, so like this whole notion like, you know, like I'm not sure whether I like this notion that like we could have a set of market mechanisms where I could get bit out of out of my property, you know? But, but you know, like if you're thinking about something like Elizabeth Warren's wealth tax for instance, like you would have, I mean it'd be really interesting and like how you would actually set the price on the assets and like you might have to have a mechanism like that if you put a tax like that in </s><s>[INST] place. It's really interesting that that kind of research at least tangentially touching Microsoft research. Yeah, you're really thinking broadly uh maybe you can speak to this connects to a I so we have a candidate. Andrew yang who kind of talks about artificial intelligence and the concern that people have about our, you know, automation impact on society. And arguably Microsoft is at the cutting edge of innovation in all these kinds of ways. And so it's pushing ai forward. How do you think about combining all our conversations together here with radical markets and socialism and innovation in ai that uh that Microsoft is doing and then Andrew Yang's worry that uh that that will that will result in job loss for the lower and so on. How do you think about that? I think [/INST] it's sort of one of the most important questions in technology, like maybe even in society right now about how is ai going to develop over the course of the next several decades and like what's it going to be used for and like what uh what benefits will it produce and what negative impact will it produce? And you know, how who gets to steer this whole thing, You know, I'll say it at the highest level. One of the real joys of of getting to do what I do at Microsoft is Microsoft has this heritage as a platform company. And so you know, like Bill Bill has this thing that he said a you know, a bunch of years ago where you know, the measure of a successful platform is that it produces far more economic value for the people who built on top of the platform than is created for the the platform owner or builder. And I think we have to think about ai that way, </s><s>[INST] like platform [/INST] Yeah, it has to like, it has to be a platform that other people can use to build businesses to fulfill their creative objectives to be entrepreneurs, to solve problems that they have in there, working in their lives. Uh it can't be a thing where there are a handful of companies uh sitting in a very small handful of cities geographically who are making all the decisions about what goes into the Ai and uh and like and then on top of like all this infrastructure, then build all of the commercially valuable uses for it. So like, I think like that's bad from a you know, sort of, you know, economics and sort of equitable distribution of value perspective, like, you know, sort of back to this whole notion of, you know, like to the markets work, But I think it's also bad from an innovation perspective because like I have infinite amounts of faith in human beings that if you give folks powerful tools, they will go do interesting things and it's more than just a few tens of thousands of people with the interesting tools that should be millions of people with the tools. So it's sort of like, you know, you think about the Um the steam engine in the late 18th century, like it was you know, maybe the first large scale substitute for human labor that we've built like a machine. Uh and uh you know, in the beginning when these things are getting deployed, the folks who uh got most of the value from the steam engines were the folks who had capital so they could afford to build them and like they built factories around them and businesses and the experts who knew how to build and maintain them. But access to that technology democratized over time like now, like uh like an engine, it's not uh it's not like a differentiated thing. Like there isn't one engine company that builds all the engines and all of the things that use engines are made by this company and like they get all the economics from all of that, like, like fully demarcated. Like they're probably, you know, we're sitting here in this room and like even though they don't, there are probably things, you know, like the, the memes, gyroscope that are in both of our folks, like they're like little engines, you know, sort of everywhere. They're just a component in how we build the modern world, like AI needs to get there. </s><s>[INST] Yeah, so that's a really powerful way to think if we think of aI as a platform versus a tool that Microsoft owns as a platform that enables creation on top of it. That's the way to democratize it. That's really, that's really interesting actually and Microsoft throughout its history has been positioned well to do that [/INST] and the you know the tie back to the to this radical markets, things like that. So my team has been working with Glenn on this and jaron Lanier actually. So jarron is the like the sort of father of virtual reality uh like he's one of the most interesting human beings on the planet, like a sweet sweet guy uh and so Jared and Glenn. Uh and folks and my team have been working on this notion of data as labor or like they call it data dignity as well. And so the the idea is that if you you know again going back to this, you know sort of industrial analogy if you think about data is the raw material that is consumed by the machine of ai in order to do useful things. Uh then like we're not doing a really great job right now and having transparent marketplaces for valuing those data contributions so and we all make them like explicitly like you go to linkedin, you sort of set up your profile on linkedin like that's an explicit contribution, like you know exactly the information that you're putting into the system and like you put it there because you have some nominal notion of like what value you're going to get in return but it's like only nominal like you don't know exactly what value you're getting in return, like services free, you know like it's low amount of like perceived out and then you've got all this indirect contribution that you're making just by virtue of interacting with all of the technology that's in your daily life. And so like what Glenn and jarron and this data dignity team are trying to do is like can we figure out a set of mechanisms that let us value those data contributions so that you could create an economy and like a set of controls and incentives that would allow people to like maybe even in the limit like earn part of their living through the data that they're creating. And like you can sort of see it in explicit ways. There are these companies like scale Ai and like there's a whole bunch of them in uh in china right now that are basically data labeling companies. So like you you're doing supervised machine learning. You need you need lots and lots of labeled training data. Yeah. And like those people are getting competent like who work for those companies are getting compensated for their data contributions into the system. And so </s><s>[INST] that's easier to put a number on their contribution because they're explicitly labeling data. But you're saying that we're all contributing data into all kinds of ways and it's fascinating to start to explicitly try to put a number on it. Do you think that's that's possible? [/INST] I don't know it's hard. It really is because you know we don't have as much transparency is uh is I think we need uh in like how the data is getting used and it's you know, super complicated, like, you know, we we you know, I think it's technologist sort of appreciate like some of the subtlety there, it's like, you know, the data, the data gets created and then it gets, you know, it's not valuable. Like the data exhaust that you give off or the, you know, the explicit data that I am putting into the system isn't value valuable. It's super valuable atomically. Like it's only valuable when you sort of aggregated together and to, you know, sort of large numbers. It's true even for these like folks who are getting compensated for like labeling things like for supervised machine learning now, like you need lots of labels to train, you know, a model that performs well. And so, you know, I think that's one of the challenges. It's like how do you, you know, how do you sort of figure out like because this data is getting combined in so many ways like through these combinations, like how the value is flowing. </s><s>[INST] Yeah, that's that's that's tough. Yeah, it's fascinating that you're thinking about this and I was I wasn't even going into this conversation expecting the breadth of uh of research really. That Microsoft broadly is thinking about you're thinking about Microsoft. So if we go back to uh 89 when Microsoft released office or 1990 when they released Windows three point oh, how's the In your view? I know you weren't there the entire, you know through his history, but how is the company changed in the 30 years since as you look at it now, [/INST] the good thing is it's started off as a platform company life, it's still a platform company. Like the parts of the business that are like thriving and most successful or those that are building platforms, like the mission of the company now is the mission's changes, like, changing a very interesting way. So, you know, back in 89 90 like they were still on the original mission which was like put a pc on every desk and in every home uh like and it was basically about democratizing access to this new personal computing technology, which when Bill started the company, integrated circuit microprocessors were a brand new thing and like people were building, you know, homebrew computers, uh you know, from kits, like the way people build ham radios right now. And I think this is sort of the interesting thing for folks who build platforms in general, Bill saw the opportunity there and what personal computers could do and it was like, it was sort of a reach. Like you just sort of imagine like where things were, you know, when they started the company versus where things are now like in success when you democratize the platform, it just sort of vanishes into the platform, you don't pay attention to it anymore. Like operating systems aren't a thing anymore, like they're super important, like completely critical and like, you know when you see one, you know fail like you you just, you sort of understand but like it's not a thing where you're, you're not like waiting for, You know, the next operating system thing uh in the same way that you were in 1995, right? Like 1995, like, you know, we had rolling stones on the stage with the Windows 95 roll out like it was like the biggest thing in the world, everybody would be lined up for the way that people used to line up for iPhone, but like, you know, eventually and like this isn't necessarily a bad thing, like it just sort of, you know, it the successes that it sort of becomes ubiquitous, it's like everywhere and like human beings when they're technology becomes ubiquitous, they just sort of start taking it for granted. So the mission now that's satya re articulated five plus years ago now when he took over his ceo of the company, our mission is to empower every individual and every organization in the world to be more successful and so, you know, again like that's a platform mission and like the way that we do it now is is different, it's like we have a hyper scale cloud, that are building applications on top of like we have a bunch of ai infrastructure that people are building their a i applications on top of we have uh you know, we have </s><s>[INST] a [/INST] productivity suite of software like Microsoft dynamics, which, you know, some people might not think is the sexiest thing in the world, but it's like helping people figure out how to automate all of their business processes and workflows and uh you know, like help those businesses using it to like grow and be more so. It's a it's a much broader vision in a way now than it was back then, like it was sort of very particular thing and like now, like we live in this world where technology is so powerful and it's like such a basic fact of life that it uh you know, that it it both exists and is going to get better and better over time, or at least more and more powerful over time, So like, you know, what you have to do is a platform player is just much </s><s>[INST] bigger, right? There's so many directions in which you can transform, you didn't mention mixed reality, you know? That's uh that's that's probably early days, it depends how you think of it, but if we think in a scale of centuries, just the early days of mixed reality for sure. And so yeah, with Hollande's uh Microsoft is doing some really interesting work there, do you do you touch that part of the effort, what's the thinking do you think of mixed reality as a platform to [/INST] Oh sure. when we look at what the platforms of the future could be so like fairly obvious that like ai is one like you don't have to I mean like that's you know you sort of say it to like someone and you know like they get it. Um But we also think of the like mixed reality and quantum is like these two interesting, you know, potentially </s><s>[INST] computing. Okay, so let's get crazy then. So [/INST] so </s><s>[INST] you're talking about some futuristic things here? Well the mixed reality Microsoft is really not even futuristic is here. It is incredible [/INST] stuff and and and look and it's having it's have an impact right now. Like one of the one of the more interesting things that's happened with mixed reality over the past a couple of years that I didn't clearly see is that it's become the computing device for for folks who for doing their work who haven't used any computing device at all to do their work before. So technicians and service folks and uh people who are doing like machine maintenance, some factory floors. So like they you know because they're mobile and like they're out in the world and they're working with their hands and you know sort of servicing these like very complicated things. Uh there they don't use their mobile phone. Uh and like they don't carry a laptop with them and you know they're not tethered to a desk and so mixed reality like where it's getting traction right now, where Hololens is selling a lot of, a lot of units is for these sorts of applications for these workers. And it's become like, I mean like the people love it, they're like, oh my God, like this </s><s>[INST] is like [/INST] for them, like the same sort of productivity boosts that, you know, like an office worker had when they got their first personal computer. </s><s>[INST] Yeah, but you did mention it's necessarily obvious ai as a platform, but can we dig into it a little bit? How does a I begin to infuse some of the products in Microsoft? So currently providing um training of, for example, neural networks in the cloud or providing uh pre trained models or uh just even providing computing resources, whatever different inference that you want to do using neural networks. Well, how do you think of ai infusing the as a platform that Microsoft can provide? [/INST] Yeah, I mean, I think it's it's super interest. It's like everywhere, and like we we run these uh we run these review meetings uh now where uh it's bee and Sasha and uh like members of SATa's leadership team and like a cross functional group of folks across the entire company who are working on like either a i infrastructure or like have some substantial part of their of their product work using ai in some significant way now. The important thing to understand is like when you think about like how the Ai is going to manifest in like an experience for something that's going to make it better. Like I think you don't want the the ai nous to be the first order thing. It's like whatever the product is and like the thing that is trying to help you do like the ai just sort of makes it better and you know this is a gross exaggeration, but like I yeah people get super excited about like where the Ai is showing up in products and I'm like, do you get that excited about like where you're using a hash table like in your code. Like it's just another, it's a very interesting programming tool but it's sort of a like it's an engineering tool and so like it shows up everywhere, so like we've got dozens and dozens of features now in office that are powered by like fairly sophisticated machine learning. Our search engine uh wouldn't work at all if you took the machine learning out of it. Uh The like increasingly, you know, things like content moderation on our Xbox and X cloud platform, you </s><s>[INST] know when you meet in moderation to be like the recommended is like showing what you wanna look at [/INST] next, It's like anti bullying, </s><s>[INST] so the usual social network stuff that you have to deal [/INST] with. But it's like really it's targeted, it's targeted towards a gaming audience. So it's like a very particular type of thing where you know the line between playful banter and like legitimate bullying is like a subtle one and like you have to like it's sort of tough. Like I have </s><s>[INST] uh I'd love to if we could dig into because you're also you lead the engineering efforts to linked in and uh if we look at, if we look at Lincoln as a social network and if you look at the xbox gaming is just the social components. The very different kinds of I imagine communication going on on the two platforms and the line in terms of bullying and so on is different on the platforms. So how do you, I mean it's such a fascinating philosophical discussion and where that line is, I don't think anyone knows the right answer. Uh, twitter folks are under fire now Jack or twitter for trying to find that line, nobody knows what that line is. But how do you try to uh find the line for, you know, uh trying to prevent abusive behavior and at the same time let people be playful and joke around and that kind of thing. [/INST] I think in a certain way like you know, if you have what I would call vertical social networks, uh it gets to be a little bit easier. So like if you have a clear notion of like what your social network should be used for or like what you are designing a community around then you don't have as many dimensions to your sort of content safety problem as uh, you know, as you do in a general purpose platform. I mean, so like on on linkedin, like the whole social network is about connecting people with opportunity, whether it's helping them find a job or to, you know, sort of find mentors or to, you know, sort of help them like find their next sales lead or uh, to just sort of allow them to broadcast their, their, you know, sort of professional identity to their, um, their network of peers and collaborators and, you know, sort of professional community like that is, I mean, like in some ways like that's very, very broad, but in other ways it's sort of, you know, it's narrow and so like you can build a eyes, uh, like machine learning systems that are, you know, capable with those boundaries of making better automated decisions about, like what is, you know, sort of inappropriate and offensive comments or dangerous comments or illegal content when you have some constraints, you know, same thing with Yeah, same thing with like the, the gaming gaming social network. So for instance, like it's about playing games not having fun. Uh, and like the thing that you don't want to have happen on the platform is why bullying is such an important thing, like bullying is not fun. So you want to do everything in your power to encourage that not to happen. And yeah, but I think it's it's sort of a tough problem in general is one where I think, you know, eventually we're going to have to have some sort of clarification from our policymakers about what it is that we should be doing, like where the lines are because it's tough. Like you don't like in democracy right? Like you don't want uh you want some sort of democratic involvement. Uh like people should have a say in like where, where the lines lines are drawn, like you don't want a bunch of people making like unilateral decisions and like we, we are in a we're in a state right now for some of these platforms where you actually do have to make unilateral decisions where the policy making isn't gonna happen fast enough in order to like prevent very bad things from happening. But we need the policy making side of that to catch up I think is as quickly as possible because you want that whole process to be a democratic thing. Not a, you know, not not some sort of weird thing where you've got a non representative group of people making decisions that have, you know, like national and global impact. </s><s>[INST] And it's fascinating because the digital space is different than the physical space in which nations and governments were established and so what policy looks like globally what bullying looks like globally, what's healthy communication looks like global is, is an open question and we're offering a freaking it out together. Yeah, I mean [/INST] with, you know sort of fake news for instance and </s><s>[INST] do fakes and fake news generated by [/INST] humans. Yeah, so we can talk about defects, like I think that is another like, you know, sort of very interesting level of complexity. But like if you think about just the written word, right? Like we have, You know, we invented papyrus, what, 3000 years ago, where we you know, you can sort of put put word on on paper and then uh 500 years ago, like we we get the printing press, like where the word gets a little bit more ubiquitous. And then like he really, really didn't get ubiquitous printed word until The end of the 19th century when the offset press was invented and then, you know, just sort of explodes and like, you know, the cross product of that and the industrial revolutions need for educated citizens resulted in like This rapid expansion of literacy and the rapid expansion of the word. But like we had 3000 years up to that point to figure out like how to, you know, like what's what's journalism, what's editorial integrity, like what's, you know, what's scientific peer review. And so like you built all of this mechanism to like try to filter through all of the Noise that the technology made possible to like, you know, sort of getting to something that society could cope with and like if you think about just the piece, the PC didn't exist 50 years ago. Uh and so in like this span of, you know like half a century, like we've gone from no digital, you know, ubiquitous digital technology to like having a device that sits in your pocket where you can sort of say whatever is on your mind to like what what would it mary having or mary meeker Just released her new uh like slide deck last week. You know, we've got 50% penetration of the Internet to the global population, like they're like 3.5 billion people who are connected now. So it's like it's </s><s>[INST] crazy, crazy [/INST] like inconceivable like how fast all of this happened. So you know, it's not surprising that we haven't figured out what to do yet, but like we gotta we gotta really like lean into this set of problems because like we basically have three millennial worth of work to do about how to deal with all of this and like probably what amounts to the next decade worth of time. </s><s>[INST] So since we're on the topic of tough, you know, tough, challenging problems, let's look at uh more on the tooling side in ai that Microsoft is looking at the face recognition software. So there's there's a lot of powerful positive use cases for face recognition, but there's some negative ones and we're seeing those in different governments in the world. So how do you, how does much, I think about the use of face recognition software as a platform and governments and companies, how do we strike an ethical balance here? [/INST] Yeah, I think we've articulated a clear point of view. So brad smith wrote a blog post uh last fall, I believe that's sort of like outlined like very specifically what uh you know what our what our point of view is there and you know, I think we believe that there are certain uses to which face recognition should not be put and we believe again that there's a need for regulation there. Like the government should like really come in and say that you know this is this is where the lines are and like we very much wanted to like figuring out where the lines are, should be a democratic process. But in the short term like we've drawn some lines where uh you know, we push back against uses of face recognition technology. Um you know like this city of SAN Francisco for instance, I think has completely outlawed any government agency uh from using face recognition tech. Uh and like that may prove to be a little bit overly broad. Um but for like certain law enforcement things like you, you really, I would personally rather be overly cautious in terms of restricting use of it until like we have, you know, sort of defined a reasonable, you know, democratically determined regulatory framework for uh like where we could and should use it and you know, the other thing there is um like we've got a bunch of research that we're doing in a bunch of progress that we've made on bias there. And like there are all sorts of like weird biases that These models can have. Like all the way from like the most noteworthy one where you know, you may have um underrepresented minorities who are like underrepresented in the training data and then you start learning uh like strange things. But like they're even, you know, other weird things like we've uh I think we've seen in the public research like models can learn strange things like all doctors are men for instance. Uh Yeah, I mean and so like it really is a thing where it's very important for everybody who is working on these things before they push publish uh they launched the experiment. They, you know, push the code, you know online. Are they even publish the paper that they are at least starting to think about what some of the potential negative consequences are. Some of this stuff. I mean this is where you know, like the deep fake stuff I find very worrisome just because they're going to be some very good beneficial uses of like gan generated uh imagery and like and funny enough like one of the places where it's actually useful is we're using the technology right now to generate synthetic synthetic visual data for training some of the face recognition models to get rid of the bias. Uh So like that's one like super good use of the tech but like you know, it's getting good enough now where you know, it's gonna sort of challenge normal human being's ability to like now you're just sort of say like it's it's very expensive for someone to fabricate a photo realistic fake video. And like gans are going to make it fantastically cheap to fabricate a photo realistic fake video. And so like what you assume you can sort of trust is true versus like be skeptical about is about to change. Uh like we're not ready for it. I don't think </s><s>[INST] the nature of truth, right? That's uh it's also exciting because I think both you and I probably would agree that the way to solve to take on that challenge with technology right? There is probably going to be ideas of ways to verify which which kind of video is legitimate, which kind of not. So uh to me that's an exciting possibility most most likely for just the comedic genius that the internet usually creates with these kinds of videos and hopefully will not result in any serious harm. [/INST] And it could be, you know, like I think we will have technology too that may be able to detect whether or not something is fake or real, although the fakes are pretty convincing even like when you subject them to machine scrutiny. But you know that we we also have these increasingly interesting social networks you know that are under fire right now. Um for some of the bad things that they do. Like one of the things you could choose to do with the social network is like you could you could use crypto and the networks to like have content signed, where you could have a like full chain of custody that accompanied every piece of content. So like when you're viewing something and like you want to ask yourself like how you know how much can I trust this? Like you can click something and like have a verified chain of custody that shows like oh this is coming from you know from this source and it's like signed by like someone whose identity I </s><s>[INST] trust. Yeah [/INST] I think having you know having that chain of custody like being able to like say oh here's this video like it may or may not have been produced using some of this deep fake technology but if you've got a verified chain of custody where you can sort of trace it all the way back to an identity and you can decide whether or not like I trust this identity. Like oh no this is really from the White House or like this is really from the you know the office of this particular presidential candidate or it's really from you know Jeff Weiner Ceo of linkedin or satya Nadella Ceo of Microsoft like that might that might be like one way that you can solve some of the problems and so like that's not the super high tech, like we've had all of this technology forever um and but I think you're right like it has to, it has to be some sort of technological thing </s><s>[INST] because the [/INST] the underlying tech that is used to create this is not going to do anything but get better over time and the genie is out of the bottle, there's no stuffing it back in </s><s>[INST] and there's a social component which I think is really healthy for a democracy where people will be skeptical about the thing they watch in general. Uh so you know, which is good skepticism in general is good in your content. So deep fakes in that sense are creating global skepticism about can they trust what they read? It encourages further research. I come from the soviet union where basically nobody trusted the media because you know it was propaganda and that encourage that kind of skepticism encouraged further research about ideas. You're supposed to just trusting anyone source. [/INST] Well look, I think it's one of the reasons why the you know, the scientific method and our apparatus of modern science is so good like because you don't have to trust anything like you like the whole notion of you know like modern science beyond the fact that you know this is a hypothesis and this is an experiment to test the hypothesis and you know like this is a peer review process for scrutinizing uh published results, but like stuff is also supposed to be reproducible. So like, you know, it's been vetted by this process, but like you also are expected to publish enough detail where you know, if you are sufficiently skeptical of the thing, you can go try to reproduce it yourself and like, I don't know what it is. Like, I think a lot of engineers are, are like this where like, you know, sort of this uh like your brain is sort of wired for for skepticism, like you don't just first order trust everything that you see an encounter and like you're sort of curious to understand, you know, the next thing, but I I think it's an entirely healthy, healthy thing. Uh and like we need a little bit more of that right now, </s><s>[INST] so I'm not a large business owner, so I'm just, I'm just a huge fan of many of the Microsoft products. I mean, I still actually, in terms of, I generate a lot of graphics and images and I still use power point to do that. It beats illustrator for me, even professional, a sort of uh, it's fascinating. So, uh, I wonder what is the future of, let's say, a Windows in office look like, do you see it? I mean, I remember looking forward to XP was an exciting when XP was released, just like you said, I don't remember when 95 was released, but uh XP for me, it was a big celebration and when 10 came out I was like, okay, well it's nice, it's a nice improvement. But uh so what do you see the future of these products? Uh [/INST] you know, I think there's a bunch of exciting, I mean on the office front there's going to be this uh like increasing productivity winds that are coming out of some of these ai powered features that are coming. Like the products will sort of get smarter and smarter and like a very subtle way, like there's not going to be this big bang moment where you know, like clip e is going to re emerge and </s><s>[INST] it's going to be in a minute. Okay, we'll [/INST] have to wait, wait, </s><s>[INST] wait, it's clipping coming back. Well it's quite seriously um so injection of Ai, there's not much or at least I'm not familiar sort of assistive type of stuff going on inside the office products in like a clip e style assistant personal assistant. Do you think that there's a possibility of that in [/INST] the future? So I think there are a bunch of like very small ways in which uh like machine learning powered assistive things are in the product right now. So there are uh there are a bunch of interesting things like um the auto response stuff is getting better and better and it's like getting to the point where you know, it can auto respond with like, okay let you know this person is clearly trying to schedule a meeting so it looks at your calendar and it automatically like tries to find like a time in a space that's mutually interesting. Um like we we have um this notion of Microsoft search where it's like not just web search but it's like search across like all of your information that's sitting inside of like your office 3 65 tenants and uh like you know potentially in other products and like we have this thing called the Microsoft graph that is basically an ap I fed aerator that uh you know, sort of like gets you hooked up across the entire breadth of like all of the, you know, like what were information silos before they got woven together with the graph. Um like that is like getting increasing with increasing effectiveness sort of plumbed into the, into some of these auto response things where you're going to be able to see the system like automatically retrieve information for you. Like if you know like I I frequently send out, you know, emails to folks were like I can't find a paper or document or what, there's no reason why the system won't be able to do that for you. And like I think the it's building towards like having things that look more like uh like a fully integrated uh you know assistant, but like you'll have a bunch of steps that you will see before you, it will not be this like big bang thing where like clip he comes back and you've got this like you know manifestation of you know, like a fully fully powered assistant. So I think that's um that's definitely coming in like all of the, you know, collaboration co authoring stuff's getting better, you know, it's like really interesting like if you look at how we use uh the office product portfolio and Microsoft, like more and more of it is happening inside of like teams as a canvas and like it's this thing where you know, you've got collaboration is like at the center of the product and uh like we we we built some really cool stuff that's some of which is about to be open source that are sort of framework level things for doing uh for doing co authoring. Uh </s><s>[INST] So uh is there a cloud component to that? So on the web or is it um forgive me if I don't already know this, but with office 265 we still the collaboration we do if we're doing word we still send the file around. No, so this is [/INST] we're already a little bit better than that and like, you know, so like the fact that you're unaware of it means we've got a better job to do, like helping you discover discover this stuff but yeah, I mean it's already like got a huge huge clock input and like part of you know, part of this framework stuff, I think we're calling it? Like I we've been working on it for a couple of years so like I know the internal code name for it, but I think when we launched it, a bill is called the fluid framework. Um and but like what fluid lets you do is like you can go into a conversation that you're having in teams and like reference like part of a spreadsheet that you're working on, where somebody is like sitting in the excel canvas, like working on the spreadsheet with a you know, chart or whatnot. And like you can sort of embed like part of the spreadsheet in the team's conversation where like you can dynamically updated, like all of the changes that you're making to the to this object are like you know, cord and everything is sort of updating in real time. So like you can be in whatever canvas is most convenient for you to get your work done. </s><s>[INST] So out of my own sort of curiosities engineer, I know what it's like to sort of lead a team of 10, 15 engineers Microsoft has uh I don't know what the numbers are maybe 50, maybe 60,000 engineers [/INST] and a lot, I don't know exactly what the number </s><s>[INST] Is. It's a lot, it's tens of thousands, right? This is more than 10 or 15 what what what I mean? You've uh you've led uh different sizes, mostly large size of engineers, What does it take to lead such a large group into uh continue innovation, continue being highly productive and yet to develop all kinds of new ideas and yet maintain like what does it take to lead such a large group of brilliant people? [/INST] I think the thing that you learn as you manage larger and larger scale is that there are three things that are like very very important for big engineering teams. Like one is like having some sort of forethought about what it is that you're going to be building over large periods of time, like not exactly like you don't need to know that like, you know, I'm putting all my chips on this one product and like this is gonna be the thing but like it's useful to know like what sort of capabilities you think you're going to need to have to build the products of the future and then like invest in that infrastructure like whether and I'm like I'm not just talking about storage systems or cloud api is it's also like what is your development process look like, what tools do you want? Like what culture do you want to build around? Like how you're, you know, sort of collaborating together to like make complicated technical things and so like having an opinion and investing in that is like it just gets more and more important and like the sooner you can get a concrete set of opinions like the better you're going to be like, you can wing it for a while, uh small scales like, you know when you start a company like you don't have to be like super specific about it, but like the biggest miseries that I've ever seen as an engineering leader are in places where you didn't have a clear enough opinion about those things soon enough and then you just sort of go create a bunch of technical debt and like culture debt that is excruciatingly painful to, to clean up. So like that's one bundle of things like the other, the other, you know, another bundle of things is like, it's just really, really important to like have a clear mission. That's not just some cute crap you say, because like you think you should have a mission, but like something that clarifies for people like where it is that you're headed together, like I know it's like probably like a little bit too popular right now, but uh, Yuval Harari book sapiens one of the central ideas in, in his book is that like storytelling is like the quintessential thing for coordinating the activities of large groups of people. Like once you get past Dunbar's number and like I really, really seen that just managing engineering teams, like you can, you can just brute force, things when you're less than 100 and 20 150 folks where you can sort of know and trust and understand what the dynamics are between all the people, but like past that, like things just sort of start to catastrophically fail if you don't have some sort of set of shared goals that you're marching towards. And so like, even though it sounds touchy feeling and you know, like a bunch of technical people will sort of balk at the idea that like, you need to like have a clear like the missions, </s><s>[INST] like very, very, very important is write, write stories, that's how our society, that's the fabric that connects us, all of us is these powerful stories and that works for companies to it works for everything. [/INST] Like it means even down to like, you know, you sort of really think about like a currency for instance, is a story, A constitution is a story. Hour laws are story. I mean like we believe very, very, very strongly in them and thank God we do, but like they are, they're just abstract things, like they're just words like if we don't believe in them, they're nothing. </s><s>[INST] And in some sense, those stories are platforms and uh the kinds some of which itself is creating, right? Platforms in which we define the future. So, last question, what do you get philosophical, maybe bigger than you have a Microsoft? What do you think The next 2030 plus years looks like for computing for technology, for devices, do you have crazy ideas about the future of the world. [/INST] Yeah, look, I think we you know, we're entering this time where we've got we have technology that is progressing at the fastest rate that it ever has and you've got you got some really big social problems like society, scale problems that we have to we have to tackle and so you know, I think we're going to rise to the challenge and like figure out how to intersect like all of the power of this technology with all of the big challenges that are facing us, whether it's uh you know global warming, whether it's like the biggest remainder of the population boom is in uh africa for the next 50 years or so. And like global warming is going to make it increasingly difficult to feed the global population in particular. Like in this place where you're going to have like the biggest population boom um I think we you know like a i is gonna uh like if we push it in the right direction, like you can do like incredible things to empower all of us to achieve our full potential and to um you know like live better lives but like that also means focus on like some super important things like how can you apply it to health care to make sure that uh you know like our quality and cost of and and sort of ubiquity of health coverage is is better and better over time. Like that's more and more important every day is like in the in the United States and like the rest of the industrialized world, western europe china, japan Korea, like you've got this population bubble of like aging, working, you know, working age folks who are, you know, at some point over the next 2030 years, they're going to be largely retired and like you're going to have more retired people and working age people and then like you've got, you know, sort of natural questions about who's gonna take care of all the old folks and who's going to do all the work. And the answer is to like all of these sorts of questions like where you're sort of running into, you know, like constraints of the, you know, the world and of society has always been like what tech is gonna help us get around this. You know, like when I was when I was a kid in the seventies and eighties, like we talked all the time about like uh like population boom, population boom, like we're gonna uh we're not gonna be able to like feed the planet. And like we were like right in the middle of the green revolution where like </s><s>[INST] this this [/INST] massive technology driven uh increase in crop productivity like worldwide and like some of that was like taking some of the things that we knew in the west and like getting them distributed to the, you know to the to the developing world and like part of it were things like um you know, just smarter biology like helping us increase and like we don't talk about like, yep overpopulation anymore because like we can more or less we sort of figured out how to feed the world, like that's a that's a technology story. Um and so like I'm super super hopeful about the future and in the ways where we will be able to apply technology to solve some of these super challenging problems, like I've I've like one of the things that I I'm trying to spend my time doing right now is trying to get everybody else to be hopeful as well because you know, back to Harari, like we we are the stories that we tell, like if we, you know, if we get overly pessimistic right now about like the the potential future of technology, like we, you </s><s>[INST] know, like [/INST] we we may fail to fail to get all of the things in place that we need to like have our best possible future </s><s>[INST] and that kind of hopeful optimism. I'm glad that you have it because you're leading large groups of engineers that are actually defining that are writing that story that are helping build that future, which is super exciting and uh I agree with everything you said except I do hope clip he comes [/INST] back, we </s><s>[INST] miss them, I speak for the people so Callan, thank you so much for talking [/INST] today, thank you so much for having me. It was </s><s>[INST] a pleasure. [/INST] </s> |
<s>[INST] I'm looking at this paper and it's talking about a puppet controlling a puppet controlling a puppet controlling a puppet controlling a puppet controlling a puppet, conceptually easy to understand, but physically impossible it's physically impossible, it's predicting a fair coin toss. I don't know what he's talking about, but there's pictures of puppets controlling puppets. Let me ask you, there's a few things that I want to ask, but we've brought up time quite a bit. You guys tweet about time quite a bit. What is time in all of this? We kind of mentioned it a bunch. Is it not important at all in terms of, is it just a word? Should we be talking about causality mostly? Like Sarah, what do you think? Is, we've talked about like memories. Is that the fundamental thing that we should be thinking about? And time is just a useful measurement device [/INST] or something like that. Well, there's different concepts of time, right? So I think in assembly theory, when we're talking about time, we're talking about the ordering of things. So that's the causal graph part. And so then the fundamental structure of the universe is that there is a certain ordering and certain things can't happen until other things happen. But usually when we colloquially talk about time, we're talking about the flow of time. And I guess Lee and I were actually debating about this this morning. So, in talking on it, walking on the river here, which is a very lovely spot for talking about time, but that the, when the universe is updating, it's transitioning between things that exist now and things that exist now. Um, uh, that's really the flow of time. Uh, so there's, there's, you have to separate out those concepts at bare minimum. And then there's also an arrow of time that people talk about in physics, which is that time doesn't appear to have a directionality in fundamental physics, but it does, um, to us, right? Like we can't go backwards in time and usually we, you know, that would be explained in physics in terms of, um, well, there's a cosmological arrow of time, but there's also the thermodynamic arrow of time of increasing entropy. Um, but what we would say in assembly theory is that there is a clear directionality that universally runs in one direction, which is why something's . It's easy to make, if the universe ones and runs in one direction, it's easy to make processes look reversible. For example, if they have no memory, they're easy to run forward and backwards, which is why the laws of physics that we have now looked the way they do, because they involve objects that have no memory, but when you get to things like us, it becomes very clear that the universe has a directionality associated to it. </s><s>[INST] So it's not reversible at all. It's the, no man ever steps in the same river. I just have to bring that out because you won't know the river. No man ever steps in the same river twice for it's not the same river and he's not the same man. So that that's not reversible, [/INST] any of the same, because you won't know the river. No, but reversibility is an emergent property. So we think of the reversibility loss as being fundamental and the irreversibility as being emergent, but I think what we would say from how we think about it and certainly if this is the case for our perception of time but also what's happening in biological evolution. You can make things reversible, but it requires work to do it. And it requires certain machines to run it forward and backward. And Chiara Marletta was working on some interesting ideas on constructor theory related to that, which is a totally different set of ideas. So you can travel back in time sometimes? Yes. You can't travel actually back in time, but you could reconstruct things that have existed in the past. You're always moving forward in time, but you can cycle through... Like, I mean I can do things- Can I clarify what you just said? Yeah, go for it. Yeah, you travel forward in time. Can I clarify what you just said quickly? You travel forward in time to travel back. </s><s>[INST] Yes. Thank you. [/INST] That really clarified your point. Well no. What Cera is saying is you don't go back in time, you recreate what happened in the past in the future and inspect it again. </s><s>[INST] So in that local pocket of time, it's as if you traveled back in time. [/INST] So I don't, how's that not traveling back in time? Because you're not going back yourself back in time. </s><s>[INST] You are, you're creating that in the future. [/INST] But everything else is the same as it was in the past. No, no, no, no, it's not in registry. I mean, it goes back to the big question. I'm saying, I mean, this is something, I was trying to look up today when I first, we first had this discussion and I was talking to Sarah on Skype and said, by the way, time is the fundamental thing in the universe. She almost hung up on me. Right, but you can even, I mean, if you want to make an analogy to computation, and I think Charles Bennett actually has a paper on this, like about reversible computation and reversible Turing machines, in order to make it reversible, you have to store memory to run the process backwards. So time is always running forward in that. Because you have to write the memory. You can erase the memory. But the point when you go back to zero, right. But the whole point is that in order to have a process that even runs in both directions, you have to start talking about memory to store the information to run it backwards. </s> |
<s>[INST] You are a man of faith christian and you are a man of science. What is the role of religion and of science and society and in the individual human mind and heart like yours? [/INST] Well, I was not a person of faith when I was growing up, I became a believer in my twenties, influenced as a medical student by recognition that I hadn't really thought through the issues of what's the meaning of life? Why are we all here? What happens when you die? Is there a god? Science is not so helpful in answering those questions. So I had to look around in other places and ultimately came to my own conclusion that Atheism, which is where I had been, was the least supportable of the choices because it was the assertion of a universal negative which scientists aren't supposed to do. And agnosticism came as an attractive option but felt a little bit like a cop out. So I had to keep going trying to figure out why the believers actually believe this stuff and came to realize it was all pretty compelling that there is no proof, I can't prove to you or anybody else that God exists. But I can say it's pretty darn plausible and ultimately, what kind of God is it caused me to search through various religions and see well, what, what do people think about that? And to my surprise, encountered the person of jesus christ as unique in every possible way and answering a lot of the questions I couldn't otherwise answer. And somewhat kicking and screaming, I became a christian even though at the time as a medical student already interested in genetics, people predicted my head would then explode because these were incompatible worldviews. They really have not been for me. I am so fortunate. I think that in a given day wrestling with an issue, It can have both the rigorous scientific component and it can have the spiritual component of COVID-19 is a great example. These vaccines are both an amazing scientific achievement and an answer to prayer. When I'm wrestling with vaccine hesitancy and trying to figure out what answers to come up with. I get so frustrated sometimes and I'm comforted, bye reassurances that God is aware of that this is I don't have to do this alone. Um so I know there are people like your friend Sam Harris who feel differently. Sam wrote a rather famous op ed in the new york times when I was nominated as the NIH director saying this is a terrible mistake. You can't you can't you can't have somebody who believes in God running the N. I H he's just gonna completely ruin the place. </s><s>[INST] Well, I have a testimonial Christopher, Hitchens, a devout atheist if I could say so, was a friend of yours and referred to you as quote, one of the greatest living americans and stated that you were one of the most devout believers he has ever met. He further stated that you were sequencing the genome of the cancer that would ultimately claim his life and that your friendship, despite their differing opinions on religion, was an example of the greatest armed truth in modern times. What did you learn from Christopher Hitchens about life or perhaps what is a fond memory you have of this man with whom you've disagreed, but who is also your friend? [/INST] Yeah, I loved hitch, I'm sorry, he's gone iron sharpens iron. There's nothing better for trying to figure out where you are with your own situation and your own opinions, your own worldviews than in counting somebody who's completely in another space and who's got the gift as hitch did of challenging everything and doing so over a glass of scotch or two or three. Uh, yeah, we got off to a rough start where in an interaction, we had a rather highbrow dinner, uh, he was really deeply insulting of a question I was asking, but you know, I was like, okay, that's fine, let's let's figure out how we could have a more civil conversation and then I really learned to greatly admire his intellect and to find the jousting with him. And it wasn't all about faith, although it often was, uh, was really inspiring and innovating energizing. And then when he got cancer, um, I became sort of his ally, trying to help him find pathways through the various options and maybe helped him uh to stay around on this planet for extra six months or so. And I have the warmest feelings of being in his apartment downtown. Um over a glass of wine, talking about whatever. Sometimes it was science who is fascinated by science. Sometimes it was thomas, jefferson, sometimes it was faith and I knew it would always be really interesting. </s><s>[INST] So he's not gone. Yeah. Do you think about your own mortality? Are you afraid of death? [/INST] I'm not afraid. I'm not looking forward to it. I don't want to rush it because I feel like I got some things I can still do here, But as a person of faith, I don't think I'm afraid I'm 71. I know I don't have an infinite amount of time left and I want to use the time I've got in some sort of way that matters. I'm not ready to become a full time golfer, but I don't quite know what that is. I do feel that I've had a chance to do amazingly powerful things as far as experiences and maybe God has something else in mind. I wrote this book 16 years ago, the language of God about science and faith, trying to explain how, from my perspective these are compatible. These are in harmony. They're complementary if you are careful about which kind of question you're asking and to my surprise, a lot of people seem to be interested in that they were tired of hearing the extreme voices Um like Dawkins at one end and people like ken ham and answers in genesis on the other end saying if you trust science, you're going to help And they thought there must be a way that these things could get along. And that's what I tried to put forward. And then I started a foundation bio logos, which then I had to step away from to become NIH director which has just flourished maybe because I stepped away. I don't know. But it now has millions of people who come to that website and they were an amazing meetings and I think a lot of people have really come to a sense that this is okay. I can love science and I can love God. And that's not a bad thing. So maybe there's something more I can do in that space. Maybe that book is ready for a second edition. </s> |
<s>[INST] And yeah, that's what athletic greens is for me. They'll give you one month supply of fish oil when you sign up at athletic greens dot com slash lex. This is the lex Friedman podcast to support it. Please check out our sponsors and now dear friends, here's andre apathy. What is a neural network and why does it seem to do such a surprisingly good job of learning what [/INST] is a neural network? It's a mathematical abstraction of the brain. I would say that's how it was originally </s><s>[INST] developed [/INST] at the end of the day. It's a mathematical expression and it's a fairly simple mathematical expression when you get down to it. It's basically a sequence of matrix multiplies which I really got products mathematically and some non linearity is thrown in. And so it's a very simple mathematical expression and it's got knobs in </s><s>[INST] it. Many knobs, [/INST] knobs and these knobs are loosely related to basically the synapses in your brain. They're trainable. They're modifiable. And so the idea is like, we need to find the setting of the knobs that makes the neural net do whatever you wanted to do, like classify images and so on and so there's not too much mystery. I was saying it like, um, you might think that basically don't want to endow it with too much meaning with respect to the brain and how it works is really just a complicated mathematical expression with knobs and does not need a proper setting for it to do something desirable. </s><s>[INST] Yeah, but poetry is just a collection of letters with spaces, but it can make us feel a certain way. And in that same way when you get a large number of knobs together with whether it's in a inside the brain or inside a computer, they seem to, they seem to surprise us with their power. [/INST] Yeah, I think that's fair. So basically I'm underselling it by a lot because you definitely do get very surprising emergent behaviors out of these neural nets when they're large enough and trained on complicated enough problems like say for example the next word prediction in a massive data set from the internet. And then these neural nets take on pretty surprising magical properties. Yeah, I think it's kind of interesting how much you can get out of even very simple mathematical formalism </s><s>[INST] when your brain right now is talking, is it doing next word prediction or is it doing something more interesting? [/INST] We'll definitely some kind of a generative model that's a gPT like and prompted by you. Um so you're giving me a prompt and I'm kind of like responding to it in a generative way </s><s>[INST] and by yourself perhaps a little bit like, are you adding extra prompts from your own memory inside your head [/INST] definitely feels like you're referencing some kind of a declarative structure of like, memory and so on. And then uh you're putting that together with your prompt and giving away some </s><s>[INST] how much of what you just said has been said by you before? [/INST] Nothing, basically. Right? </s><s>[INST] No, but if you actually look at all the words you've ever said in your life and you do a search, you'll probably said a lot of the same words in the same order before. [/INST] Yeah, could be, I mean, I'm using phrases that are common etcetera, but I'm remixing it into pretty sort of unique sentence at the end of the day, but you're right, definitely. There's like a ton of remixing, </s><s>[INST] why you didn't you? It's like magnus Carlsen said, I'm redid 2900, whatever, which is pretty decent. I think you're talking very uh you're not giving enough credit in your own nuts here, Why do they seem to what's your best intuition about this emergent behavior? [/INST] It's kind of interesting because I'm simultaneously underselling them, but I also feel like there's an element to which I'm over. Like it's actually kind of incredible that you can get so much emergent magical behavior out of them despite them being so simple mathematically. So, I think those are kind of like two surprising statements that are kind of just juxtaposed together and I think basically what it is is we are actually fairly good at optimizing these neural nuts and when you give them a hard enough problem, they are forced to learn very interesting solutions in the optimization and those solutions basically have these emergent properties that are very interesting. There's </s><s>[INST] wisdom and knowledge in the knobs [/INST] and so </s><s>[INST] this representation that's in the knobs doesn't make sense to you. Intuitively the large number of knobs can hold the representation that captures some deep wisdom about the data. It has looked at a lot of knobs, it's [/INST] a lot of knobs and somehow, you know, so speaking concretely, um one of the neural nets that people are very excited about right now are are G. P. T. S, which are basically just next word prediction networks. So you consume a sequence of words from the internet and you try to predict the next word. And once you train these on a large enough dataset, um they you can basically uh prompt these neural nets in arbitrary ways and you can ask them to solve problems and they will so you can just tell them you can you can make it look like you're trying to um solve some kind of a mathematical problem and they will continue what they think is the solution based on what they've seen on the internet. And very often those solutions look very remarkably consistent. Look correct potentially. </s><s>[INST] Do you still think about the brain side of it? So as neural nets as an abstraction or mathematical abstraction of the brain, you still draw wisdom from from the biological neural networks or even the bigger question. So, you're a big fan of biology and biological computation. What impressive thing is biology do doing to you that computers are not yet. That gap, I [/INST] would say I'm definitely on, I'm much more hesitant with the analogies to the brain than I think you would see potential in the field. Um, and I kind of feel like certainly the way your network started is everything stemmed from inspiration of the brain by the brain. But at the end of the day, the artifacts that you get after training they are arrived at by a very different optimization process than the optimization process that gave rise to the brain. And so I think, uh, I kind of think of it as a very complicated alien artifact. Um, it's something different. I'm sorry, the neural nets that were </s><s>[INST] training, they [/INST] are complicated uh, alien artifact. I do not make analogies to the brain because I think the optimization process that gave rise to it is very different from the brain. So there was no multi agent self play kind of setup. Uh, an evolution, it was an optimization that is basically a what amounts to a compression objective on a massive amount of data. </s><s>[INST] Okay, so artificial neural networks are doing compression and biological neural networks now [/INST] to survive and they're </s><s>[INST] not really doing anything. They're they're an agent in a multi agent self place system that's been running for a very, very [/INST] long time. That said, evolution has found that it is very useful to to predict and have a predictive model in the brain. And so I think our brain utilizes something that looks like that as as a part of it but it has a lot more you know catches and gizmos and value functions and ancient nuclei that are all trying to like make it survive and reproduce and everything else. </s><s>[INST] And the whole thing through embryo genesis is built from a single cell. I mean it's just the code is inside the D. N. A. And it just builds it up like the entire organism with [/INST] crazy in </s><s>[INST] the head and legs. And like it does it pretty well it may [/INST] not be possible. </s><s>[INST] So there's some learning going on. There's some there's some there's some kind of competition going through that building process. I mean I don't know where if you were just to look at the entirety of history of life on earth, what do you think is the most interesting invention? Is it the origin of life itself? Is it just jumping to your carry? It's is it mammals as the humans themselves almost sapiens the the origin of intelligence or highly complex intelligence or what or is it all just a continuation of the same kind of process? [/INST] Certainly I would say it's an extremely remarkable story that I'm only like briefly learning about recently all the way from actually like you almost have to start at the formation of earth and all of its conditions and the entire solar system and how everything is arranged with jupiter and moon and the habitable zone and everything. And then you have an active earth that's turning over material and then you start with a bit of genesis and everything. So it's all like a pretty remarkable story. I'm not sure that I can pick like a single unique piece of it that I find most interesting. Um, I guess for me as an artificial intelligence researcher, it's probably the last piece. We have lots of animals that uh, you know, are, are not building technological society, but we do. And um, it seems to have happened very quickly. It seems to have happened very recently and uh, something very interesting happened there that I don't fully understand. I almost understand everything else kind of I think intuitively, but I don't understand exactly that part and how quick it was. </s><s>[INST] Both explanations would be interesting. One is that this is just a continuation of the same kind of process. There's nothing special about humans that would be deeply understanding. That would be very interesting that we think of ourselves as special. But it was obvious it was already written in the, in the code that you would have greater and greater intelligence emerging and then the other explanation, which is something truly special happened. Something like a rare event whether it's like crazy rare event, like a space odyssey, what would it be if you say like the invention of fire or the as Richard Wrangham says the beta males deciding a clever way to kill the alpha males by collaborating. So just optimizing the collaboration really the multi agent aspect of the multi agent and that really being constrained on resources and trying to survive the collaboration aspect is what created the complex intelligence, but it seems like it's a natural outgrowth of the evolution process. What could possibly be a magical thing that happened? Like a rare thing that would say that humans are actually human level intelligence is actually a really rare thing in the universe. [/INST] Yeah, I'm hesitant to say that it is rare by the way, but it definitely seems like it's kind of like a punctuated equilibrium where you have um lots of exploration and then you have certain leaps, sparse leaps in between. So of course like origin of life would be one um you know, D. N. A. Sex eukaryotic eukaryotic life. Um The endo symbiosis event where the eight bacteria, you know, just the whole thing and then of course emergence of consciousness and so on. So it seems like differently. There are sparse events where massive amount of progress was made. But yeah it's kind of hard to pick one. So </s><s>[INST] you don't think humans are unique. I gotta ask you, how many intelligent alien civilizations do you think are out there? And uh is their intelligence different or similar to ours? [/INST] Yeah, I've been preoccupied with this question quite a bit recently basically the fermi paradox and just thinking through and and the reason actually that I am very interested in the origin of life is fundamentally trying to understand how common it is that there are technological societies out there uh in space. And the more I study it, the more I I think that um uh there should be quite a few quite a lot. </s><s>[INST] Why haven't we heard from them? Because I I agree with you. It feels like I just don't see why what we did here on earth is so difficult to do. [/INST] Yeah. And especially when you get into the details of it, I used to think origin of life was very um it was this magical rare event. But then you read books like for example nick Lane, um the vital question uh life ascending etcetera. And he really gets in and he really makes you believe that this is not that rare </s><s>[INST] basic chemistry. [/INST] You have an active earth and you have your alkaline vents and you have lots of alkaline waters mixing whether the devotion and you have your proton gradients and you have the little poorest pockets of these alkaline vents that concentrate chemistry. And um basically as he steps through all of these little pieces, you start to understand that actually this is not that crazy, you could see this happen on other systems. Um and he really takes you from just a geology to primitive life. And he makes it feel like it's actually pretty plausible and also like the origin of life um didn't uh was actually fairly fast after formation of earth. Um I think if I were correctly just a few 100 million years or something like that after basically when it was possible life actually arose. And so that makes me feel like that is not the constraint that is not the limiting variable and that life should actually be fairly common. Um And then you know where the drop offs are is very um it's very interesting to think about. I currently think that there's no major drop offs basically. And so there should be quite a lot of life. And basically what where that brings me to then is the only way to reconcile the fact that we haven't found anyone and so on is that um we just can't we can't see them, we can't observe them. </s><s>[INST] Just a quick brief comment nick Lane and a lot of biologists I talked to, they really seem to think that the jump from bacteria to more complex organisms is the hardest jump. [/INST] Yeah. </s><s>[INST] Which I don't I get it they're much more knowledgeable than me about like the intricacies of biology but that seems like crazy because how much how many single cell organisms are there like and how much time you have. Surely it's not that difficult like in a billion years. It's not even that long of a time really. Just all these bacteria under constrained resources battling it out. I'm sure they can invent more complex. Like I don't understand. It's like how to move from a Hello World program to like like invent a function or something like that. I don't Yeah. So I don't Yeah. So I'm with you. I just feel like I don't see any if the origin of life, that would be my intuition, that's the hardest thing. But if that's not the hardest thing because it happened so quickly then it's gotta be everywhere. And yeah, maybe we're just too dumb to see it. [/INST] It's just we don't have really good mechanisms for seeing this life. I mean by what? How? So I'm not an expert just to pre face this, but just from aliens. And it </s><s>[INST] was I [/INST] want to meet an expert </s><s>[INST] on alien intelligence and how to communicate. [/INST] Very suspicious of our ability to to find these intelligences out there and to find these Earth like radio waves for example, are are terrible. They're power drops off as basically one over R square. So I remember reading that our current radio waves would not be the ones that we are broadcasting would not be uh measurable by our devices today. Only like was it like 1/10 of a light year away. Like not even basically tiny distance because you really need like a targeted transmission of massive power directed somewhere for this to be picked up on long, long distances. And so I just think that our ability to measure is um, is not amazing. I think there's probably other civilizations out there. And then the big question is, why don't they build one moment probes and why don't they interstellar travel across the entire galaxy? And my current answer is, it's probably interstellar travel is like really hard. Uh you have the interstellar medium if you want to move at close to the speed of light, you're going to be encountering bullets along the way because even like tiny hydrogen atoms and little particles of dust are basically have like massive kinetic energy at those speeds. And so basically you need some kind of shielding. You need, you have all the cosmic radiation. It's just like brutal out there. It's really hard. And so my thinking is maybe interstellar travel is just extremely hard years </s><s>[INST] to build hard. It feels like uh it feels like we're not a billion years away from doing [/INST] that. It just might be that it's very, you have to go very slowly potentially as an example through space, </s><s>[INST] right? As opposed to close to the speed of [/INST] light. So I'm suspicious basically of our ability to measure life and I'm suspicious of the ability to just permeates all of space in the galaxy or across Galaxies. And that's the only way that I can, I can currently see a way around it. </s><s>[INST] It's kind of mind blowing to think that there is trillions of intelligent alien civilizations out there kind of slowly traveling through space [/INST] to </s><s>[INST] meet each other and some of them meet, some of them go to war. Some of them collaborate [/INST] and or they're all just independent. They are just like little pockets. Well, </s><s>[INST] statistically if there's like, if it's, there's trillions of them, surely some of them, some of the pockets are close enough to get in close enough to see each other. And then once you see, once you see something that is definitely complex life, like if we see something, we're probably going to be severe, like intensely aggressively motivated to figure out what the hell that is and try to meet them. But what would be your first instinct to try to like at a generational level, meet them or defend against them or what would be your instinct as a president of the United States [/INST] and </s><s>[INST] the scientist? I don't know which hat you prefer in this question? Yeah, [/INST] I think the question it's really hard. Um, I will say like for example, for us, um, we have lots of primitive life forms on earth. Um, next to us, we have all kinds of ants and everything else and we share space with them and we are hesitant to impact on them and to, we are trying to protect them by default because they are amazing, interesting dynamical systems that took a long time to evolve and they are interesting and special and I don't know that you wanna um destroy that by default. And so um I like complex dynamical systems that took a lot of time to evolve. I think um I'd like to, I'd like to preserve it if I can afford to and I'd like to think that the same would be true about the galactic resources and that uh they would think that we're kind of incredible interesting story that took time. It took a few billion years to unravel and you don't want to just destroy it. I </s><s>[INST] could see two aliens talking about Earth right now and saying I'm I'm a big fan of complex dynamical systems. So I think it was a value to preserve these and well basically our video game they watch or show a tv show that they watch. [/INST] Yeah I think you would need like a very good reason I think to to destroy it. Like why don't we destroy these and farms and so on. It's because we're not actually like really in direct competition with them right now we do it accidentally and so on. But um there's plenty of resources and so why would you destroy something that is so interesting and precious? </s><s>[INST] Well from a scientific perspective you might probe it, you might interact with it lately, [/INST] you might want to learn something from it, right? </s><s>[INST] So I wonder there could be certain physical phenomena that we think is a physical phenomenon but it's actually interacting with us to like poke the finger and see what happens. [/INST] I think it should be very interesting to scientists, other alien scientists. What happened here? Um, and you know, it's a, what we're seeing today is a snapshot basically. It's a, it's a result of a huge amount of computation uh, over a billion years or something like that. So </s><s>[INST] it could have been initiated by aliens. This could be a computer running a program like okay if you had the power to do this when you okay For sure. At least I would, I would pick earth like planet that has the conditions based my understanding of the chemistry prerequisites for life. And I would see that with life and run it right? Like wouldn't you 100% do that and observe it and then protect. I mean that that's not just a hell of a good TV show. It's, it's a good scientific experiment and that it's physical simulation. Right? Maybe [/INST] maybe </s><s>[INST] the evolution is the most like actually running it, uh, is the most efficient way to understand computation or to compute stuff [/INST] to understand life or you know what life looks like And what branches it can take. </s><s>[INST] It doesn't make me kind of feel weird that we're part of a science experiment, but maybe it's everything is a science experiments that does that change anything for us for a science experiment. [/INST] Um, I don't know, </s><s>[INST] two descendants of apes talking about being inside of the science. [/INST] I'm suspicious of this idea of like deliberate panspermia as you described it service. And I don't see a divine intervention in some way in the in the historical record right now. I do feel like um the story in these in these books like Nicholas books and so it sort of makes sense. Uh and it makes sense how life arose on earth uniquely. And uh yeah I don't need uh I need I don't need to reach for more exotic explanations right now. </s><s>[INST] Sure. But N. P. C. S inside a video game. Don't [/INST] don't don't </s><s>[INST] observe any divine intervention either. We might just be all in Pcs or any kind of code. Maybe [/INST] actually they will. Currently in pCS are really dumb but once they're running G. P. T. S um maybe they will be like hey this is really suspicious with the hell. </s><s>[INST] So you famously tweeted, it looks like if you bombard earth with photons for a while you can admit a Roadster. So if like in Hitchhiker's Guide to the Galaxy, we would summarize the story of Earth. So in in that book it's mostly harmless. Uh what do you think all the possible stories like a paragraph long or sentence long that Earth could be summarized as once it's done it's computation. So like all the possible full [/INST] if </s><s>[INST] Earth is a book right? Uh Probably there has to be an ending. I mean there's going to be an end to Earth and it could end in all kinds of ways. It can end soon. You can end later. What do you think of the possible stories? [/INST] We'll definitely there seems to be, yeah, you're sort of </s><s>[INST] it's [/INST] pretty incredible that these self replicating systems will basically arise from the dynamics and then they perpetuate themselves and become more complex and eventually become conscious and build a society. And I kind of feel like in some sense it's kind of like a deterministic wave that, you know, that kind of just like happens on any, you know, any sufficiently well arranged system like Earth. And so I kind of feel like there's a certain sense of inevitability in it. Um and it's really beautiful </s><s>[INST] and it ends somehow. Right? So it's a it's a chemically a diverse environment where complex dynamical systems can evolve and become more more further and further complex. But then there's a certain um what is it? There's certain terminating conditions. [/INST] Yeah, I don't know what the terminating conditions are, but definitely there's a trend line of something and we're part of that story and like where does that, where does it go? So, you know, we're famously described often as a biological boot loader for ai and that's because humans, I mean, you know, we're an incredible biological system and we're capable of computation and uh you know, and love and so on. Um but we're extremely inefficient as well. Like we're talking to each other through audio. It's just kind of embarrassing honestly that we're manipulating like seven symbols serially. We're using vocal cords. It's all happening over like multiple seconds. It's just like kind of embarrassing when you step down to the frequencies at which computers operate or are able to cooperate on. And so basically it does seem like um synthetic intelligences are kind of like the next stage of development and um I don't know where it leads to like at some point I suspect the universe is some kind of a puzzle. And these synthetic a eyes will uncover that puzzle and solve it. And </s><s>[INST] then what happens after right? Like what? Because if you just like fast forward earth many billions of years it's like it's quiet and then it's like normal you see like city lights and stuff like that. And then what happens at like at the end like is it like a pro is it or is it like a calming is it explosion? Is it like [/INST] Earth? Like open like </s><s>[INST] a giant because you said emit roasters will start admitting like like a giant number of like satellites? [/INST] Yes. It's some kind of a crazy explosion and we're living we're like we're stepping through a explosion and we're like living day to day and it doesn't look like it but it's actually if you I saw a very cool animation of Earth and life on Earth and basically nothing happens for a long time. And then the last like two seconds like basically cities and everything and in the lower orbit just gets cluttered and just the whole thing happened in the last two seconds and you're like this is exploding statement explosion. </s><s>[INST] Yeah if you play it at normal speed is it'll just look like an explosion. [/INST] It's a firecracker. We're living in a firecracker </s><s>[INST] where it's going to start emitting all kinds of interesting things and then the explosion doesn't, it might actually look like a little explosion with with lights and fire and energy emitted all that kind of stuff. But when you look inside the details of the explosion there's actual complexity happening where there's like yeah human life or some kind of life [/INST] we hope it's another destructive firecracker. It's kind of like a constructive firecracker. Alright </s><s>[INST] so given that [/INST] I think is </s><s>[INST] hilarious, it [/INST] is a really interesting to think about like what the puzzle of the universe is that the creator of the universe. Give us a message. Like for example in the book contact um Carl Sagan, there's a message for humanity for any civilization in the digits In the expansion of Pi in base 11 eventually which is kind of interesting thought maybe maybe we're supposed to be giving a message to our creator, maybe we're supposed to somehow create some kind of a quantum mechanical system that alerts them to our intelligent presence here because if you think about it from their perspective, it's just say like quantum field theory massive like cellular automaton like thing. And like how do you even notice that we exist? You might not even be able to pick us up in that simulation. And so how do you how do you prove that you exist, that you're intelligent and that you're part of the universe? So </s><s>[INST] this is like a touring test for intelligence from earth. [/INST] Like the </s><s>[INST] creator is uh I mean maybe this is like trying to complete the next word in the sentence. This is a complicated way of that. Like Earth is just is basically sending a message back. [/INST] Yeah, the puzzle is basically like alerting the creator that we exist or maybe the puzzle is just to just break out of the system and just you know, stick it to the creator in some way basically like if you're playing a video game, you can um you can somehow find an exploit and find a way to execute on the host machine any arbitrary code. There's some uh for example, I believe someone got Mario game of Mario to play pong just by uh exploiting it and then um creating uh basically writing writing code and being able to execute arbitrary code in the game. And so maybe we should be, maybe that's the puzzle is that we should be um find a way to exploit it. So so I think like some of these synthetic ads will eventually find the universe to be some kind of a puzzle and then solve it in some way. And that's kind of like the endgame somehow. </s><s>[INST] Do you often think about it as um as a simulation? So as the universe being kind of computation that has might have bugs and exploits? [/INST] Yes. Yeah, I think so. Physics </s><s>[INST] is essentially, [/INST] I think it's possible that physics has exploits and we should be trying to find them arranging some kind of a crazy quantum mechanical system that somehow gives you buffer overflow somehow gives you a rounding error in the floating point. Uh </s><s>[INST] Yeah, that's right. And like more and more sophisticated exploits. Those are jokes, but that could be actually [/INST] very close. We'll find some way to extract infinite energy. For example, when you train reinforcement learning agents um in physical simulations and you ask them to say run quickly on the flat ground, they'll end up doing all kinds of weird things. Um in part of that optimization, right? They'll get on their back legs and they will slide across the floor and it's because the optimization, um the enforcement learning optimization on that agent has figured out a way to extract infinite energy from the friction forces and um basically they're poor implementation and they found a way to generate infinite energy and just slide across the surface and it's not what you expected. It's just a it's sort of like a perverse solution. And so maybe we can find something like that. Maybe we can be that little dog in this physical simulation, </s><s>[INST] the cracks or escapes the intended consequences of the physics that the universe came up with. We'll figure out some kind of shortcut to some weirdness and then but see the problem with that weirdness is the first person to discover the weirdness, like sliding in the back legs, that's all we're gonna do. [/INST] Yeah, </s><s>[INST] it's very quickly because everybody does that thing. So like the paperclip maximize er is a ridiculous idea, but that very well could be what then we'll just we'll just all switch that because it's so fun. [/INST] Well no person will discover it. I think by the way I think it's going to have to be uh some kind of a super intelligent A. G. I. Of a third generation, like we're building the first generation A. G. I. You know, </s><s>[INST] third generation. Yeah, so the bootloader for an ai that ai will be a bootloader for another [/INST] guy </s><s>[INST] and then there's no way for us to introspect. Like what might even, I [/INST] think it's very likely that these things for example, like so you have these G. I. S. It's very likely, for example, they will be completely inert. I like these kinds of sci fi books sometimes where these things are just completely inert. They don't interact with anything and I find that kind of beautiful because they probably they probably figured out the meta meta game of the universe in some way, potentially they're they're doing something completely beyond our imagination. Um and they don't interact with simple chemical life forms, like why would you do that? So I find those kinds of ideas compelling. </s><s>[INST] What's their source of fun? What are they, what are they doing [/INST] solving in the universe? </s><s>[INST] But inert. So can you define what it means inert? So they escape the interaction, [/INST] physical realities as in um uh They will behave in some very like strange way to us because they're they're beyond their playing the meta game. And the meta game is probably say like arranging quantum mechanical systems in some very weird ways to extract infinite energy uh solve the digital expansion of pi to whatever amount they will build their own like little fusion reactors or something crazy. Like they're doing something beyond comprehension and not understandable to us and actually brilliant under the hood. </s><s>[INST] What if quantum mechanics itself is the system? And we're just thinking it's physics but we're really parasites on not parasite. We're not really hurting physics. We're just living on this organism and this organism and we're like trying to understand it. But really it is an organism. And with a deep deep intelligence maybe physics itself is uh the the organism that's doing the super interesting thing and we're just like one little thing. [/INST] Yeah </s><s>[INST] and sitting on top of it trying to get energy from it. [/INST] We're just kind of like these particles in the wave that I feel like it's mostly deterministic and takes universe from some kind of a big bang to some kind of a super intelligent replicator, some kind of a stable point in the universe given these laws of physics, you </s><s>[INST] don't think uh as Einstein said, God doesn't play dice, so you think it's mostly deterministic, there's no randomness and the thing [/INST] I think is deterministic, there's tons of, well, I want to be careful with randomness, </s><s>[INST] pseudo random. [/INST] Yeah, I don't like random. I think maybe the laws of physics are deterministic. Um Yeah, I think they're determined </s><s>[INST] just got really uncomfortable with this question. I do you have anxiety about whether the universe is random or not? What's there's [/INST] no randomness </s><s>[INST] you like? Good will hunting, It's not your fault andrei [/INST] it's </s><s>[INST] your fault, man. Um So you you don't like randomness uh [/INST] yeah, I think it's unsettling, I think it's a deterministic system. I think that things that look random, like say the collapse of the wave function, et cetera. I think they're actually deterministic, just entanglement and so on and some kind of a multiverse theory something something. </s><s>[INST] Okay, so why does it feel like we have a free will, like if if I raised his hand, I chose to do this now. Um what that doesn't feel like a deterministic thing, it feels like I'm making a choice, [/INST] it feels like it </s><s>[INST] okay, so it's all feelings, it's just feelings. So when RL agent is making a choice is that um it's not really making a choice, the choice is already there. [/INST] Yeah. You're interpreting the choice and you're creating a narrative for for having made it. </s><s>[INST] Yeah. And now we're talking about the narrative, it's very matter. Looking back what is the most beautiful or surprising idea in deep learning or ai in general that you've come across, you've seen this field explode uh and grow in interesting ways. Just what what cool ideas like, like we made you sit back and go small, big or small. [/INST] Well, the one that I've been thinking about recently the most probably is the </s><s>[INST] the [/INST] transformer architecture. Um so basically uh neural networks have a lot of architectures that were trendy have come and gone for different sensory modalities, like for vision, audio, text, you would process them with different looking neural nuts. And recently we've seen these this convergence towards one architecture, the transformer and you can feed it video or you can feed it, you know, images or speech or text and it just gobbles it up and it's kind of like a Bit of a general purpose computer that is also trainable and very efficient to run in our hardware. And so this paper came out in 20 </s><s>[INST] 16, I [/INST] want to say </s><s>[INST] attention is all you need [/INST] attention is all you need, You </s><s>[INST] criticized the paper title in retrospect that it wasn't um it didn't foresee the bigness of the impact that it was going to [/INST] have. I'm not sure if the authors were aware of the impact that that paper would go on to have probably weren't, but I think they were aware of some of the motivations and design decisions behind the transformer and they chose not to, I think expand on it in that way in the paper. And so I think they had an idea that there was more um, than just the surface of just like, oh we're just doing translation and here's a better architecture. You're not just doing translation. This is like a really cool differentiable optimism, efficient computer that you've proposed and maybe they didn't have all of that foresight. But I think it's really interesting, </s><s>[INST] isn't it funny, Sorry to interrupt that that title is memorable that they went for such a profound idea. They went with the I don't think anyone used that kind of title before. Right, [/INST] attention is all you need. It's like a meme or something. </s><s>[INST] It's not funny that one. Like maybe [/INST] if </s><s>[INST] it was a more serious title didn't have the impact. [/INST] Honestly, I yeah, there is an element of me that honestly agrees with you and prefers it this way. </s><s>[INST] If [/INST] it was two grand it would over promise and under deliver potentially. So you want to just mean your way to greatness </s><s>[INST] that should be a t shirt. So you, you tweeted the transformers, a magnificent neural network architecture because it is a general purpose differential computer. It is simultaneously expressive in the forward pass optimize Herbal via back propagation and gradient descent and efficient. High parallelism compute graph. Can you discuss some of those details, expressive. Optimize herbal, efficient, yeah. From memory or or in general whatever comes to your [/INST] heart. You want to have a general purpose computer that you can train on arbitrary problems like say the task of next word prediction or detecting if there's a cat in the image or something like that and you want to train this computer so you want to set its its weight. And I think there's a number of design criteria that sort of overlap in the transformer simultaneously. That made it very successful. And I think the authors were kind of deliberately trying to make this really powerful architecture and so basically it's very powerful in the forward past because it's able to express um very general computation as sort of something that looks like message passing, You have notes and they all store vectors and these nodes get to basically look at each other and it's each other's vectors and they get to communicate and basically notes, get to broadcast. Hey, I'm looking for sir, certain things and then other nodes get to broadcast. Hey, these are the things I have those are the keys in the values. </s><s>[INST] So it's not just attention. [/INST] Exactly, transformer is much more than just the attention component that's got many pieces architectural that went into it. The residual connection of the way it's arranged, there's a multi layer perceptron and there the way it's stacked and so on. Um but basically there's a message passing scheme where nodes get to look at each other, decide what's interesting and then update each other. And so I think the, when you get to the details of it, I think it's a very expressive function so it can express lots of different types of algorithms and forward pass. Not only that, but the way it's designed with the residual connections. Later normalization is the softmax attention and everything. It's also optimize double this is a really big deal because there's lots of computers that are powerful that you can't optimize or they are not easy to optimize using the techniques that we have, which is back propagation ingredients sent. These are first order methods, very simple optimizers really. And so um you also needed to be optimized. Herbal. Um and then lastly, you wanted to run efficiently in our hardware. Our hardware is a massive throughput machine, like gps, they prefer lots of parallelism. So you don't want to do lots of sequential operations. You want to do a lot of operations serially. And the transformer is designed with that in mind as well. And so it's designed for our hardware and is designed to both be very expressive in the forward pass but also very optimized able in the backward pass. </s><s>[INST] And you said that uh the residual connections support a kind of ability to learn short algorithms fast and first and then gradually extend uh longer during training. Yeah. What's what's the idea of learning short algorithms? [/INST] Right. Think of it as So basically a transformer is a series of blocks, right? And these blocks have attention and a little multi layer perception. And so you you go off into a block and you come back to this residual pathway and then you go off and you come back and then you have a number of layers arranged sequentially. And so the way to look at it I think is because of the residual pathway in the backward pass, the gradients sort of flow along it uninterrupted because addition distributes the gradient equally to all of its branches. So the gradient from the supervision at the top just floats directly to the first layer and the all the residual connections are arranged so that in the beginning of during initialization they contribute nothing to the residual pathway. Um so what it kind of looks like is imagine the transformer is kind of like a python function like a death. And um you get to do various kinds of like lines of code. Um so you have 100 layers deep transformer, typically they would be much shorter, say 20 so 20 lines of code and you can do something in them. And so think of during the optimization, basically what it looks like is first you optimize the first line of code and then the second line of code can kick in and the third line of code can kick in and I kind of feel like because of the residual pathway and the dynamics of the optimization, You can sort of learn a very short algorithm that gets the approximate answer, but then the other layers can sort of kick in and start to create a contribution and at the end of it you're you're optimizing over an algorithm that is 20 lines of code Except these lines of code are very complex because it's an entire block of a transformer. You can do a lot in there. What's really interesting is that this transformer architecture actually has been remarkably resilient, basically a transformer that came out in 2016 as the transformer you would use today except you reshuffle from the layer norms, the normalization have been reshuffled to a pre norm um formulation and so it's been remarkably stable, but there's a lot of bells and whistles that people have attached on it and try to improve it. I do think that basically it's a, it's a big step in simultaneously optimizing for lots of properties of a desirable neural network architecture and I think people have been trying to change it, but it's proven remarkably resilient. Um but I do think that there should be even better architecture is potentially, </s><s>[INST] but it's your, you admire the resilience here, there's something profound about this architecture that at least maybe we can everything can be turned into a uh into a problem that transformers can solve [/INST] currently definitely looks like the transformers taking over AI and you can feed basically arbitrary problems into it and it's a general differentiable computer and it's extremely powerful and uh this convergence in AI has been really interesting to watch uh for me personally </s><s>[INST] what else do you think could be discovered here about transformers? Like what surprising thing or is it a stable I was in a stable place, is there something interesting we might discover about transformers like aha moments maybe has to do with memory, maybe knowledge representation, that kind of stuff [/INST] definitely The zeitgeist today is just pushing like basically right now the guys is do not touch the transformer, touch everything else. So people are scaling up the data sets making them much much bigger, they're working on the evaluation, making the evaluation much much bigger and uh um they're basically keeping the architecture unchanged and that's how we've that's the last five years of progress in ai kind of </s><s>[INST] what do you think about one flavor of it? Which is language models? Have you been surprised uh has your sort of imagination been captivated by, you mentioned D. P. T. And all the bigger and bigger and bigger language models and uh what are the limits of those models do you think? So just for the task of natural language, [/INST] basically the way GPT is trained, right, is you just down a massive amount of text data from the internet and you try to predict the next word in the sequence, roughly speaking, you're predicting a little word chunks but roughly speaking, that's it. Um and what's been really interesting to watch is Basically it's a language model. Language models have actually existed for a very long time. Um there's papers on language modeling from 2003 even earlier. Can </s><s>[INST] you explain that case what language model is? [/INST] Yeah, so language model, just basically the rough idea is um just predicting the next word in a sequence, roughly speaking. Uh so there's a paper from for example, Benji Oh and the team from 2003 where for the first time they were using a neural network to take say like three or five words and predict the next word. And they're doing this on much smaller data sets. And the neural net is not a transformer, it's a multi layer perceptron but it's the first time that a neural network has been applied in that setting. But even before neural networks that were um, language models except they were using um N gram models. So N gram models are just count based models. So um if you try if you try to take two words and predict the third one, you just count up how many times you've seen any two word combinations and what came next and what you predict is coming next is just what you've seen the most of in the training set. And so language modeling has been around for a long time, neural networks have done language modeling for a long time. So really what's new or interesting or exciting is just realizing that when you scale it up with powerful enough neural net transformer, you have all these emergent properties. Where basically what happens is if you have a large enough dataset of text, you are in the task of predicting the next word, you are multitasking a huge amount of different kinds of problems you are multitasking understanding of, you know, chemistry physics, human nature, lots of things are sort of clustered in that objective. It's a very simple objective, but actually you have to understand a lot about the world to make that prediction. </s><s>[INST] You just said the u word understanding. Are you in terms of chemistry and physics and so on. What do you feel like it's doing is it's searching for the right context uh in in what, what is it, what is the actual process happening here? Yes. [/INST] So basically it gets 1000 words and trying to predict 1001st and in order to do that very, very well over the entire dataset available on the internet, you actually have to basically kind of understand the context of what's going on in there. Um and uh it's a sufficiently hard problem that you uh if you have a powerful enough computer, like a transformer, you end up with interesting solutions and uh you can ask it to do all kinds of things and it shows a lot of emergent properties like in context learning. That was the big deal with GPT and the original paper when they published it is that you can just sort of uh prompt it in various ways and ask it to do various things and it will just kind of complete the sentence. But in the process of just completing the sentence it's actually solving all kinds of really interesting problems that we care about. Do </s><s>[INST] you think it's doing something like understanding like when we use the word understanding for us humans? [/INST] I think it's doing some understanding it in its weights, it understands I think a lot about the world and it has to in order to predict the next word in a sequence. </s><s>[INST] So it's trained on the data from the internet. What do you think about this this approach in terms of data sets of using data from the internet. Do you think the internet has enough structured data to teach ai about human civilization? [/INST] Yes, I think the internet is a huge amount of data. I'm not sure if it's a complete enough set. I don't know that text is enough for having a sufficiently powerful A. G I has an outcome. Um </s><s>[INST] Of course there is audio and video and images and all that kind of stuff. [/INST] Yeah. So, text by itself. I'm a little bit suspicious about. There's a ton of things we don't put in text in writing just because they're obvious to us about how the world works in the physics of it and that things fall, we don't put that stuff in text because why would you we share that understanding? And so taxes communication medium between humans and it's not a all encompassing medium of knowledge about the world. But as you pointed out, we do have video and we have images and we have audio. And so I think that that definitely helps a lot, but we haven't trained models sufficiently across both across all these modalities yet. So I think that's what a lot of people are interested in, </s><s>[INST] but I wonder what that shared understanding of, like, what we might call common sense has to be learned inferred in order to complete the sentence correctly. So maybe the fact that it's implied on the internet, the model's gonna have to learn that not by reading about it by inferring in the representation. So like, common sense, just like we I don't think we learned common sense, like nobody says tells us explicitly, we just figure it all out by interacting with the [/INST] world. </s><s>[INST] And so here's a model of reading about the way people interact with the world and might have to infer that I wonder [/INST] Yeah, </s><s>[INST] you you briefly worked on a project called the world of bits, training in RL system to take actions on the internet um versus just consuming the internet, like we talked about, do you think there's a future for that kind of system interacting with the internet to help the learning? [/INST] Yes, I think that's probably the final frontier for a lot of these models because um so as you mentioned, there was an opening, I was working on this project world of bits and basically it was the idea of giving neural networks access to a keyboard and a mouse and the idea </s><s>[INST] possibly go wrong. [/INST] So basically you you perceive the input of the uh screen pixels and basically the state of the computer is sort of visualized for human consumption in images of the web browser and stuff like that. And then you give them the ability to press keyboards and use the mouse and we're trying to get it to for example, complete bookings and you know, interact with the user interfaces and what </s><s>[INST] did you learn from that experience? Like what was some fun stuff? There's a super cool idea. I mean, it's like yeah, I mean the step between observer to actor is a super fascinating [/INST] step. Well, the universal interface in the digital realm, I would say. And there's a universal interface in like the physical realm which in my mind is a humanoid form factor kind of thing. We can later talk about optimus and so on, but I feel like there's a they're kind of like a similar philosophy in some way where the human the world, the physical world is designed for the human form and the digital world is designed for the human form of seeing the screen and using keyword, keyboard and mouse. And so is the universal interface that can basically command the digital infrastructure we've built up for ourselves. And so it feels like a very powerful interface to command and to build on top of now to your question as to like what I learned from that. It's interesting because the world of bits was basically too early I think at open ai at the time um this is around 2015 or so, and the zeitgeist at that time was very different in ai from the zeitgeist today at the time everyone was super excited about reinforcement learning from scratch. This is the time of the Atari paper where neural networks were playing Atari games um and beating humans in some cases uh Alphago and so on. So everyone is very excited about training neural networks from scratch using reinforcement learning. Um directly it turns out that reinforcement learning is extremely inefficient way of training neural networks because you're taking all these actions and all these observations and you get some sparse rewards once in a while. So you do all this stuff based on all these inputs and once in a while you're like told you did a good thing, you did a bad thing. It's just an extremely hard problem. You can't learn from that, You can burn the forest and you can sort of brute force through it. And we saw that I think with, you know, with Go and Dota and so on and does work, but it's extremely inefficient uh and not how you want to approach problems, practically speaking. And so that's the approach that at the time we also took the world of bits, we would have an agent initialized randomly. So with keyboard mash and mouse mash and try to make a booking. And it's just like revealed the insanity of that approach very quickly where you have to stumble by the correct booking in order to get a reward of you did it correctly and you're never gonna stumble by it by chance at random. So </s><s>[INST] even with a simple web interface, there's too many options, [/INST] there's just too many options, uh and uh it's too sparse of reward signal and you're starting from scratch at the time. And so you don't know how to read, you don't understand pictures, images, buttons, you don't understand what it means to like make a booking, but now what's happened is uh it is time to revisit that and opening, I was interested in this. Uh, Companies like Adept are interested in this and so on and uh the idea is coming back because the interface is very powerful, but now you're not training an agent from scratch, you are taking the GPT as initialization. So GPT is pre trained on all of text and it understands what's a booking, it understands what to submit, it understands um quite a bit more and so it already has those representations, they are very powerful and that makes all the training significantly more efficient um and makes the problem tractable. </s><s>[INST] Should the interaction be with like the way humans see it with the buttons and the language or should be with the html javascript and CSS what's what do you think is the better? So [/INST] today, all of this interaction is mostly on the level of html CSS and so on. That's done because of computational constraints. Uh but I think ultimately, um everything is designed for human visual consumption and so at the end of the day there's all the additional information is in the layout of the web page and what's next to you and what's a red background and all this kind of stuff and what it looks like visually. So I think that's the final frontier is we're taking in pixels and we're giving out keyboard mouse commands, but I think it's impractical Still today, </s><s>[INST] do you worry about bots on the internet, Given, given these ideas, given how exciting they are. Do you worry about bots on twitter being not the stupid bots that we see now with the crypto bots, but the bots that might be out there actually that we don't see that they're interacting in interesting ways. So this kind of system feels like it should be able to pass the I'm not a robot click button whatever um which you actually understand how the test works. I don't quite like there's there's a there's a check box or whatever that you click, it's presumably tracking like mouse movement and the timing and so on. So exactly this kind of system we're talking about should be able to pass that. So yeah, what do you feel about? Um bots that are language models plus have some interact ability and are able to tweet and reply and so on. Do you worry about that world? [/INST] Yeah, I think it's always been a bit of an arms race between sort of the attack and the defense. So the attack will get stronger but the defense will get stronger as well, our ability to detect that. How </s><s>[INST] do you defend? How do you detect? How do you know that your account on twitter is is human? [/INST] How </s><s>[INST] did you approach that? Like if people will claim you know uh how would you defend yourself in the court of law that I am a human? [/INST] Um </s><s>[INST] this account at some [/INST] point I think uh it might be I think the society society will evolve a little bit like we might start signing digitally signing some of our correspondence or you know things that we create right now. It's not necessary but maybe in the future it might be I do think that we are going towards the world where we share we share the digital space with a eyes, </s><s>[INST] synthetic beings. [/INST] Yeah. And they will get much better and they will share our digital realm and they'll eventually share our physical realm as well. It's much harder. But that's kind of like the world we're going towards and most of them will be benign and hopeful and some of them will be malicious and it's going to be an arms race trying to detect them. </s><s>[INST] So I mean the worst isn't the ai is the worst is the Ai is pretending to be human. So I don't know if it's always malicious. There's obviously a lot of malicious applications but it could also be, you know, if I was an ai I would try very hard to pretend to be human because we're in a human world. I wouldn't get any respect as an A I want to get some love and respect. [/INST] I don't think the problem is intractable. People are people are thinking about the proof of personhood and uh we might start digitally signing our stuff and we might all end up having like uh yeah, basically some some solution for proof of personhood. It doesn't seem to be intractable. It's just something that we haven't had to do until now. But I think once the need like really starts to emerge which is soon. I think people think about it much more </s><s>[INST] so, but that too will be a race because obviously you can probably uh spoof or fake the proof of personhood. So you have to try to figure out how to [/INST] probably. I </s><s>[INST] mean it's weird that we have like social security numbers and like passports and stuff. It seems like it's harder to fake stuff in the physical space, in the visual space. It just feels like it's gonna be very [/INST] tricky, very </s><s>[INST] tricky to out because it seems to be pretty low cost fake stuff. What are you gonna put an ai in jail for like trying to use a fake fake personhood proof? I mean, okay fine, you put a lot of guys in jail, but there will be more as our like exponentially more, the cost of creating bot is very low. Unless there's some kind of way to track accurately. Like you're not allowed to create any program without showing uh tying yourself to that program. Like any program that runs on the internet, you'll be able to uh trace every single human program that was involved with that [/INST] program. Yeah. Maybe you have to start declaring when uh you know, we have to start drawing those boundaries and keeping track of okay, what are digital entities versus human entities and uh what is the ownership of human entities and digital entities and uh something like that. Um I don't know, but I'm I think I'm optimistic that this is uh this is uh possible and in some, in some sense, we're currently in like the worst time of it because um all these bots suddenly have become very capable, but we don't have defenses yet built up as a society and but I think that doesn't seem to be intractable. It's just something that we have to deal with. </s><s>[INST] It seems weird that the twitter, but like really crappy twitter bots are so numerous. Is it? So I presume that the engineers of twitter are very good. So it seems like what I would infer from that uh is it seems like a hard problem if they're probably catching all right. If I were to sort of steel man the case. It's a hard problem and there's a huge cost to uh false positive to removing a post by somebody that's not about, that creates a very bad user experience. So they're very cautious about removing. So maybe it's um, and maybe the boss are really good at learning what gets removed and not such that they can stay ahead of the removal process very quickly. My [/INST] impression of it honestly is there's a lot of longing for. I mean, </s><s>[INST] yeah, just that's what I [/INST] it's not subtle my impression of it. It's not so, but </s><s>[INST] you have, yeah, that's my impression as well, but it feels like maybe you're seeing the tip of the iceberg, maybe the number of boxes and like the trillions and you have to like just, it's a constant assault of bots and yeah I don't know um you have to steal in the case because the bots, I'm seeing a pretty like obvious I could write a few lines of code to catch these bots. [/INST] I mean definitely there's a lot of low hanging fruit but I will say I agree that if you are a sophisticated actor you could probably create a pretty good but right now um you know using tools like G. P. T. S because it's a language model, you can generate faces that look quite good now uh and you can do this at scale and so I think um it's quite possible it's going to be hard to defend. </s><s>[INST] There was a google engineer that claim that the lambda was essentially, do you think there's any inkling of truth to what he felt? And more importantly to me at least, do you think language models will achieve sentience or the illusion of sentience soon? Ish. Yeah. [/INST] To me it's a little bit of a canary in a coal mine kind of moment. Honestly a little bit because uh so this engineer spoke to like a chatbot at google and uh became convinced that </s><s>[INST] asked some existential philosophical question [/INST] and it gave like reasonable answers and looked real and uh and so on so to me it's a he was he was he wasn't sufficiently trying to stress the system I think and exposing the truth of it as it is today. Um But I think this will be increasingly harder over time. So uh yeah, I think more and more people will basically become um yeah, I think more and more there will be more people like that over time as as this gets better, like </s><s>[INST] form an emotional connection to an ai perfectly [/INST] plausible in my mind. I think these guys are actually quite good at human human connection. Human emotion. A ton of text on the internet is about humans and connection and love and so on. So I think they have a very good understanding in some in some sense of how people speak to each other about this and they're very capable of creating a lot of that kind of text. The there's a lot of like sci fi from fifties and sixties that imagined Ai is in a very different way. They are calculating cold Vulcan like machines, that's not what we're getting today. We're getting pretty emotional ai that actually are very competent and capable of generating, you know, possible sounding text with respect to all of these topics. </s><s>[INST] See I'm really hopeful about ai systems that are like companions that help you grow, develop as a human being, help you maximize long term happiness. But I'm also very worried about ai systems that figure out from the internet, the humans get attracted to drama and so this would just be like shit talking eyes, they just constantly did you hear, like they'll do gossip, they'll do they'll try to plant seeds of suspicion to other humans that you love and trust and just kind of mess with people. Uh you know, because because that's going to get a lot of attention to drama maximize drama on the path to maximizing engagement. And us humans will feed into that machine and get it will be a giant drama shit storm. Uh so I'm worried about that. So it's the objective function really defines the way that human civilization progresses with the eyes in [/INST] it. I think right now at least today they're not sort of it's not correct to really think of them as goal seeking agents that want to do something. They have no long term memory or anything. They it's literally a good approximation of it is you get 1000 words and you're trying to predict 1000 and first and then you continue feeding it in and you are free to prompt it in whatever way you want. So in text, so you say, okay you are a psychologist and you are very good and you love humans and here's a conversation between you and another human human column something, you something and then it just continues the pattern and suddenly you're having a conversation with the fake psychologist who's trying to help you. And so it's still kind of like an aroma of a tool is a people can prompt in arbitrary ways and it can create really incredible text but it doesn't have long term goals over a long period of time, it doesn't try to uh so it doesn't look that way right now </s><s>[INST] but you can do short term goals that have long term effects. So if my prompting, short term goal is to get Andrew capacity to respond to me on twitter when I like I think a I might that's the goal but it might figure out that talking shipped to you. It would be the best in a highly sophisticated interesting way and then you build up a relationship when you were responding once and then it like over time it gets to not be sophisticated and just like just talk shit and [/INST] okay </s><s>[INST] maybe you won't get to andre but it might get to another celebrity and might get into other big accounts and then it'll just so with just that simple goal get them to respond to maximize the probability of actual response. [/INST] Yeah, I mean you could prompt a powerful model like this with their its opinion about how to do any possible thing you interested. So they will just, they're kind of on track to become these oracles, I could sort of think of it that way they are. Oracles currently it's just text but they will have calculators, they will have access to google search, they will have all kinds of gadgets and gizmos, they will be able to operate the internet and find different information and yeah, in some sense, that's kinda like currently what it looks like in terms of the development, </s><s>[INST] do you think it will be an improvement eventually over what google is for access to human knowledge, Like it'll be a more effective search engine to access human knowledge. [/INST] I think there's definite scope in building a better search engine today and I think google, they have all the tools, all the people, they have, everything they need, all the puzzle pieces, they have people training transformers at scale, they have all the data. Uh it's just not obvious if they are capable as an organization to innovate on their search engine right now and if they don't someone else will, there's absolute scope for building a significantly better search engine built on these tools. </s><s>[INST] It's so interesting. A large company where the search, there's already an infrastructure, it works as brings out a lot of money, so where structurally inside the company is their motivation to pivot to say we're going to build a new search engine. [/INST] Yeah, that's hard. </s><s>[INST] So it's usually going to come from a startup, [/INST] right? That's um, that would be yeah, or some other more competent organization. Um so, uh, I don't know, so currently, for example, maybe being has another shot at it, you know, </s><s>[INST] Go Microsoft, [/INST] we're talking </s><s>[INST] offline um I [/INST] mean, I definitely it's really interesting because search engines used to be about, okay, here's some query, here's, here's here's web pages that look like the stuff that you have but you could just directly go to answer and then have supporting evidence. Um and these uh these models basically they've read all the text and they read all the web pages and so sometimes when you see yourself going over to search results and sort of getting like a sense of like the average answer to whatever you're interested in like that just directly comes out, you don't have to do that work. Um So they're kind of like uh yeah, I think they have a way to this of distilling all that knowledge </s><s>[INST] into like [/INST] some level of insight basically. </s><s>[INST] Do you think of prompting as a kind of teaching and learning like this whole process like another layer, you know, because maybe that's what humans are already have that background model and you're the world is prompting you? [/INST] Yeah, exactly. I think the way we are programming these computers now, like GPS is converging to how you program humans. I mean how do I program humans via prompt? I go to people and I prompt them to do things. I prompt them from information. And so natural language prompt is how we program humans and we're starting to program computers directly in that interface, it's like pretty remarkable honestly. </s><s>[INST] So you've spoken a lot about the idea of software 2.0 um all good ideas become like cliches so quickly, like the terms it's kind of hilarious um it's like I think Eminem once said that like if he gets annoyed by song, he's written very quickly, that means it's gonna be a big hit because it's too catchy. But can you describe this idea and how you're thinking about it has evolved over the months and years since, since you coined it? [/INST] Yeah. Yes. I had a block post on software 2.0, I think several years ago now. Um and the reason I wrote that post is because I kept, I kind of saw something remarkable happening in like software development and how a lot of code was being transitioned to be written, not in sort of like C plus plus and so on, but it's written in the weights of the neural net. Basically just saying that neural nets are taking over software, the realm of software and um taking more more more tasks. And at the time I think not many people understood this deeply enough that this is a big deal. It's a big transition uh neural networks were seen as one of multiple classification algorithms you might use for your dataset problem on cattle. Like this is not that this is a change in how we program computers and I saw neural nuts as uh this is going to take over the way we program computers is going to change is not going to be people writing software in C plus plus or something like that and directly programming the software, it's going to be accumulating training sets and datasets and crafting these objectives by which we train these neural nets and at some point there's going to be a compilation process from the data sets and the objective and the architecture specification into the binary, which is really just uh the neural nut, you know, weights and the forward path of the neural nut. And then you can deploy that binary. And so I was talking about that sort of transition and uh that's what the post is about. And I saw this sort of play out in a lot of fields, uh you know, being one of them, but also just a simple image classification. People thought originally, you know, in the 80s and so on that they would write the algorithm for detecting a dog in an image. And they had all these ideas about how the brain does it. And first we detect corners and then we detect lines and then we stitch them up and they were like really going at it. They were like thinking about how they're gonna write the algorithm and this is not the way you build it. Um there was a smooth transition where okay first we thought we were gonna build everything. Then we were building the features. So like hog features and things like that that detect these little statistical patterns from image patches and then there was a little bit of learning on top of it like support vector machine or binary classifier for cat versus dog and images on top of the features. So we wrote the features but we trained the last layer sort of the classifier. And then people are like actually let's not even design the features because we can't honestly we're not very good at it. So let's also learn the features. And then you end up with basically a convolutional neural net where you're learning most of it, you're just specifying the architecture and the architecture has tons of fill in the blanks which is all the knobs and you let the optimization right? Most of it. And so this transition is happening across the industry everywhere. And uh suddenly we end up with a ton of code that is written in neural net weights. And I was just pointing out that the analogy is actually pretty strong. And we have a lot of developer environments for software like we have I. D. S. Um how you work with code? How you debug code. How do you how do you run code? How do you maintain code? We have Git hub. So I was trying to make those analogies in the new realm like what is the Git hub? Software 2.0, turns out it's something that looks like hugging face right now. </s><s>[INST] Uh, [/INST] you know, And so I think some people took it seriously and build cool companies and uh many people originally attacked the post. It actually was not well received when I wrote it and I think maybe it has something to do with the title but the post was not well received and I think more people sort of have been coming around to it over time. </s><s>[INST] Yeah so you were the director of Ai at Tesla where I think this idea was really implemented a scale which is how you have engineering teams doing software two point Oh so can you sort of linger on that idea of I think we're in the really early stages of everything you just said which is like get hub I. D. S. Like how how do we build engineering teams that that work in software 2.0. System and and the the data collection and the data annotation which is All part of that software 2.0. Like what do you think is the task of programming software? 2.0 is it debugging in the space of hyper parameters or is it also debugging the space of data? [/INST] Yeah. The way by which you program the computer and influence its algorithm is not by writing the commands yourself. You're changing mostly the dataset. You're changing the loss functions of like what the neural net is trying to do, how it's trying to predict things but they're basically the data sets and the architecture of the neural net and um so in the case of the autopilot, a lot of the data sets had to do with for example detection of objects and lane lane markings and traffic lights and so on. So accumulate massive data sets of, here's an example, here's the desired label and then uh, here's roughly how the architect, here's roughly what the algorithm should look like. And that's a convolutional neural net. So the specification of the architecture is like a hint as to what the algorithm should roughly look like. And then the fill in the blanks process of optimization is the training process and then you take your neural net that was trained. It gives all the right answers on your dataset and you deploy it. </s><s>[INST] So there's in that case perhaps at all machine learning cases, there's a lot of tasks. So is coming up formulating a task like uh, for a multi headed neural network, is formulating a task part of the programming, [/INST] how </s><s>[INST] you break down a problem into a set of tasks. [/INST] I'm on the high level, I would say if you look at the software running in in the autopilot, I gave a number of talks on this topic, I would say originally a lot of it was written in software one point No, there's imagine lots of c plus plus, right. And then gradually there was a tiny neural net that was for example, predicting given a single image. Is there like a traffic light or not? Or is there a landline marking or not. And this neural net didn't have too much to do in the in the scope of the software was making tiny predictions on individual little image and then the rest of the system stitched it up. So okay, we're actually we don't have just a single camera with eight cameras. We actually have eight cameras over time. And so what do you do with these predictions? How do you put them together? How do you do the fusion of all that information and how do you act on it? All of that was written by humans um in C plus plus. And then we decided, okay, we don't actually want uh to do all of that fusion in C. Plus plus code because we're actually not good enough to write that algorithm. We want the neural nets to write the algorithm and we want to support all of that software into the stack. And so then we actually have neural nets that now take all the eight camera images simultaneously and make predictions for all of that. So um and actually they don't make predictions in the in the space of images. They now make predictions directly in three D. And actually they don't in three dimensions around the car. And now actually we don't um manually fuse the predictions over in three D. Over time. We don't trust ourselves to write that tracker. So actually we give the neural net uh the information over time. So it takes these videos now and makes these predictions. And so you're sort of just like putting more and more power into the neural net neural processing and at the end of it, the eventual sort of goal is to have most of the software potentially be in the and um because it works significantly better, humans are just not very good at writing software basically. So </s><s>[INST] The prediction is happening in this like 4D land with three dimensional world over time. How do you do annotation in that world? What what what have you as the data annotation whether it's self supervised or manual by humans is um is a big part of the world, [/INST] right? I would say by far in the industry if you're like talking about the industry and how what is the technology of what we have available, Everything is supervised learning. So you need a data set of input, desired output and you need lots of it. And um there are three properties of it that you need, you need to be very large, you needed to be accurate, no mistakes and you needed to be diverse. You don't want to uh just have a lot of correct examples of. One thing you need to really cover the space of possibility as much as you can and the more you can cover the space of possible inputs, the better the algorithm will work at the end. Now, once you have really good data sets that you're collecting curating, um and cleaning you can train your neural net um on top of that. So a lot of the work goes into cleaning those datasets now as you pointed out it's probably it could be. The question is how do you achieve a ton of uh if you want to basically predict in three D. You need data and three D. To back that up. So in this video we have eight videos coming from all the cameras of the system and this is what they saw. And this is the truth of what actually was around there was this car, there was this car this car, these are the main line markings, this is the geometry of the road, there's traffic light in this three dimensional position. You need the ground truth. Um And so the big question that team was solving of course is how do you how do you arrive at that ground truth? Because once you have a million of it and it's large, clean and diverse. Then training neural net on it works extremely well and you can ship that into the car. And uh so there's many mechanisms by which we collected that training data. You can always go for human annotation. You can go for a simulation as a source of ground truth. You can also go for what we call the offline tracker um </s><s>[INST] that [/INST] we've spoken about at the A. I. D. And so on. Which is basically an automatic reconstruction process for taking those videos and recovering the three dimensional sort of reality of what was around that car. So basically think of doing like a three dimensional reconstruction as an offline thing and then understanding that okay there's 10 seconds of video. This is what we saw and therefore here's all the landlines cars and so on. And then once you have that annotation you can train a neural net to imitate it. </s><s>[INST] And how difficult is the reconstruct the three reconstruction? It's [/INST] difficult but it can be done. </s><s>[INST] So there's so there's overlap between the cameras and you do the reconstruction and there's uh perhaps if there's any inaccuracy. So that's caught in the annotation step. [/INST] Yes. The nice thing about the annotation is that it is fully offline. You have infinite time, you have a chunk of one minute and you're trying to just offline in a supercomputer somewhere figure out where were the positions of all the cars, all the people and you have your full one minute video from all the angles and you can run all the neural nets you want. And they can be very efficient massive neural nuts. There can be neural nets that can't even run in the car later at test time. So they can be even more powerful neural nets than what you can eventually deploy. So you can do anything you want three dimensional reconstruction neural nets anything you want just to recover that truth and then you supervise that truth. </s><s>[INST] What have you learned, You said no mistakes about humans doing annotation because I assume humans are uh there's like a range of things that are good at in terms of clicking stuff on screen. Is that how interesting is that you have a problem with designing an annotator where humans are accurate, enjoy it, like what are they, even the metrics are efficient or productive? All that kind of stuff. [/INST] Yeah. So I grew the annotation team at Tesla from basically 0 2000 uh while I was there, that was really interesting. You know, my background as a PhD student researcher. So growing that kind of organization was pretty crazy. Uh but uh yeah I think it's extremely interesting and part of the design process very much behind the autopilot as to where you use humans. Humans are very good at certain kinds of annotations. They're very good for example, at two dimensional annotations of images. They're not good at annotating uh cars over time in three dimensional space. Very, very hard. And so that's why we were very careful to design the tasks that are easy to do for humans versus things that should be left to the offline tracker. Like maybe the maybe the computer will do all the triangulation and three D reconstruction, but the human will say exactly these pixels of the image, our car, exactly these pixels are human. And so co designing the data annotation pipeline was very much bread and butter was what I was doing daily, </s><s>[INST] Do you think there's still a lot of open problems in that space? [/INST] Um </s><s>[INST] just in general annotation where the stuff the machines are good at machines do and the humans do what they're good at and there's maybe some iterative process, right? [/INST] I think to a very large extent we went through a number of iterations and we learned a ton about how to create these datasets. Um I'm not seeing big open problems like originally when I joined, I was like, I was really not sure how this will turn out, but by the time I left I was much more secure in actually we sort of understand the philosophy of how to create these data sets and I was pretty comfortable with where that was at the time. </s><s>[INST] So what our strengths and limitations of cameras for the driving task and your understanding when you formulate the driving task as a vision task with eight cameras. you've seen the entire, you know, most of the history of the computer vision field when it has to do with neural networks. What just if you step back, what are the strengths and limitations of pixels of using pixels to drive? [/INST] Yeah, pixels I think are a beautiful sensory, beautiful sensor. I would say the things like cameras are very, very cheap and they provide a ton of information, ton of bits. Also it's uh extremely cheap sensor for a ton of bits and each one of these bits as a constraint on the state of the world. And so you get lots of megapixel images, very cheap and it just gives you all these constraints for understanding what's actually out there in the world. So vision is probably the highest bandwidth sensor. It's very high bandwidth sensor. And um </s><s>[INST] I love that pixels is uh, is a constraint in the world, is highly complex, high bandwidth constraint in the world, on the state of the world. It's [/INST] not just that, but again, this real real importance of it's the sensor that humans use. Therefore everything is designed for that sensor </s><s>[INST] text, [/INST] the writing, the flashing signs, everything is designed for vision. And so you just find it everywhere. And so that's why that is the interface you want to be in. Um talking again about these universal interfaces and that's where we actually want to measure the world as well. And then develop software for that sensor. </s><s>[INST] But there's other constraints on the state of the world that humans use to understand the world. I mean, vision ultimately is the main one, but we were like, we're like referencing our understanding of human behavior and some common sense and physics that could be inferred from vision from, from a perception perspective, but it feels like we're using some kind of reasoning to predict the world, not just the pixels, [/INST] you have a powerful prior for how the world evolves over time, etcetera. So it's not just about the likelihood term coming up from the data itself, telling you about what you are observing but also the prior term of like where, where are the likely things to see and how do they likely move and so on? And </s><s>[INST] the question is how complex is the the the range of possibilities that might happen in the driving task? That still is that you still an open problem of how difficult is driving? Like philosophically speaking, do all [/INST] the time you worked on </s><s>[INST] driving. Do you understand how hard driving is [/INST] driving is really hard because it has to do with the predictions of all these other agents and the theory of mind and you know what they're gonna do and are they looking at you, are they, where are they looking, what are they thinking? Yeah, there's a lot that goes there at the, at the full tale of you know the expansion of the nine that we have to be comfortable with it. Eventually the final problems are of that form, I don't think those are the problems that are very common I think eventually they're important but it's like really in the tail end </s><s>[INST] in the tail end the rare edge cases from the vision perspective, what are the toughest parts of the vision problem of driving? [/INST] Um well basically the sensor is extremely powerful but you still need to process that information. Um, and so going from brightness is of the special values to hey, here's the three dimensional world is extremely hard and that's what the neural networks are fundamentally doing. And so, um, the difficulty really is in just doing an extremely good job of engineering the entire pipeline, the entire data engine, having the capacity to train these neural nuts, having the ability to evaluate the system and iterate on it. So I would say just doing this in production at scale is like the hard part, it's an execution problem. </s><s>[INST] So the data engine, but also the, um, the sort of deployment of the system such that has low latency performance. So it has to do all these steps [/INST] for general, not specifically just making sure everything fits into the chip on the car and you have a finite budget of flops that you can perform and uh, and memory bandwidth and other constraints. And you have to make sure it flies and you can squeeze in as much computer as you can into the tiny, </s><s>[INST] what have you learned from that process because maybe that's one of the bigger, like new things coming from a research background where there's, there's a system that has to run under heavily constrained resources has to run really fast. What, what kind of insights have you learned from that? [/INST] Yeah, I'm not sure if it's, if there's too many insights, you're trying to create a neural nets that will fit in what you have available and you're always trying to optimize it and we talked a lot about it on the ai day and uh basically the triple back flips that the team is doing to make sure it all fits and utilizes the engine. Uh, so I think it's extremely good engineering. Um, and then there's all kinds of little insights peppered in on how to do it properly. </s><s>[INST] Let's actually zoom out because I don't think we talked about the data engine, the entirety of the layout of this idea that I think is just beautiful with humans in the loop. Can you describe the data engine? [/INST] Yeah, the data engine is what I call the almost biological feeling like process by which you perfect the training sets for these neural networks. Um, so because most of the programming now is on the level of these data sets and make sure they're large, diverse and clean basically you have a data set that you think is good. You train your neural net, you deploy it and then you observe how well it's performing and you're trying to always increase the quality of your data set. You're trying to catch scenarios basically there are basically rare and it is in these scenarios that will typically struggling because they weren't told what to do in those rare cases in the dataset. But now you can close the loop because if you can now collect all those at scale, you can then feed them back into the reconstruction process I described and reconstruct the truth in those cases and added to the dataset. And so the whole thing ends up being like a staircase of improvement of perfecting your training set and you have to go through deployments so that you can mine uh the parts that are not yet represented well in the data set. So your data says basically in perfect, it needs to be diverse, it has pockets, there are missing and you need to pad out the pockets, you can sort of think of it that way in the data. </s><s>[INST] What role do humans play in this? So what's the this biological system like a human body is made up of cells? What what role like how do you optimize the human system, the multiple engineers collaborating, figuring out what to focus on what to contribute, which which task to optimize in this neural network, who is in charge of figuring out which task needs more data. What can you can you speak to the hyper parameters the human uh system, [/INST] it really just comes down to extremely good execution from an engineering team, who knows what they're doing, they understand intuitively the philosophical insights underlying the data engine and the process by which the system improves and uh how to again like delegate the strategy of the data collection and how that works and then just making sure it's all extremely well executed and that's where most of the work is, is not even the philosophizing or the research or the ideas of it is just extremely good execution is so hard when you're dealing with data at that </s><s>[INST] scale. So your role in the data engine, executing well on it. It is difficult and extremely important. Is there a priority of like uh like a vision board of saying like we really need to get better at stoplights. Like the prioritization of tasks is that essentially, and that comes from the data [/INST] that comes to um a very large extent to what we are trying to achieve in the product for a map where we're trying to the release, we're trying to get out um in the feedback from the QA team where the system is struggling or not, the things we're trying to improve and </s><s>[INST] the QA team gives some signal some information in aggregate about the performance of the system in various conditions [/INST] and then of course all of us drive it and we can also see it, it's really nice to work with the system that you can also experience yourself and you know, it drives you home. It's </s><s>[INST] is there some insight you can draw from your individual experience that you just can't quite get from an aggregate statistical Yeah, it's so weird. Right? Yes. It's not scientific in a sense because you're just one anecdotal sample. [/INST] Yeah, I think there's a ton of uh it's a source of truth is your interaction with the system and you can see it, you can play with it, you can perturb it, you can get a sense of it, you have an intuition for it. I think numbers just like have a way of numbers and plots and graphs are much harder. Uh, it hides a lot of </s><s>[INST] it's like if you train a language model, what it's a really powerful way is, is by you interacting with [/INST] it and try </s><s>[INST] to build up an intuition. [/INST] Yeah, I think like Ellen also like he always wanted to drive the system himself. He drives a lot and uh, I want to say almost daily. So, uh, he also sees this as a source of truth. You driving the system um, and performing and </s><s>[INST] yeah, so what do you think? Tough questions here? Uh, so Tesla last year removed radar from, from the sensor suite and now just announced there's going to remove ultrasonic sensors relying solely on vision. So camera only does that make the perception problem harder or easier? I [/INST] would almost reframe the question in some way. So the thing is basically you would think that additional sensors </s><s>[INST] by the way, can I just interrupt. I wonder if a language model will ever do that If you prompted? Let me reframe your question. That would be epic. [/INST] That's the wrong </s><s>[INST] problem. Sorry. [/INST] It's like a little bit of a wrong question because basically you would think that these sensors are an asset to you. But if you fully consider the entire product in its entirety. These sensors are actually potentially liability because the sensors aren't free. They don't just appear on your car. You need suddenly you need to have an entire supply chain. You have people procuring it. There can be problems with them. They may need replacement. They are part of the manufacturing process. They can hold back the line in the production, you need to source them to maintain them. You have to have teams that write the firmware, all of the, all of it. And then you also have to incorporate and fuse them into the system in some way. And so it actually like bloats the a lot of it. And I think Ellen is really good at simplify, simplify best part is no part. And he always tries to throw away things that are not essential because he understands the entropy in organizations and an approach And I think in this case the cost is high and you're not potentially seeing it if you're just a computer vision engineer and I'm just trying to improve my network and you know, is it more useful or less useful? How how useful is it? And the thing is once you consider the full cost of a sensor, it actually is potentially a liability and you need to be really sure that it's giving you extremely useful information in this case. We looked at using it or not using it and the delta was not massive. And so it's not useful </s><s>[INST] is it also blow in the data engine like having more sense [/INST] percent </s><s>[INST] and [/INST] these sensors, you know, they can change over time. For example, you can have one type of say radar, you can have other type of radar. They change over time now you suddenly need to worry about it. Now suddenly you have a column in your sequel light telling you, oh, what sensor type was it? And they all have different distributions and then uh, they can, they just, they contribute noise and entropy into everything and they bloat stuff And also organizationally has been really fascinating to me that it can be very distracting. Um, if you, if you, if you only want to get to work is vision, all the resources are on it and you're building out a data engine and you're actually making forward progress because that is the, the sensor with the most bandwidth, the most constraints in the world and you're investing fully into that and you can make that extremely good if you're, you're only a finite amount of sort of spend of focus across different facets of the system. </s><s>[INST] And uh, this kind of reminds me of which Sutton's a bitter lesson. It just seems like simplifying the system in the long run. Of course you don't know what the long run and it seems to be always the right solution [/INST] in </s><s>[INST] that case it was for RL but it seems to apply generally across all systems that do computation. So where what do you think about the lidar as a crutch debate. The battle between point clouds and pixels. [/INST] Yeah, I think this debate is always like slightly confusing to me because it seems like the actual debate should be about like do you have the fleet or not? That's like the really important thing about whether you can achieve really good functioning of an Ai system at the scale. So </s><s>[INST] data collection systems. [/INST] Yeah. Do you have a fleet or not? Is significantly more important whether you have lighter or not? It's just another sensor. Um and uh yeah, I think similar to the radar discussion basically. I um yeah I don't think it it basically doesn't offer extra, extra information is extremely costly. It has all kinds of problems you have to worry about it, you have to calibrate it etcetera. It creates bloat and entropy. You have to be really sure that you need this uh this um sensor in this case. I basically don't think you need it and I think honestly I will make a stronger statement. I think the others, some of the other companies who are using it are probably going to drop it. Yeah. </s><s>[INST] So you have to consider the sensor in the full in considering can you build a big fleet that collects a lot of data and can you integrate that sensor with that? That data and that sensor into a data engine that's able to quickly find different parts of the data that then continuously improves whatever the model that you're using [/INST] another way to look at it is like vision is necessary in the sense that uh dr uh the world is designed for human visual consumption so you need vision it's necessary and then also it is sufficient because it has all the information that you that you need for driving and humans obviously has a vision to drive. So it's both necessary and sufficient. So you want to focus resources and you have to be really sure if you're going to bring in other sensors, you could you could you could add sensors to infinity at some point you need to draw the line and I think in this case you have to really consider the full cost of any one sensor that you're adopting and do you really need it? And I think the answer in this case, you know, </s><s>[INST] so what do you think about the idea that that the other companies are forming high resolution maps and constraining heavily the geographic regions in which they operate? Is that approach not in your in your view um not going to scale over time to the entirety of the United States. [/INST] I think you mentioned like they pre map all the environments and they need to refresh the map and they have a perfect centimeter level accuracy map of everywhere they're gonna drive. It's crazy how are you going to, We've been talking about autonomy actually changing the world. We're talking about the deployment on the on the global scale autonomous systems for transportation and if you need to maintain sending your accurate map for earth or like many cities and keep them updated. It's a huge dependency that you're taking on huge dependency. It's it's a massive massive dependency and now you need to ask yourself do you really need it and humans don't need it? Um Right so it's it's very useful to have a low level map of like okay the connectivity of your road. You know that there's a four coming up when you drive an environment, you sort of have that high level understanding. It's like a small google map and Tesla uses google map like similar kind of resolution information in the system but it will not pre map environments centimeter level accuracy. It's a crutch. It's a distraction, it costs entropy and it diffuses the team. It dilutes the team and you're not focusing on what's actually necessary which is computer vision problem. </s><s>[INST] What did you learn about machine learning about engineering about life? About yourself? As one human being from working with Elon musk. [/INST] I think the most I've learned is about how to sort of run organizations efficiently and how to create efficient organizations and how to fight entropy in an organization. </s><s>[INST] So human engineering in the fight against [/INST] entropy. There's a there's a, I think Ellen is a very efficient warrior in the fight against entropy in organizations. </s><s>[INST] The organization look like exactly [/INST] it's process, it's it's process and </s><s>[INST] inefficiencies and and that kind of stuff. [/INST] Yeah meetings, he hates meetings, he keeps telling people to skip meetings if they're not useful. Um He basically runs the world's biggest startups I would say uh Tesla SpaceX are the world's biggest startups. Tesla actually multiple startups. I think it's better to look at it that way. And so I think he's he's extremely good at at that. And uh yeah he's a very good intuition for streamline processes, making everything efficient. Best part is no part simplifying focusing um and just kind of removing barriers moving very quickly, making big moves all this is very startup, p sort of seeming things but at scale </s><s>[INST] so strong drive to simplify from from your perspective. I mean that um that also probably applies to just designing systems and machine learning and otherwise like simplify simplify. What do you think is the secret to maintaining the startup culture in a company that grows. Is there, can you introspect that? [/INST] I do think he needs someone in a powerful position with a big hammer. Like Ellen who's like the cheerleader for that idea and ruthlessly ruthlessly pursues it. If no one has a big enough hammer, everything turns into committees, Democracy within the company process, talking to stakeholders decision making. Just everything just crumbles. If you have a big person who is also really smart and has a big hammer. Things move quickly. </s><s>[INST] So you said your favorite scene in Interstellar is the intense stocking scene with the ai and cooper talking saying cooper, what are you doing docking? It's not possible, No, it's necessary such a good line by the way. Just so many questions there, why an Ai in that scene presumably is supposed to be able to compute a lot more than the human is saying it's not optimal. Why the human, I mean that's a movie but shouldn't they? I know much better than the human anyway. What do you think is the value of setting seemingly impossible goals? So like uh our initial intuition which seems like something that you have taken on that Ellen espouses that where the initial intuition of the community might say this is very difficult and then you take it on anyway with a crazy deadline. You just from a human engineering perspective, uh have you seen the value of that? [/INST] I wouldn't say that setting impossible goals. Exactly is is a good idea, but I think setting very ambitious goals is a good idea. I think there's a, what I call sub linear scaling of difficulty, which means that 10 X problems are not 10 X hard, usually 10 X 10 X harder problem is like two or three X harder to execute on because if you want to actually like if you want to improve the system by 10% it costs some amount of work and if you want to improve the system it doesn't cost You know 100 x. Amount of work and because you fundamentally change the approach and if you start with that constraint then some approaches are obviously dumb and not going to work and it it forces you to reevaluate. Um and I think it's a very interesting way of approaching problem solving </s><s>[INST] but it requires a weird kind of thinking it's just going back to your like PhD days, it's like how do you think which ideas in the machine learning community are solvable? [/INST] Yes it's </s><s>[INST] uh it requires what is that? I mean there's the cliche first principles thinking but like it requires to basically ignore what the community is saying because it doesn't the community doesn't. A community in science usually draw lines of what is and isn't possible and like it's very hard to break out of that without going crazy. [/INST] Yeah I mean I think a good example here is you know the deep learning revolution in some sense because you could be in computer vision at that time when during the deep learning sort of revolution of 2012 and so on. Uh you could be improving a computer vision stack by 10% or we can just be saying actually all this is useless and how do I do 10 X. Better computer vision? Well it's not probably by tuning Hog feature detector I need a different approach um I need something that is scalable. Going back to uh Richard Sutton's um and understanding sort of like the philosophy of the bitter lesson and then being like actually I need much more scalable system. Like a neural network that in principle works and then having some deep believers that can actually execute on that mission and make it work. So that's the solution. </s><s>[INST] What do you think is the timeline to solve the problem of autonomous driving? That's still in part open question. [/INST] Yeah. I think the tough thing with timelines of self driving obviously is that no one has created self driving. </s><s>[INST] Yeah. [/INST] So it's not like what do you think is the timeline to build this bridge? Well we've built million bridges before. Here's how long that takes its. You know it's uh no one has built autonomy. It's not obvious. Uh some parts turned out to be much easier than others so it's really hard to forecast you do your best based on trend lines and so on and based on intuition but That's why fundamentally it's just really hard to forecast this. No one is </s><s>[INST] even still like being inside of it is hard to to do [/INST] some things turn out to be much harder and some things turn out to be much easier. </s><s>[INST] Do you try to avoid making forecasts? Because like Ellen doesn't avoid them right? And heads of car companies in the past have not avoided it either forward in other places have made predictions that we're going to solve level four. Driving by 2020 2021 whatever. And they all kind of backtracking that prediction. Are you as a as an ai person, do you for yourself privately make predictions or do they get in the way of like your actual ability to think about a thing? [/INST] Yeah, I would say like what's easy to say is that this problem is tractable and that's an easy prediction to make. It's tractable. It's going to work. Yes, it's just really hard. Something turned out to be harder than something turned out to be easier. So but it's it definitely feels tractable and it feels like at least the team at Tesla which is what I saw internally is definitely on track to that. How </s><s>[INST] do you form a strong representation that allows you to make a prediction about attract ability? So like you're the leader of a lot, a lot of humans, you have to kind of say this is actually possible. Like how do you build up that intuition? It doesn't have to be even driving, it could be other tasks. It could be. Um and I want to, what difficult task did you work on your life? I mean classification, achieving certain just an image in that certain level of super human level performance. Yeah, [/INST] expert intuition just intuition, it's belief. </s><s>[INST] So just like thinking about it long enough. Like studying looking at sample data, like you said driving my intuition is really flawed on this. Like I don't have a good intuition about tracked ability. It could be either, it could be anything, it could be solvable. Like uh, you know, the driving task could be simplified into something quite trivial. Like uh, the solution to the problem would be quite trivial and at scale more and more cars driving perfectly might make the problem much easier. The more cars you have drive, like people learn how to drive correctly, not correctly, but in a way that's more optimal for heterogeneous system of autonomous and semi autonomous and manually driven cars that could change stuff. Then again also have spent a ridiculous number of hours just staring at pedestrians crossing streets, thinking about humans and it feels like the way we use our eye contact, it sends really strong signals and there's certain quirks and edge cases of behavior and of course a lot of the fatalities that happen have to do with drunk driving and um, both on the pedestrian side and the driver's side. So there's that problem of driving at night and all that kind of so I wonder, you know, it's like the space of possible solution to autonomous driving includes so many human factor issues that it's almost impossible to predict. There could be super clean, nice solutions. [/INST] Yeah, I would say definitely like to use a game analogy, there's some fog of war, but you definitely also see the frontier of improvement and you can measure historically how much you've made progress. And I think for example, at least what I've seen and roughly five years of Tesla when I joined, it barely kept laying on the highway. I think going up from Palo Alto to SF was like three or four interventions anytime the road would do anything geometrically or turn too much, it would just like not work. And so going from that to like a pretty competent system in five years and seeing what happens also under the hood and what the scale of what the team is operating now with respect to data and computer and everything else, uh, is just a massive progress. So </s><s>[INST] just you're climbing a mountain and fog, but you're making a lot of progress, [/INST] You're making progress and you see what the next directions are and you're looking at some of the remaining challenges and they're not like, uh, they're not perturbing you and they're not changing your philosophy and you're not control contorting yourself. You're like actually these are the things that we still need to do. </s><s>[INST] The fundamental components of solving the problems seem to be there for the data engine, to the computer, the computer on the card, to the computer, the training, all that kind of stuff. So you've done over the years, you've been a test, you've done a lot of amazing breakthrough ideas and engineering, all of it. Um, from the data engine to the human side, all of it can you speak to why you chose to leave Tesla, [/INST] basically, as I described, that ran? I think over time, during those five years I've kind of gotten myself into a little bit of managerial position, most of my days where you know, meetings and growing the organization and making decisions about sort of high level strategic decisions about the team and what it should be working on and so on and uh it's kind of like a corporate executive role and I can do it, I think I'm okay at it, but it's not like fundamentally what I what I enjoy. And so I think when I joined um there was no computer vision team because Tesla was just going from the transition of using mobile, I a third party vendor for all of its computer vision to having to build this computer vision system. So when I showed up there were two people training deep neural networks and they were training them at a computer at their at their legs. Like down there was a </s><s>[INST] basic classification task [/INST] and so I kind of like grew that into what I think is a fairly respectable deep learning team, massive computer cluster, very good data annotation organization. And uh I was very happy with where that was, it became quite autonomous. And so I kind of stepped away and I, you know, I'm very excited to do much more technical things again. Yeah, and kind of like we focused on a G I </s><s>[INST] what was this soul searching like you took a little time off and think like what how many mushrooms did you take? I mean what what was going through your mind, the human lifetime is finite? [/INST] You </s><s>[INST] did a few incredible things, You're you're one of the best teachers of ai in the world, You're one of the best and I don't mean that, I mean that in the best possible way you're one of the best tinkerers in the Ai world, meaning like understanding the fundamental fundamentals of how something works by building it from scratch and playing with the basic intuitions, it's like Einstein Feinman, we're all really good at this kind of stuff, like a small example of a thing to to play with it, to try to understand it. So that and obviously now will you help build a team of machine learning um uh like engineers and a system that actually accomplished something in the real world, given all that, like what what was the soul [/INST] searching? Like? It was hard because obviously I love the company a lot and I love I love Ellen, I love Tesla, I want um it was hard to leave, I love the team basically. Um but yeah, I think actually I would potentially like interested in revisiting it when you're coming back at some point, are working optimist, working on a gi at Tesla, I think Tesla is going to do incredible things. It's basically like uh it's a massive large scale robotics kind of company with a ton of in house talent for doing really incredible things and I think uh human robots are going to be amazing. I think autonomous transportation is going to be amazing. All this is happening at Tesla. So I think it's just a really amazing organization. So being part of it and helping it along I think was very, basically enjoyed that a lot. Yeah, it was basically difficult for those reasons because I love the company but you know, I'm happy to potentially at some point come back for Act two. But I felt like at this stage I built the team, it felt autonomous and I became a manager and I wanted to do a lot more technical stuff, I wanted to learn stuff, I want to teach stuff. Uh and uh I just kind of felt like it was a good time for for a change of pace a little bit. </s><s>[INST] What do you think is the best movie sequel of all time? Speaking of Part 2? Because like because most of them suck [/INST] movie </s><s>[INST] Sequels and you tweet about movies. So just a tiny tangent is there? What's your, what's like a favorite movie sequel? Godfather? Part two. Um Are you a fan of Godfather cause you didn't even tweet or mention the Godfather. [/INST] Yeah, I don't love that movie. I know it hasn't </s><s>[INST] edit that out. We're going to edit out the hate towards The Godfather? How dare you [/INST] Think will make a strong statement? I don't know why. I don't know why, but I basically don't like any movie before 1995. Something like that. </s><s>[INST] Didn't you mention Terminator [/INST] 2? Okay, Okay. That's like a terminator two was a little bit later, 1990. </s><s>[INST] No, I think terminator two was in there [/INST] And I like terminator one as well. So okay, so like a few exceptions, but by and large, for some reason I don't like movies before 1995 or something, they feel very slow. The camera is like zoomed out, it's boring, it's kind of naive. It's kind of weird. </s><s>[INST] And also Terminator was very much ahead of its time. [/INST] Yes. And The Godfather, there's like no, A G. I joe </s><s>[INST] I mean, but you have Good Will Hunting was one of the movies you mentioned and that doesn't have any A G. I either. I guess that's mathematics. I [/INST] guess occasionally I do in the movies that don't feature or </s><s>[INST] like anchorman that has no, that's that [/INST] is so good. </s><s>[INST] I don't understand. Um Speaking of A G. I because I don't understand why Will Ferrell is so funny. It doesn't make sense. It doesn't compute there's just something about him and he's a singular human because you don't get that many comedies these days and I wonder if it has to do about the culture or the like The Machine of Hollywood or does it have to do it just we got lucky with certain people in comedy came together because he is a singular human. That was a ridiculous tangent, I apologize, but you mentioned humanoid robots. So what do you think about optimus? About Tesla body? Do you think we'll have robots in the factory in in in the home in 10 2030, 40, 50 years? [/INST] Yeah, I think it's very hard project, I think it's going to take awhile but who else is going to build human robots at scale? And I think it is a very good form factor to go after because like I mentioned that the world is designed for human in form factor. These things will be able to operate our machines, they will be able to sit down in chairs, potentially even drive cars. Uh basically the world is designed for humans, that's the form factor you want to invest into and make work over time. Uh I think you know, there's another school of thought which is okay, take a problem and design a robot to it. But actually designing a robot and getting a whole data engine and everything behind it to work is actually incredibly hard problem. So it makes sense to go after general interfaces that okay, they are not perfect for any one given task but they actually have the generality of just with a prompt with english able to do something across and so I think it makes a lot of sense to go after a general uh interface um, in the physical world. And I think it's a very difficult project. It's going to take time. Um, but I see no other, no other company that can execute on that vision. I think it's going to be amazing. Like uh basically physical labor, like if you think transportation is a large market, try physical labor insane, </s><s>[INST] but it's not just physical labor. To me, the thing that's also exciting is the social robotics. So the relationship will have on different levels with those robots. That's why I was really excited to see optimists like um people have criticized me for the excitement, but I've I've worked with uh a lot of research labs that do humanoid legged robots, boston dynamics, unitary. A lot there's a lot of companies that do legged robots, but that's the the elegance of the movement is a tiny tiny part of the big picture. So integrating the two big exciting things to me about Tesla doing humanoid or any legged robots is clearly integrating into the data engine. So the data engine aspect, so the actual intelligence for the perception and the and the control and the planning and all that kind of stuff integrating into this huge the fleet that you mentioned, right. Um and then speaking of fleet, the second thing is the mass manufacturers just knowing uh culturally uh driving towards a simple robot that's cheap to produce at scale and doing that well, having experience to do that. Well, that changes everything. That's why that's a very different culture and style than boston dynamics who by the way, those those robots are just the way they move. It's like, it'll be a very long time before Tessa can achieve the smoothness of movement. But that's not what it's about. It's, it's about, it's about the entirety of the system. Like we talked about the data engine and the fleet. That's super exciting. Even the initial sort of models but that too was really surprising that in a few months you can get a prototype [/INST] and the reason that happened very quickly is as you alluded to, there's a ton of copy paste from what's happening in the autopilot a lot. The amount of expertise that like came out of the woodworks at Tesla for building the human robot was incredible to see like basically Ellen said at one point, we're doing this. And then next day, basically like all these cad models started to appear and people talking about like the supply chain and manufacturing and people showed up with like screwdrivers and everything like the other day and started to like put together the body and I was like, whoa! Like all these people exist at Tesla and fundamentally building a car is actually not that different from building a robot the same. And that is true, not just for the hardware pieces and also, let's not forget hardware, not just for a demo, but um, manufacturing of that hardware at scale. It's like a whole different thing, but for software as well, basically this robot currently thinks it's a car. Uh, it's gonna </s><s>[INST] have a midlife crisis or something. It [/INST] thinks it's a car. Um, some of the earlier demos actually, we're talking about potentially doing them outside in the parking lot because that's where all of the computer vision that was like working out of the </s><s>[INST] box instead [/INST] of like inside. Um, but all the operating system, everything just copy paste. Computer vision mostly copy paste. I mean, you have to retrain the neural nuts, but the approach and everything and data engine and offline trackers. And the way we go about the occupancy tracker and so on everything copy paste. You just need to retrain your neural lots. Uh, and then the planning control of course has to change quite a bit, but there's a ton of copy paste from what's happening at Tesla. And so if you were to, if you were to go with the goal of like, okay, let's build a million human robots and you're not Tesla, that's that's a lot to ask if your Tesla, it's actually like, it's not, it's not that crazy. And </s><s>[INST] then the follow up question is, and how difficult just like we're driving. How difficult is the manipulation task such that it can have an impact at scale. I think depending on the context. The really nice thing about robotics is that unless you do a manufacturing that kind of stuff is there's more room for error. Driving is so safety critical and so that and also time critical robot is allowed to move slower which is nice. [/INST] I think it's going to take a long time. But the way you want to structure the development is you need to say okay it's going to take a long time. How can I set up the uh product development roadmap so that I'm making revenue along the way. I'm not setting myself up for a 01 loss function where it doesn't work until it works. You don't wanna be in that position. You want to make it useful almost immediately and then you want to slowly deploy it uh and uh </s><s>[INST] at [/INST] scale and you want to set up your data engine, your improvement loops, the telemetry, the evaluation, the harness and everything. Um and you want to improve the product over time incrementally and you're making revenue along the way. That's extremely important because otherwise you cannot build these large undertakings just like don't make sense economically. And also from the point of view of the team working on it, they need the dopamine along the way. They're not just going to make a promise about this being useful. This is going to change the world in 10 years when it works, it's not where you want to be, you want to be in a place like I think about a politics today where it's offering increased safety and um and uh convenience of driving today, people pay for it, people like it, people purchase it and then you also have the greater mission that you're working towards </s><s>[INST] and you see that so the doping you for the team that that was a source of happiness, a [/INST] 100% you're deploying this. People like it. People drive it, people pay for it, they care about it. There's all these YouTube videos, your grandma drives it. She gives you feedback people like it, people engage with it, you engage with it huge. </s><s>[INST] Do people that drive Tesla's like recognize you and give you love like like hey thanks for the for the this nice feature that is doing [/INST] Yeah, I think the tricky thing is like some people really love you, some people unfortunately like you're working on something that you think is extremely valuable, useful etcetera. Some people do hate you. There's a lot of people who like hate me and the team and everything the whole project and I </s><s>[INST] think Tesla drivers in [/INST] many cases they're not actually </s><s>[INST] yeah, that's that's actually makes me sad about humans or the current, the ways that humans interact, I think that's actually fixable. I think humans want to be good to each other. I think twitter and social media is part of the mechanism that actually somehow makes the negativity more viral that it doesn't deserve like disproportionately uh add like a viral viral boost the negativity, but I got, I wish people would just get excited about uh so suppressed some of the jealousy, some of the ego and just get excited for others. And then there's a karma aspect to that you get excited for others, they'll get excited for you the same thing in academia. If you're not careful, there is a like a dynamical system there. If you, if you think of in silos and get jealous of somebody else being successful, that actually perhaps counterintuitively uh leads to less productivity of you as a community and you individually, I feel like if you keep celebrating [/INST] others </s><s>[INST] that actually makes you more successful. Yeah, I think people haven't, depending on the industry haven't quite learned that [/INST] yet. Some people are also very negative and very vocal, so they're very prominently featured, but actually there's a ton of people who are cheerleaders, but they're silent, clearly cheerleaders and when you talk to people just in the world, they will tell you it's amazing. It's great, especially like people who understand how difficult it is to get this stuff working like people who have built products and makers, entrepreneurs entrepreneurs, like making making this work and changing something is incredibly hard, those people are more likely to cheerlead you. </s><s>[INST] Well, well, one of the things that makes me sad is some folks in the robotics community don't do the cheerleading than they should. There's uh because they know how difficult it is. Well they actually sometimes don't know how difficult it is to create a product that scale. Right? They actually deployed in the real world. A lot of the development of robots and AI system is done on very specific small benchmarks. Um, and as opposed to real world conditions. [/INST] Yeah, I think it's really hard to work on robotics in academic setting </s><s>[INST] or AI systems that apply in the real world. You, you've criticized you um flourished and loved for a time. The image net, the famed image that data set and I've recently had some words of criticism that the academic research ml community gives a little too much love still to the Image net or like those kinds of benchmarks. Can you speak to the strengths and weaknesses of datasets used in machine learning research? [/INST] Actually I don't know that I recall the specific instance where I was unhappy or criticizing Image net. I think image that has been extremely valuable. Uh it was basically a benchmark that allowed the deep learning community to demonstrate that deep neural networks actually work. It was uh there's a massive value in that. So I think that was useful but basically it's become a bit of an amnesty at this point. So feminist is like equal to 28 by 28 grayscale digits there's kind of a joke dataset that everyone like just crushes, </s><s>[INST] there's still papers written on amnesty, right? [/INST] Maybe like strong papers </s><s>[INST] like papers that focus on like how do we learn the small amount of data that [/INST] I could see that being helpful but not in sort of like mainline computer vision research anymore. Of course. I </s><s>[INST] think the way I've heard you somewhere, maybe I'm just imagining things, but I think you said like that was a huge contribution to the community for a long time and now it's time to move past those kinds of [/INST] that has been crushed. I mean, you know, the air rates are uh, Yeah, we're getting like 90% accuracy in in 1000 classification way, uh, prediction and I've seen those images and it's like really high. That's really, that's really good. If I'm correctly, the top five air rate is now like 1% or something. </s><s>[INST] Given your experience with the gigantic real world dataset, would you like to see benchmarks moving in certain directions that the research community uses? [/INST] Unfortunately, I don't think academics currently have the next image net. We've obviously, I think we've crushed em n'est, we've basically kind of crushed image net. Uh, and there's no next sort of big benchmark that the entire community rallies behind and uses. Um, you know, for further development of these networks. </s><s>[INST] Yeah. What what it takes for data set to captivate the imagination of everybody. Like where they all get behind it? That that could also need like a virus like a leader, right. Somebody with popularity. I mean, yeah. Why did image of that take off? Is there or is it just the accident of history? [/INST] It was the right amount of difficult. Uh, it was the right amount of difficult and simple and interesting enough. It just kind of like it was it was the right time for that kind of a data set </s><s>[INST] question from Reddit. Uh, what are your thoughts on the role that synthetic data and game engines will play in the future of neural net model development? [/INST] I think. Um, as neural nets converge to humans, uh, the value of simulation to neural nets will be similar to value of simulation to humans. So people use simulation for uh, people use simulation because they can learn something in that kind of a system. Um, and without having to actually experience it. Um, </s><s>[INST] but are you referring to the simulation we do in our head. No, [/INST] sorry, simulation. I mean, like video games or, you know, um, other forms of simulation for various professionals. </s><s>[INST] So let me push back on that because maybe there's simulation that we do in our heads. Like simulate if I do this. What do I think will happen? [/INST] Okay. That's like internal simulation. </s><s>[INST] Yeah. Internal. Isn't that what we're doing assuming before we act? [/INST] Yeah, but that's independent from like the use of simulation in the sense of like computer games or using simulation for training set creation or </s><s>[INST] is it independent or is it just loosely correlated? Because like isn't that useful to do like um counterfactual or like educators simulation to like, you know, what happens if there's a nuclear war? What happens if there's, you know like those kinds of things? Because yeah, [/INST] that's a different simulation from like Unreal engine. That's how I interpreted the question. </s><s>[INST] Ah So like simulation of the average case, is that what's unreal engine? What what what what what do you mean by unreal engine? So simulating a world physics of that world? Why is that different? Like because you also can add behavior to that world and you can try all kinds of stuff, right? Like you could throw all kinds of weird things into it was a real engine is not just about similar. I mean I guess it is about submitting the physics of the world. It's also doing something with that. [/INST] Yeah, the graphics, the physics and the agents that you put into the environment and stuff like that. </s><s>[INST] See, I think you I feel like you said that it's not that important I guess for the future of ai development, is that is that correct to interpret it that way? [/INST] I think humans use simulators for humans use simulators and they find them useful. And so computers will use simulators and find them useful. Okay, </s><s>[INST] so you're saying it's not that I I don't use simulators very often. I play a video game every once in a while, but I don't think I derive any wisdom about my own existence from, from those video games? It's a momentary escape from reality versus a source of wisdom about [/INST] reality. So </s><s>[INST] I don't, so I think that's a very polite way of saying simulation is not that useful. [/INST] Yeah, maybe. Maybe not. I don't see it as like a fundamental, really important part of like training neural nets currently. Uh, but I think as neural nets become more and more powerful, I think you will need fewer examples to train additional behaviors. And uh, simulation is of course there's a domain gap in a simulation that is not the real world is slightly something different. But with a powerful enough neural net you need um, the domain gap can be bigger I think because neural net will sort of understand that even though it's not the real world, it like has all this little structure that I'm supposed to be like, learn from. </s><s>[INST] So the neural net will actually, yeah, we'll be able to leverage the synthetic data better by closing the gap by understanding in which ways this is not real data. Exactly. Right. To do better questions next time. That was that was a question. But I'm just kidding. All right. Um, so is it possible do you think speaking of amnesty to construct neural nets and training processes that require very little data. So we've been talking about huge data sets, like the international training. I mean, one way to say that it's like you said like the aquarium itself as another level of training I guess and that requires a little data but do you see any value in doing research and kind of going down the direction of can we use very little data to train to construct a knowledge base? [/INST] 100%. I just think like at some point you need a massive data set and then when you pre train your massive neural nut and get something that you know it is like a GPT or something then you're able to be very efficient at training any arbitrary new task. So a lot of these G. P. T. S you know you can do tasks like sentiment analysis or translation or so on just by being prompted with very few examples. Here's the kind of thing I want you to do like here's an input sentence, here's the translation into german input sentence, translation to german input sentence blank and the neural net will complete the translation to german just by looking at sort of the example you've provided. And so that's an example of a very few shot learning in the activations of the neural net instead of the weights of the neural net. And so I think um basically just like humans neurons will become very data efficient at learning any other new task but at some point you need a massive data set to pre train your network </s><s>[INST] to get that. And probably we humans have something like that. Do we do we have something like that? Do we have a passive in the background background model constructing thing that just runs all the time in the self supervised way? We're not conscious of it. [/INST] I think humans definitely. I mean obviously we have we learnt a lot during during our life span but also we have a ton of hardware that helps us initial initialization coming from sort of evolution. And so I think that's also a really big big component. A lot of people in the field I think they just talk about the amount of like seconds and the you know that person has lived pretending that this is a sort of like a zero initialization of the neural nut. And it's not like you can look at a lot of animals like for example zebras, zebras get born and they see and they can run there's zero training data in their lifespan. They can just do that. So somehow I have no idea how evolution has found a way to encode these algorithms. And these neural net initialization are extremely good into A. T. C. Gs and have no idea how this works. But apparently it's possible because here's approved by existence. There's </s><s>[INST] something magical about going from a single cell to organism that is born to the first few years of life. I kind of like the idea that the reason we don't remember anything about the first few years of our life is that it's a really painful process. Like it's a very difficult challenging training process like intellectually like and maybe. Yeah, I mean I don't why don't we remember any of that? There might be some crazy training going on and that maybe that's the background model training. That is very painful. And so it's best for the system once it's trained not to remember how it's constructed. I [/INST] think it's just like the hardware for long term memory is just not fully developed. I kind of feel like the first few years of uh of infants is not actually like learning its brain </s><s>[INST] maturing [/INST] um were born premature. Um there's a theory along those lines because of the birth canal and the swelling of the brain and so were born premature and then the first few years were just the brain's maturing and then there's some learning eventually. Um it's my current view on it. </s><s>[INST] What do you think? Do you think can have long term memory? I like that approach is something like humans. Do you think, do you think there needs to be another meta architecture on top of it to add something like a knowledge base that learns facts about the world and all that kind of stuff? [/INST] Yes, but I don't know to what extent it will be explicitly constructed. Um It might take on intuitive forms where you are telling the gPT like, hey you have a, you have a declarative memory bank to which you can store and retrieve data from and whenever you encounter some information that you find useful, just save it to your memory bank and here's an example of something you have retrieved and here's how you say it and here's how you load from it, you just say load whatever you teach it in text in english and then it might learn to use a memory bank from from that. Oh, </s><s>[INST] so so the neural net is the architecture for the background model, the base thing and then everything else is just on top of that. [/INST] It's not just the text, right? It's you're giving it gadgets and gizmos. So uh you're teaching some kind of special language by which we can, we can save arbitrary information and retrieve it at a later time and you're, you're telling about the special tokens and how to arrange them to use these interfaces. It's like, hey you can use a calculator, here's how you use it. Just do 53 plus 41 equals and when equals is there a calculator will actually read out the answer and you don't have to calculate it yourself and you just like tell it in english this might actually work, </s><s>[INST] do you think in that sense Gado is interesting. The deepmind system that it's not just the language but actually throws it all uh in the same pile, images, actions, all that kind of stuff that's basically what we're moving towards. Yeah, [/INST] I think so. So Gatto is, is very much a kitchen sink approach to like um reinforcement learning lots of different environments with a single fixed transformer model. Right. Um I think it's a very sort of early result in that in that realm. But I think yeah, it's along the lines of what I think things will eventually look like. </s><s>[INST] Right? So this is the early days of a system that eventually will look like this. Like from a rich, rich, sudden perspective. [/INST] I'm not super huge fan of, I think all these interfaces that look very different. Um I would want everything to be normalized into the same ap I so for example, screen pixels very same api instead of having like different world environments that are very different physics and joint configurations and appearances and whatever and you're having some kind of special tokens for different games that you can plug. I'd rather just normalize everything to a single interface. So it looks the same to the neural net, if that makes sense. So it's </s><s>[INST] all gonna be pixel based pong in the end. [/INST] I think so, okay, </s><s>[INST] uh let me ask you about your own personal life. A lot of people want to know you're one of the most productive and brilliant people in the history of ai what is a productive day in the life of Andre Karpati look like. What time do you wake up? Because imagine um some kind of dance between the average productive day and a perfect productive day. So the perfect productive day is the thing we strive towards and the average is kind of what it kind of converges to giving all the mistakes and human eventualities and so on. So what time do you wake up? Like a morning person? [/INST] I'm not a morning person, I'm a night owl for sure </s><s>[INST] it's stable or not, [/INST] it's semi stable like eight or nine or something like that. During my PhD it was even later I used to go to sleep usually at three am I think uh am hours are are precious and very interesting time to work because everyone is asleep um at at eight AM or seven AM. The East coast is awake. So there's already activity, there's already some text messages, whatever there's stuff happening, you can go on like some news website and there's stuff happening distracting at three a.m. Everything is totally quiet and so you're not gonna be bothered and you have solid chunks of time to do work. Um So I like those periods night all by default and then I think like productive time basically um what I like to do is you need you need to like build some momentum on the problem without too much distraction and um you need to load your RAM uh your working memory with that problem and then you need to be obsessed with it when you're taking a shower, when you're falling asleep. You need to be obsessed with the problem and it's fully in your memory and you're ready to wake up and work on it right there. So </s><s>[INST] it is the scale of uh is this in a scale temporal scale of a single day or a couple of days a week? A month. [/INST] So I can't talk about one day basically in isolation because it's a whole process when I want to get when I want to get productive in the problem. I feel like I need a span of a few days where I can really get in on that problem and I don't wanna be interrupted and I'm going to just be completely obsessed with that problem and that's where I do most of my good workouts. </s><s>[INST] You've done a bunch of cool like little projects in a very short amount of time very quickly so that that requires you just focusing on it. [/INST] Yeah basically I need to load my working memory with the problem and I need to be productive because there's always like a huge fixed cost to approach any problem. Uh you know like I was struggling with this for example at Tesla because I want to work on the small side project but okay you first need to figure out okay I need to associate into my cluster, I need to bring up a V. S. Code editor so I can work on this. I need to I run into some stupid error because of some reason like you're not at a point where you can be just productive right away you are facing barriers and so it's about really removing all that barrier and you're able to go into the problem and you have the full problem loaded in your memory </s><s>[INST] and somehow avoiding distractions of all different forms like uh news stories, emails but also distractions from other interesting projects that you previously worked on are currently working on and so on. You just want to really focus your mind [/INST] and I mean I can take some time off for distractions and in between but I think it can't be too much uh you know most of your day is sort of like spent on that problem and then you know I drink coffee, I had my morning routine. I look at some news twitter, hacker news, Wall Street Journal etcetera so it's great. </s><s>[INST] So basically you wake up you have some coffee. Are you trying to get to work as quickly as possible? Do you do taking this diet of of like what the hell is happening in the world? 1st [/INST] I am I do find it interesting to know about the world. I don't know that it's useful or good but it's part of my routine right now. So I do read through a bunch of news articles and I want to be informed and um I'm suspicious of it. I'm suspicious of the practice but currently that's where I am. Oh </s><s>[INST] you mean suspicious about the positive effect of that practice on your productivity and your well being is [/INST] my well being psychologically and </s><s>[INST] also on your ability to deeply understand the world because there's a bunch of sources of information, you're not really focused on deeply integrating [/INST] distracting your Yeah, </s><s>[INST] in terms of perfectly productive day for how long of a stretch of time In one session, do you try to work and focus on the thing? A couple of hours? There's one hours and 30 minutes, 10 minutes. [/INST] I can probably go like a small few hours and then I need some breaks in between for like food and stuff. And yeah, but I think like it's still really hard to accumulate hours. I was using a tracker that told me exactly how much time I spent coding any one day and even on a very productive day I still spent only like six or eight hours. Yeah, and it's just because there's so much padding, commute talking to people, food etcetera, there's like the cost of life just living and sustaining and homeostasis and just maintaining yourself as a human is very high and </s><s>[INST] and that there seems to be a desire within the human mind to to participate in society that creates that pattern because I yeah, the most productive days I've ever had is just completely from start to finish, just tuning out everything and just sitting there and then and then you could do more than six and eight hours, Is there some wisdom about what gives you strength to do? Like uh tough days of long focus? [/INST] Yeah, just like whenever I get obsessed about a problem, something just needs to work, something just needs to exist. </s><s>[INST] It needs to exist. And so you're able to deal with bugs and programming issues and technical issues and uh design decisions that turned out to be the wrong ones, you're able to think through all of that given, given that you want to think to exist, [/INST] it needs to exist. And then I think to me also a big factor is, you know, our other humans are going to appreciate it. Are they going to like it? That's a big part of my motivation if I'm helping humans and they seem happy, they say nice things, they tweet about it or whatever that gives me pleasure because I'm doing something useful. </s><s>[INST] So like you do see yourself sharing it with the world, like what? Get help with the blog posts or videos? [/INST] Yeah, I was thinking about it like suppose I did all these things but did not share them. I don't think I would have the same amount of motivation that I can build up. </s><s>[INST] You enjoy the feeling of other people gaining value and happiness from the stuff you've created. What about diet is there? I saw you play with intermittent fasting you fast. Does that help [/INST] with everything </s><s>[INST] with the things you played? What's been most beneficial to the, your ability to mentally focus on the thing and just mental mental productivity and happiness. You're still fast [/INST] so fast. But I do intermittent fasting. But really what it means at the end of the day is I skip breakfast. So I do uh 18 6 roughly by default when I'm in my steady state, if I'm traveling or doing something else, I will break the rules. But in my steady state I do 18 6, so I eat only from 12 to 6. Not a hard rule and I break it often, but that's my default. And then um yeah, I've done a bunch of random experiments for the most part right now where I've been for the last year and a half I want to say is um plant based or plant forward. I heard plant forward. It sounds </s><s>[INST] better. Exactly. I [/INST] don't actually know the difference is, but it sounds better in my mind, but it just means that I prefer plant based food and uh </s><s>[INST] or cooked, I [/INST] prefer cooked and plant </s><s>[INST] based. So plant based. Forgive me, I don't actually know how why the category of plant entails [/INST] just means that you're not </s><s>[INST] about it [/INST] and you can flex and uh you just prefer to eat plants and you know, you're not making, you're not trying to influence other people and if someone is, you come to someone's house party and they serve you a steak that they're really proud of. You will eat it. It's </s><s>[INST] just not judgmental. That's beautiful. I mean that's um on the flip side of that, but I'm very sort of flexible. Have you tried doing one meal a day? [/INST] I have accidentally, not consistently, but I've accidentally had that. I don't, I don't like it. I think it makes me feel not good. It's too, it's too much, too much of a hit. And so currently I have about two meals a day, 12 and six. </s><s>[INST] I do that nonstop. I'm doing it. No one meal a day. It's interesting. It's an interesting feeling. Have you ever fasted longer than a day? [/INST] Yeah, I've done a bunch of water fasts because I was curious what happens. Uh </s><s>[INST] Anything interesting? Yeah, [/INST] I would say so. I mean, you know what's interesting is that you're hungry for two days and then starting day three or so you're not hungry. It's like such a weird feeling because you haven't eaten in a few days and you're not hungry. Is </s><s>[INST] that weird? It's really one of the many weird things about human biology. It figure something out. It finds, finds another source of energy or something like that or relaxes the system. I don't know how [/INST] the body is, like you're hungry, you're hungry and then it just gives up. It's like, okay, I guess we're fasting now. </s><s>[INST] There's nothing [/INST] and then it just kind of like focuses on trying to make you not hungry and you know not feel the damage of that and trying to give you some space to figure out the food situation. </s><s>[INST] So are you still to this day most productive at [/INST] night? I would say I am, but it is really hard to maintain my PhD schedule um especially when I was working at Tesla and so on, it's a nonstarter, so, but even now like you know, people want to meet for various events, they society lives in a certain period of time and you sort of have to like so that </s><s>[INST] it's hard to like do a social thing and then after that return and do work. [/INST] Yeah, it's just really hard. </s><s>[INST] That's why I tried when I do social things, I try not to do too too much drinking so I can return and continue doing work. Um but Tesla, is there is there a convergence Tesla, but any company, is there a convergence to him as a schedule or is there more, is that how humans behave when they collaborate? I need to learn about this. Do they try to keep us consistent schedule? You're all awake at the same time? [/INST] I'm gonna do try to create a routine and I try to create a steady state in which I'm comfortable in, so I have a morning routine. I have a day routine. I try to keep things to a steady state and things are predictable and then you can sort of just like your body just sort of like sticks to that and if you try to stress that a little too much it will create uh you know when you're traveling and you're dealing with Jetlag, you're not able to really ascend to, you know where you need to go. </s><s>[INST] Yeah. Yeah. That's what about humans with the habits and stuff? What are your thoughts on work life balance throughout a human lifetime? So testing part was known for sort of pushing people to their limits in terms of what they're able to do in terms of what they're trying to do in terms of how much they work, all that kind of stuff. [/INST] Yeah, I mean I will say Tesla is still too much bad rep for this because what's happening is Tesla is a it's a bursty environment. So I would say the baseline, my only point of reference is google, like where I've entered three times and I saw what it's like inside google and deepmind. Um I would say the baseline is higher than that, but then there's a punctuated equilibrium where once in a while there's a fire and someone like people work really hard and so it's spiky and bursty and then all the stories get collected </s><s>[INST] and [/INST] then it gives the appearance of like total insanity, but actually it's just a bit more intense environment and there are fires and sprints and so I think, you know, it definitely though. I would say um it's a more intense environment than something you would get </s><s>[INST] your person forget all of that just in your own personal life. Um What do you think about the happiness of a human being, a brilliant person like yourself about finding a balance between work and life or is it such a thing? Not a good thought experiment? [/INST] Yeah, I think, I think balance is good, but I also love to have sprints that are out of distribution and that's when I think I've been pretty uh creative. And um as well, </s><s>[INST] sprints out of distribution means that most of the time you have uh quote unquote balance, [/INST] I have balanced most of the time. I like being obsessed with something once in a while, </s><s>[INST] once in a while, once a week, Once a month, once a year. [/INST] Probably like once a month or something. </s><s>[INST] Yeah. And that's when we get and you get [/INST] that's when you really care about a problem, it must exist. This will be awesome. You're obsessed with it. And now you can't just do it on that day. You need to pay the fixed cost of getting into the groove and then you need to stay there for a while and then society will come and they will try to mess with you and they will try to distract you. Yeah. The worst thing is like a person who's like, I just need five minutes of your time. Yeah. This is the cost of, that is not five minutes and society needs to change how it thinks about just five minutes of your </s><s>[INST] time. Right? It's never it's never just one minute. It's just just a quick what's the big [/INST] deal are you being so? </s><s>[INST] Yeah. No. What's your computer setup? What what's like the perfect. Are you somebody that's flexible too, no matter what laptop, four screens or do you prefer a certain set up that you're most productive? [/INST] Um I guess the one that I'm familiar with is one large screen 27 in um and my laptop on the side, </s><s>[INST] what operating system [/INST] I do max that's my primary </s><s>[INST] for all tasks [/INST] I would say was x. But when you're working on deep learning everything as Linux, your ssh into a cluster and you're working remotely, </s><s>[INST] but what about the actual development, like they're using the I. D. [/INST] You would use I think a good way is you just run V. S. Code um My favorite editor right now on your Mac, but you are actually you have a remote folder through ssh. Um So the actual files that you're manipulating on the cluster somewhere else. </s><s>[INST] So what's the best I. D. V. S code? What else do people? So I use imax still uh It may be cool. I don't know if it's maximum productivity. Um So what what do you recommend in terms of editors? You worked a lot of software engineers, editors for python, c plus plus machine machine learning applications. [/INST] I think the current answer is V. S. Code. Currently I believe that's the best. Um I. D. It's got a huge amount of extensions. It has Guitar, co pilot um uh integration which I think is very valuable. </s><s>[INST] What do you think about the co pilot integration? I was actually uh I got to talk a bunch with was a creative python and he loves co pilot. You like programs a lot with it. Uh Do you [/INST] use co pilot? I love it and it's free for me but I would pay for it. Yeah I think it's very good and the utility that I found with it was isn't isn't I would say there's a learning curve and you need to figure out when it's helpful and when to pay attention to its outputs and when it's not going to be helpful where you should not pay attention to it because if you're just reading it suggestions all the time it's not a good way of interacting with it. But I think I was able to like mold myself to it. I found it very helpful number one in copy paste and replace some parts. So I don't know when the pattern is clear, it's really good at completing the pattern and number two. Sometimes it suggests a p. I is that I'm not aware of. So it it tells you about something that you didn't know. So </s><s>[INST] and that's an opportunity to discover and [/INST] it's an opportunity to so I would never take copilot code has given, I almost always copy copy paste into a google search and you see what this function is doing and then you're like, oh it's actually actually exactly what I need, thank you co pilot. So you learn something </s><s>[INST] so it's in part of search engine apart, maybe getting the exact syntax correctly that once you see it. Yeah, it's that and be hard things like once you see it, you know, [/INST] correct. </s><s>[INST] You yourself [/INST] can struggle, </s><s>[INST] you can verify efficiently, but you can't generate efficiently and [/INST] co pilot really, I mean it's it's autopilot for programming, right? And currently is doing the link following, which is like the simple copy paste and sometimes suggest, but over time it's going to become more and more autonomous. And so the same thing will play out and not just coding, but actually across many, many different things, probably </s><s>[INST] coding is an important one, right? Writing programs. How do you see the future of that, developing the program synthesis? Like being able to write programs that are more and more complicated because right now it's human supervised in interesting ways like what it feels like the transition will be very painful. [/INST] My mental model for it is the same thing will happen as with the autopilot. So currently doing the following is doing some simple stuff and eventually we'll be doing autonomy and people will have to intervene less and less </s><s>[INST] and there could be like like testing mechanisms like if it writes a function and that function looks pretty damn correct. But how do you know it's correct because you're like getting lazier and lazier as a programmer, like your ability to cause like little bugs but I guess it won't make a little [/INST] it will it co pilot will make off by one subtle bugs. It has done that to me. But </s><s>[INST] do you think future systems will or is it really off by one is actually a fundamental challenge of programming. [/INST] In that case it wasn't fundamental and I think things can improve but yeah I think humans have to supervise. I am nervous about people not supervising what comes out and what happens to for example the proliferation of bugs in all of our systems. I'm nervous about that but I think there will probably be some other co pilots for bug finding and stuff like that at some point because there will be like a lot more automation for </s><s>[INST] man, [/INST] it's </s><s>[INST] like a program, [/INST] a co </s><s>[INST] pilot that generates a compiler, one that does a lint er one that does like a type checker, [/INST] it's a committee of like GPT sort of like </s><s>[INST] and then they'll be like a manager for the committee and then there would be somebody that says a new version of this is needed. We need to regenerate it. [/INST] Yeah there were 10 G. P. T. S. They were forwarded and get 50 suggestions. Another one looked at it and picked a few that they like a bug, one looked at it and it's like it's probably a bug. They got re ranked by some other thing and then a final ensemble GPT comes in. It's like okay given everything you guys have told me this is probably the next token. You </s><s>[INST] Know the feeling is the number of programmers in the world has been growing and growing very quickly. Do you think it's possible that it will actually level out and drop to like a very low number with this kind of world? Because then you'll be doing software 2.0 programming. Um And you'll be doing this kind of generation of co pilot type systems, programming but you won't be doing the old school software one point oh programming. I [/INST] don't currently think that they're just going to replace human programmers. Um I'm so hesitant saying stuff like this right because </s><s>[INST] this is going to be replaced [/INST] in five years. </s><s>[INST] I know it's going to show [/INST] that like this </s><s>[INST] is where we thought because I I agree with you but I think we might be very [/INST] surprised </s><s>[INST] right? Like what are the next i what's your sense of where we stand with language models? Like does it feel like the beginning or the middle or the end? [/INST] The beginning? 100%. I think the big question in my mind is for sure. GPT will be able to program quite well competently and so on. How do you steer the system? You still have to provide some guidance to what you actually are looking for. And so how do you steer it? And how do you say, how do you talk to it? How do you um audited and verified that what is done is correct. And how do you like work with this? And it's as much not just an ai problem but a ui ux problem. Yeah. Um So beautiful fertile ground for so much interesting work for V. S. Code plus plus where you're not just not just human programming anymore. It's amazing. </s><s>[INST] Yeah. So you're interacting with the system so not just one prompt but its iterative prompting trying to figure out having a conversation with the system. That actually I mean to me that's super exciting to have a conversation with the program I'm writing. [/INST] Yeah maybe at some point you're just conversing with it. It's like okay here's what I wanna do actually this variable. Maybe it's not even that low level is variable but you </s><s>[INST] can also imagine like can you translate this to C. Plus plus and back to python [/INST] already kind of existence </s><s>[INST] but just like doing it as part of the program experience. Like I think I'd like to write this function in C Plus plus or like you just keep changing from different different programs because there's different syntax, maybe I want to convert this into a functional language and so like you get to become multi lingual as a programmer and dance back and forth officially. [/INST] I think the Ui ux of it though is like still very hard to think through because it's not just about writing code on a page, you have an entire developer environment, you have a bunch of hardware on it. Uh you have some environmental variables, you have some scripts that are running in a Cron job like there's a lot going on to like working with computers and how do these systems set up environment flags and work across multiple machines and set up screen sessions and automate different processes like how all that works and is auditable by humans and so on is like massive questions. The moment </s><s>[INST] you've built archive sanity, what is archive and what is the future of academic research publishing that you would like to see? [/INST] So archive is this pre print server. So if you have a paper you can submit it for publication to journals or conferences and then wait six months and then maybe get a decision pass or fail or you can just uploaded to archive and then people can tweet about it three minutes later and then everyone sees it. Everyone reads it and everyone can profit from it in their own ways </s><s>[INST] and you can cite it and it has an official look to it. It feels like a like it feels like a publication process. It feels different than if you just put in a blog post. [/INST] Oh yeah, I mean it's a paper and usually the bar is higher for something that you would expect on archive as opposed to and something you would see in a blog post </s><s>[INST] or the culture created the bar because you could probably host a pretty crappy fix for an archive. Um So what's that make you feel like, what does that make you feel about peer review? So rigorous peer review by 23 experts versus the peer review of the community? Right? As it's written [/INST] basically, I think the community is very well able to peer review things very quickly on twitter and I think maybe it just has to do something with A. I. Machine learning field specifically though, I feel like things are more easily auditable. Um and the verification is easier potentially than the verification somewhere else. So it's kind of like um you can think of these scientific publications as like little block chains where everyone's building on each other's work and setting each other and you sort of have A I which is kind of like this much faster and loose Blockchain, but then you have, and any one individual entry is like very, very cheap to make and then you have other fields where maybe that model doesn't make as much sense. Um And so I think in a I at least things are pretty easily verifiable and so that's why when people upload papers, they're a really good idea and so on, People can try it out the next day and they can be the final arbiter of whether it works or not on their problem and the whole thing just moves significantly faster. So I kind of feel like academia still has a place, sort of just like conference journal process still has a place, but it's sort of like um it lags behind I think and it's a bit more maybe higher quality process, but it's not sort of the place where you will discover cutting edge work anymore. It used to be the case when I was starting my PhD that you go to conferences and journals and you discuss all the latest research now when you go to a conference or general like no one discusses anything that's there because its already like three generations ago irrelevant. </s><s>[INST] Which makes me sad about like Deepmind, for example, where they still they still publish in nature and these big prestigious, I mean there's still value I suppose to the prestige that comes with these big venues, but the result is that they will announce some breakthrough performance and it will take like a year to actually publish the details, I mean, and those details in if they were published immediately would inspire the community to move in certain directions with that [/INST] let's speed up the rest of the community, but I don't know to what extent that's part of their objective function also, </s><s>[INST] that's true. So it's not just the prestige. A little bit of the delay is uh, is part of Yeah, [/INST] they certainly deepmind specifically has been um, working in the regime of having slightly higher quality, basically process and latency and publishing those papers that way. Another </s><s>[INST] question from Reddit, Do you or have you suffered from imposter syndrome being the director of Ai. Tesla, being this person when you're stanford, where like the world looks at you as the expert in ai to teach, teach the world about machine learning. [/INST] When I was leaving Tesla after five years, I spent a ton of time in meeting rooms. Uh, and you know, I would read papers in the beginning when I joined Tesla, I was writing code and then I was writing less and less code and I was reading code and then I was reading less and less code. And so this is just a natural progression that happens, I think. And uh, definitely I would say near the tail end, that's when it sort of starts to hit you a bit more that you're supposed to be an expert. But actually the source of truth is the code that people are writing the guitar and the actual, the actual code itself and you're not as familiar with that as it used to be. And so I would say maybe there's some like insecurity there. </s><s>[INST] Yeah, that's actually pretty profound that a lot of the insecurity has to do with not writing the code in the computer science space like that, because that is the truth, that that code [/INST] is the source of truth, the papers and everything else. It's a high level summary. I don't yeah, just a high level summary. But at the end of the day you have to read code, it's impossible to translate all that code into actual, you know, paper form. So when when things come out, especially when they have a source code available, that's my favorite place to </s><s>[INST] go. So like I said, you're one of the greatest teachers of machine learning ai ever from CS to 31 end to today, what advice would you give to beginners? Interested in getting into machine learning, [/INST] beginners are often focused on </s><s>[INST] like what [/INST] to do, and I think the focus should be more like how much you do. So I'm kind of like believer on the high level in this 10,000 hours kind of concept where you just kind of have to just pick the things where you can spend time and you you care about and you're interested in, you literally have to put in 10,000 hours of work. Um it doesn't even matter as much like where you put it and you'll iterate and you'll improve and you'll waste some time. I don't know if there's a better way you need to put in 10,000 hours, But I think it's actually really nice because I feel like there's some sense of determinism about being an expert at a thing. If you spend 10,000 hours you can literally pick an arbitrary thing. And I think if you spend 10,000 hours of deliberate effort and work you actually will become an expert at it. And so I think it's kind of like a nice thought. Um and so uh basically I would focus more on like are you spending 10,000 hours? and </s><s>[INST] So and then thinking about what kind of mechanisms maximize your likelihood of getting to 10,000 hours? [/INST] Exactly. Which </s><s>[INST] for us silly humans means probably forming a daily habit of like every single day actually doing the thing [/INST] whatever helps you. So I do think to a large extent is a psychological problem for yourself. One other thing that I help that I think is helpful for the psychology of it is many times people compare themselves to others in the area I think is very harmful. Only compare yourself to you from some time ago, like say a year ago, are you better than you year ago is the only way to think. Um And I think this then you can see your progress and it's very motivating. </s><s>[INST] That's so interesting that focus on the quantity of hours because I think a lot of people in the beginning stage but actually throughout get paralyzed by uh the choice like which one do I pick this path of this path, like they'll literally get paralyzed by like which I d to use. Well [/INST] they're worried they're worried about all these things. But the thing is some of the you will waste time doing something wrong, you will eventually figure out it's not right, you will accumulate scar tissue and next time you'll grow stronger because next time you'll have the scar tissue and next time you'll learn from it and now next time you come to a similar situation you'll be like oh I I messed up, I've spent a lot of time working on things that never materialized into anything and I have all that scar tissue and I have some intuitions about what was useful wasn't useful how things turned out. So all those mistakes were were not that work, you know? So I just think you should just focus on working what have you done, what have you done last week? </s><s>[INST] That's a good question actually to ask for for a lot of things, not just machine learning. Um It's a good way to cut the, I forgot what the term we use but the fluff the blubber, whatever the uh the inefficiencies in life. What do you love about teaching, You seem to find yourself often in the like drawn to teaching. You're very good at it, but you're also [/INST] drawn to it. I mean I don't think I love teaching, I love happy humans </s><s>[INST] and [/INST] happy humans. Like when I teach, </s><s>[INST] I [/INST] wouldn't say I hate teaching, I tolerate teaching but it's not like the act of teaching that I like. It's it's that um You know I I have some I have something I'm actually okay at it. I'm okay at teaching and people appreciate a lot. And so I'm just happy to try to be helpful and teaching itself is not like the most. I mean it's really it's can be really annoying frustrating. I was working on a bunch of lectures just now. I was reminded back to my days of 231 and just how much work it is to create some of these materials and make them good. The amount of iteration and thought and you go down blind alleys and just how much you change it. So creating something good. Um in terms of educational value is really hard and it's not fun. It's </s><s>[INST] difficult. So people should definitely go watching new stuff you put out. There are lectures where you actually building the thing like from like you said the code is truth. So discussing back propagation by building it by looking through and just the whole thing. So how difficult that to prepare for. I think it's a really powerful way to teach. How did you have to prepare for that or are you just live thinking through it? [/INST] I will typically do like say three takes and then I take like the better take. So I do multiple takes and I take some of the better takes and then I just build out a lecture that way. Sometimes I have to delete 30 minutes of content because it just went down an alley that I didn't like too much. There's about a bunch of iteration and it probably takes me, you know, somewhere around 10 hours to create one hour of content </s><s>[INST] to give one hour. It's interesting. I mean, is it difficult to go back to the, like, the basics? Do you draw a lot of, like, wisdom from going back to the basics? [/INST] Yeah. Going back to back propagation loss functions where they come from. And one thing I like about teaching a lot, honestly is it definitely strengthens your understanding. So it's not a purely altruistic activity. It's a way to learn if you have to explain something to someone, uh you realize you have gaps in knowledge. Uh And so I even surprised myself in those lectures, like, also the result will obviously look at this and then the result doesn't look like it. And I'm like, okay, I thought I understood this, but that's why </s><s>[INST] it's really cool to literally code. You run it in the notebook and it gives you a result and you're like, oh wow, like actual numbers, actual input, you know, actual code, [/INST] it's not mathematical symbols, etcetera. The source of truth is the code. It's not slides, it's just like, let's build </s><s>[INST] it, It's beautiful, You're a rare human in that sense. What advice would you give to researchers trying to develop and published idea that have a big impact in the world of aI so maybe um undergrads, maybe early graduate students. [/INST] I mean, I would say like they definitely have to be a little bit more strategic than I had to be as a PhD student because of the way I is evolving. It's going the way of physics where, you know, in physics, you used to be able to do experiments on your bench top and everything was great and you can make progress and now you have to work in like LHC or like cern and and so ai is going in that direction as well. Um, so there's certain kinds of things that's just not possible to do on the bench top anymore. And uh, I think um that didn't used to be the case at the time. Do </s><s>[INST] you still think that there's like gan type papers to be written, were like, like very simple idea that requires just one computer to illustrate a simple example. [/INST] I mean, one example that's been very influential recently is diffusion models. The fusion models are amazing. The fusion models are six years old for the longest time. People are kind of ignoring them as far as I can tell and they're an amazing generative model, especially in images and so stable diffusion and so on. It's all diffusion based. The fusion is new, it was not there and came from, well, it came from google, but a researcher could have come up with it. In fact, some of the first actually, no, those came from google as well. But a researcher could come up with that in an academic institution. </s><s>[INST] Yeah. What do you find most fascinating about diffusion models? So from the societal impact of the technical architecture, [/INST] what I like about the fusion is it works so well. </s><s>[INST] Was that is that surprising to you? The amount of the variety almost the novelty of the synthetic data is generating? [/INST] Yes, So the stable diffusion images are incredible. It's the speed of improvement in generating images has been insane. Uh We went very quickly from generating like tiny digits to the tiny faces and it all looked messed up. And now you're stable diffusion. And that happened very quickly. There's a lot that academia can still contribute, you know, for example, um Flash attention is a very efficient colonel for running the attention operation inside the transformer that came from academic environment. It's a very clever way to structure the colonel. Uh that that's the calculation, so it doesn't materialize the attention matrix. Um and so there's I think there's still like lots of things to contribute, but you have to be just more strategic. Do </s><s>[INST] you think neural networks can be made to reason? [/INST] Uh Yes. Do </s><s>[INST] you think they're already reason. [/INST] What's </s><s>[INST] your definition of reasoning? [/INST] Uh information processing. </s><s>[INST] So in the way that humans think through a problem and come up with novel ideas, it feels like reasoning. So the novelty, I don't want to say, But out of distribution ideas you think it's possible? [/INST] Yes. And I think we're seeing that already in the current you're able to remix the training set information into true generalization in some sense. </s><s>[INST] That doesn't appear it doesn't appear in the in [/INST] the training set. Like you're doing something interesting, algorithmic lee, you're manipulating, you know, some symbols and you're coming up with some correct a unique cancer in a new setting. </s><s>[INST] What would [/INST] uh </s><s>[INST] illustrate to you? Holy ship? This thing is definitely thinking [/INST] to me thinking or reasoning is just information processing and generalization. And I think the neural nets already do that today. </s><s>[INST] So being able to perceive the world or perceive the whatever the inputs are and to make predictions based on that or actions based on that. That's that's the reason [/INST] you're giving correct answers in novel settings. By manipulating information, you've learned to correct algorithm, You're not doing just some kind of a look up table and there's neighbor search something like </s><s>[INST] that. Let me ask you about A. G. I. What what are some Moonshot ideas you think might make significant progress towards A G. I. Maybe in other ways, what are big blockers that we're missing now. So [/INST] basically I am fairly bullish on our ability to build a G. I. S. Uh basically automated systems that we can interact with and are very human like and we can interact with them in the digital realm or physical realm. Currently it seems most of the models that sort of do these sort of magical tasks are in a text realm. Um I think uh as I mentioned I'm suspicious that text realm is not enough to actually build full understanding of the world. I do actually think you need to go into pixels and understand the physical world and how it works. So I do think that we need to extend these models to consume images and videos and train on a lot more data that is multimodal in that way. Do </s><s>[INST] you think you need to touch the world to understand it also? Well that's [/INST] the big open question I would say in my mind is if you also require the embodiment and the ability to sort of interact with the world run experiments and have a data of that form then you need to go to optimists </s><s>[INST] or something [/INST] like that. And so I would say optimist in some way is like a hedge. Um in A. G. I. Because it seems to me that it's possible that just having data from the internet is not </s><s>[INST] enough if [/INST] that is the case then optimists may lead to a G. I. Uh because optimists would to me there's nothing beyond optimists you have like this humanoid form factor that can actually like do stuff in the world, you can have millions of them interacting with humans and so on. And uh if that doesn't give rise to a G. I. At some point, like I'm not sure what will. Um So from a completeness perspective, I think that's the that's a really good platform, but it's a much harder platform because you are dealing with atoms and you need to actually like build these things and integrate them into society. So I think that path takes longer, </s><s>[INST] but it's [/INST] much more certain. And then there's a path of the internet and just like training these compression models effectively on trying to compress all the internet. And that might also give these agents as well compress </s><s>[INST] the internet, but also interact with the internet, so it's not obvious to me. In fact, I suspect you can reach A. G. I. Without ever entering the physical world and which is a little bit more uh concerning because it might that results in it happening faster. So it just feels like we're like in boiling water we won't know as it's happening. I [/INST] would like to </s><s>[INST] I'm not afraid of a G. I I'm excited about it, There's always concerns, but I would like to know when it happens. [/INST] Yeah. </s><s>[INST] And have like hints about when it happens like a year from now, it will happen, that kind of thing. Yeah, I just feel like in the digital realm it just might happen. [/INST] Yeah, I think all we have available to us because no one has built A G. I. Again so all we have available to us is uh is there enough for child ground on the periphery I would say yes and we have the progress so far which has been very rapid and there are next steps that are available. And so I would say yeah it's quite likely that we'll be interacting with digital entities. </s><s>[INST] How will you know that somebody is [/INST] it's going to be a slow I think it's going to be a slow incremental transition is going to be product based and focused is going to be a top copilot. Getting better and then GPT is helping you right and then these oracles that you can go to with mathematical problems. I think we're on a on the verge of being able to ask very complex questions in chemistry, physics math of these oracles and have them complete solutions. </s><s>[INST] So A G. I. To use primarily focus on intelligence. So consciousness doesn't enter into into it. [/INST] So in my mind consciousness is not a special thing, You will you will figure out and bolt on. I think it's an emergent phenomenon of a large enough and complex enough um generative model sort of So um if you have a complex in the world model that understands the world then it also understands its predicament in the world as being a language model which to me is a form of consciousness or self awareness. And </s><s>[INST] so in order to understand the world deeply, you probably have to integrate yourself into the world and in order to interact with humans and other living beings. Consciousness is a very useful [/INST] tool. I think consciousness is like a modeling insight </s><s>[INST] modeling insight. [/INST] Yeah, it's a you have a powerful enough model of understanding the world that you actually understand, that you are an entity in it. Yeah, </s><s>[INST] but there's also this um perhaps just the narrative we tell ourselves there's a it feels like something to experience the world, the hard problem of consciousness, but that could be just a narrative that we tell ourselves. [/INST] Yeah, I don't think, yeah, I think it will emerge. I think it's going to be something very boring, like we'll be talking to these digital a eyes, they will claim their conscious, they will appear conscious, they will do all the things that you would expect of other humans. And it's going to just be a stalemate. I think there will </s><s>[INST] be a lot of actual fascinating ethical questions like Supreme Court level questions of whether you're allowed to turn off a conscious ai if you're allowed to build a conscious ai uh maybe there would have to be the same kind of the base that you have around um sorry to bring up a political topic, but, you know, abortion, which is the deeper question with abortion is what is life. And the deep question with ai is also what is life and what is conscious and I think that would be very fascinating to bring up. It might become illegal to build systems that are capable the like of such level of intelligence that consciousness would emerge and therefore the capacity to suffer would emerge. And some a system that says, no, please don't kill me. [/INST] That's what the lambda compute the lambda chatbot already told this google engineer, right? Like it was talking about not wanting to die or so on. </s><s>[INST] So that might become illegal to do that, right? Because otherwise, you might have a lot of, a lot of creatures that don't want to die and they will just [/INST] spawn infinity of them on the cluster. </s><s>[INST] And then that might lead to like horrible consequence because then there might be a lot of people that secretly love murder and they will start practicing murder in those systems. I mean, there's just to [/INST] me all of this </s><s>[INST] stuff just brings a beautiful mirror to the human condition and human nature will get to explore it. And that's what like the best of the Supreme Court of all the different debates we have about ideas of what it means to be human. We get to ask those deep questions that we've been asking throughout human history. There has always been the other in human history. Uh we're the good guys and that's the bad guys. And we're going to, you know, throughout human history, let's murder the bad guys. And the same will probably happen with robots that will be the other at first and then we'll get to ask questions. What does it mean to be alive? What does it mean to be conscious? [/INST] And I think there's some Canary in the coal mines even with what we have today. Um And uh you know like for example these there's these like y foods that you can like work with and some people are trying to like this company is going to shut down, but this person really like love their wife. Ooh and like it's trying to like put it somewhere else and like it's not possible. And like I think like definitely people will have feelings towards uh towards these systems because in some sense they are like a mirror of humanity because they are like sort of like a big average of humanity in a way that it's trained </s><s>[INST] but we can that average we can actually watch it. So it's nice to be able to interact with the big average of humanity and do like a search query on it. [/INST] Yeah, that's very fascinating. And we can also of course also like shape it. It's not just a pure average. We can mess with the training data, we can mess with the objective, we can find them in various ways. So we have some um you know, impact on what those systems look like </s><s>[INST] if you want to achieve a G. I. Um And you could have a conversation with her and ask her talk about anything. Maybe ask her a question. What kind of stuff would you would you ask? [/INST] I would have some practical questions in my mind like uh do I or my loved ones really have to die? What can we do about that? </s><s>[INST] Do you think it will answer clearly or would it answer poetically? [/INST] I would expect it to give solutions. I would expect it to be like well I've read all of these textbooks and I know all these things that you've produced and it seems to me like here are the experiments that I think it would be useful to run next and here's some gene therapies that I think would be helpful and uh here are the kinds of experiments you should run. </s><s>[INST] Okay let's go to this thought experiment. Okay imagine that mortality is actually uh like a prerequisite for happiness. So if we become immortal will actually become deeply unhappy and the model is able to know that. So what is this supposed to tell you stupid human about it? Yes you can become immortal but you will become deeply unhappy if the model is if the A. G. I. System Is trying to empathize with you human, what is it supposed to tell you that? Yes you don't have to die but you're really not gonna like it is that is it going to be deeply honest? Like there's a interstellar, what is it the Ai says like humans want 90% honesty, so like you have to pick how honest I want to answer these practical questions. [/INST] I love Interstellar by the way, I think it's like such a sidekick to the entire story but </s><s>[INST] at the [/INST] same time it's like really interesting. It's kind </s><s>[INST] of limited in certain ways, right? [/INST] Yeah it's limited and I think that's totally fine by the way, I don't think uh I think it's fine and plausible to have limited and imperfect A G. I. S. </s><s>[INST] Is that the feature almost as [/INST] an example, like it has a fixed amount of compute on its physical body and it might just be that even though you can have a super amazing mega brain, super intelligent. Ai also can have like you know less intelligent ai so you can deploy in a power efficient way and then they're not perfect, they might make mistakes. </s><s>[INST] No I meant more like say you have infinite compute and it's still good to make mistakes sometimes. Like in order to integrate yourself? Like [/INST] um what </s><s>[INST] is it going back to good Will hunting Robin Williams character says like the human imperfections, that's the good stuff, right? Isn't that isn't that like we don't want perfect, we want flaws in [/INST] part </s><s>[INST] two to form connections with each other because it feels like something you can attach to your feelings too, the flaws in that same way you want ai that's flawed, I don't know, I feel like a perfectionist [/INST] But then you're saying okay </s><s>[INST] but that's not a G I C A G. I would need to be intelligent enough to give answers to humans that humans don't understand. And I think perfect isn't something humans can't understand because even science doesn't give perfect answers. There's always gaps and mysteries and I don't know. I don't know if humans want perfect. [/INST] Yeah. I can imagine just having a conversation with this kind of oracle entity as you'd imagine them. And yeah, maybe it can tell you about you know based on my analysis of human condition. Uh You might not want this and here's some of the things that might but </s><s>[INST] every every dumb human will say, yeah trust me I can give me the truth. I can handle it. [/INST] But that's the beauty like people can choose. So But </s><s>[INST] then I think [/INST] the </s><s>[INST] old marshmallow test with the kids and so on. I feel like too many people like I can't handle the truth. Probably including myself like the deep truth of the human condition. I don't I don't know if I can handle it. Like what if there's some dark, what if we are an alien science experiment and it realizes that what if it had I [/INST] mean I mean this is the matrix. You know all over again. </s><s>[INST] I don't know I would what would I talk about? I don't even Yeah I [/INST] probably I will </s><s>[INST] go with the safer scientific questions at first that have nothing to do with my own personal life immortality, just like about physics and so [/INST] on. </s><s>[INST] Uh to build up like let's see where is that or maybe see if it has a sense of humor. That's another question. Would it be able to presumably in order to, if it understands humans deeply would be able to generate uh [/INST] yeah, to </s><s>[INST] generate humor. [/INST] Yeah, I think that's actually a wonderful benchmark almost. Like is it able, I think that's a really good point basically </s><s>[INST] to make you laugh [/INST] if it's able to be like a very effective stand up comedian that is doing something very interesting computational e I think being funny is extremely hard. </s><s>[INST] Yeah, [/INST] because </s><s>[INST] it's hard in a way like a touring test. The original intent of the touring test is hard because you have to convince humans and there's nothing that's why that's what comedians talk about this. Like there's this is deeply honest because if people can't help but laugh and if they don't laugh that means you're not funny, they laugh. That's funny [/INST] and you're showing you need a lot of knowledge to create to create humor about like the occupation, human condition and so on. And then you need to be clever with it. </s><s>[INST] You mentioned a few movies, you tweeted movies that I've seen five plus times but I'm ready and willing to keep watching Interstellar Gladiator, contact good will hunting the Matrix Lord of the Rings all three avatar, fifth element So on it goes on to make [/INST] it to </s><s>[INST] me and girls. I'm not gonna ask [/INST] about </s><s>[INST] great [/INST] um </s><s>[INST] what are some of the jump onto your memory that you love and why? Like you mentioned the Matrix as as a computer person. Why do you love the Matrix? [/INST] There's so many properties that make it like beautiful interesting. So there's all these philosophical questions but then there's also a G. I. S and there's simulation and it's cool and there's you know the black you know </s><s>[INST] the look of it, the feel [/INST] the feel of it, the action, the bullet time. It was just like innovating in so many ways </s><s>[INST] and then uh good will good will hunting. What do you like that one? Yeah [/INST] I just I really like this uh tortured genius sort of character who's like grappling with whether or not he has like any responsibility or like what to do with this gift that he was given or like how to think about the whole thing and uh </s><s>[INST] there's also a dance between the genius and the the personal like what it means to love another human being and [/INST] there's a lot of things there it's just a beautiful movie </s><s>[INST] and then the fatherly figure the mentor in the in the psychiatrist and it [/INST] like really like uh it messes with you you know there's some movies that just like really messed with you uh on a deep level </s><s>[INST] do you relate to that movie at all? [/INST] No </s><s>[INST] It's not your fault. As I said lord of the rings. That's self explanatory terminator two. Which is interesting. You? Re watch that a lot. Is that better than terminator 1? You like? You like I [/INST] do like terminator one as well. Uh I like Terminator two a little bit more but in terms of like its surface properties </s><s>[INST] do you think china is at all a possibility? [/INST] Yes. </s><s>[INST] Like the actual sort of autonomous weapons system kind of thing. Do you worry about that stuff [/INST] being used </s><s>[INST] for war? [/INST] 100% worry about it. And so the I mean the you know some of these fears of A. Gs and how this will plan out. I mean this will be like very powerful entities probably at some point. And so um for a long time they're going to be tools in the hands of humans. You know people talk about like alignment of A. G. I. S. And how to make the problem is like even humans are not aligned. So how this will be used and what this is going to look like is um yes troubling. So </s><s>[INST] do you think it'll happen slow slowly enough that we'll be able to as a as a human civilization? Think through the problems [/INST] yes that's my hope is that it happens slowly enough and an open enough way where a lot of people can see and participate in it. Just figure out how to deal with this transition I think, which is gonna be interesting. I </s><s>[INST] draw a lot of inspiration from nuclear weapons because I sure thought it would be would be fucked once they develop nuclear weapons. But like it's almost like uh when uh when the systems are not so dangerous, they destroy human civilization, we deploy them and learn the lessons and then we quickly, if it's too dangerous, we quickly quickly we might still deploy but very quickly learned not to use them. So they'll be like this balance achieved. Humans are very clever as a species. It's interesting we exploit the resources as much as we can, but we don't, we avoid destroying ourselves. It seems like, [/INST] well, I don't know about that actually. </s><s>[INST] I hope it continues. [/INST] Um </s><s>[INST] I [/INST] mean I'm definitely concerned about nuclear weapons and so on, not just as a result of the recent conflict even before that, that's probably like my number one concern for humanity. </s><s>[INST] So if humanity destroys itself or destroys, you know, 90% of people, that would be because of nukes. I [/INST] think so. Um it's not even about full destruction. To me, it's bad enough if we reset society that would be like terrible. It would be really bad. And I can't believe we're like so close to it. </s><s>[INST] It's like [/INST] so crazy to me. It </s><s>[INST] feels like we might be a few tweets away from something like that. Yeah, [/INST] basically it's extremely unnerving, but and has been for me for a long time </s><s>[INST] it seems unstable that world leaders just having a bad mood can like um take one step towards a bad direction and it escalates because of a collection of bad moods. It can escalate without being able to stop. [/INST] Yeah. It's just it's a huge amount of power. And then also with the proliferation basically I don't I don't actually really see, I don't actually know what the good outcomes are here. So I'm definitely worried about it a lot. And then a gi is not currently there. But I think at some point will more and more </s><s>[INST] become something [/INST] like it. The danger with A G. I even is that I think it's even like slightly worse in the sense that uh there are good outcomes of A. G. I. And then the bad outcomes are like an epsilon away like a tiny runway. And so I think um capitalism and humanity and so on will drive for the positive ways of using that technology. But then if bad outcomes are just like a tiny like flipping minus sign away. That's a really bad position to be in </s><s>[INST] a tiny perturbation of the system results in the destruction of the human species. It's a weird line to walk. Yeah. [/INST] I think in general it's really weird about like the dynamics of humanity in this explosion we talked about is just like the insane coupling afforded by </s><s>[INST] technology and [/INST] just the instability of the whole dynamical system. I think it just doesn't look good honestly. </s><s>[INST] Yes, that explosion could be destructive and constructive and the probabilities are non zero in both. I'm [/INST] gonna have to I do feel like I have to try to be optimistic and so on. And I think even in this case I still am predominantly optimistic, but there's definitely </s><s>[INST] me too. Do you think will become a multi planetary species? [/INST] Probably, yes, but I don't know if it's a dominant feature of uh future humanity. Uh there might be some people on some planets and so on, but I'm not sure if it's like, yeah, it's like a major player in our culture and so on. We </s><s>[INST] still have to solve the drivers of self destruction here on earth. So just having a backup on MArs is not going to solve the problem. [/INST] So by the way, I love the backup on MArs, I think that's amazing. We should absolutely do that. And I'm so thank you. </s><s>[INST] Uh would you go to MARS [/INST] personally? No, I do like Earth quite a lot. I'll </s><s>[INST] go to MArs I'll go for [/INST] you. I'll </s><s>[INST] tweet at you from [/INST] maybe eventually I would, once it's safe enough, but I don't actually know if it's on my lifetime scale unless I can extend it by a lot. I do think that for example, a lot of people might disappear into virtual realities and stuff like that and I think that could be the major thrust of um sort of the cultural development of humanity if it survives. So it might not be, it's just really hard to work in physical realm and go out there and I think ultimately all your experiences are in your brain and so it's much easier to to disappear into digital realm. And I think people will find them more compelling easier, safer, more interesting. </s><s>[INST] So you're a little bit captivated by virtual reality by the possible worlds, whether it's the metaverse or some other manifestation of [/INST] that. </s><s>[INST] Yeah, it's really interesting. It's uh I'm interested just just talking a lot to Carmack where's the where's the thing that's currently preventing that? [/INST] Yeah. I mean to be clear, I think what's interesting about the future is um it's not that I kind of feel like the variance in the human condition grows, that's the primary thing that's changing. It's not as much the mean of the distribution is like the variance of it. So there will probably be people on MArs and there will be people in VR and there are people here on Earth, it's just like there will be so many more ways of being. And so I kind of feel like I see it as like a spreading out of a human experience. </s><s>[INST] There's something about the internet that allows you to discover those little groups and they gravitate to something about your biology, likes that kind of world that you find each other [/INST] and we'll have trans humanists and then we'll have the Amish and they're gonna, everything is just gonna coexist </s><s>[INST] the cool thing about it because I've interacted with a bunch of internet communities is um they don't know about each other. Like you can have a very happy existence just like having a very close knit community and not knowing about each other. Even you haven't sensed this just having traveled to Ukraine. There's, they don't know so many things about America you like when you travel across the world, I think you experienced this too. There are certain cultures, they're like they have their own thing going on. They don't and so you you can see that happening more and more and more and more in the future. We have little communities. [/INST] Yeah. Yeah, I think so, that seems to be the, that seems to be how it's going right now and I don't see that trend like really reversing. I think people are diverse and they're able to choose their own path in existence. And I sort of like celebrate that. Um and so </s><s>[INST] we spend so much time in the metaverse in the virtual reality or which community are you, are you the physical ist the physical reality [/INST] enjoy or or </s><s>[INST] do you see drawing a lot of pleasure and fulfillment in the digital world. [/INST] Yeah, I think currently the virtual reality is not that compelling. </s><s>[INST] I do [/INST] think it can improve a lot but I don't really know to what extent maybe you know, there's actually like even more exotic things you can think about like neural links or stuff like that. So um currently I kind of see myself as mostly a team human person. I love nature, I love harmony, I love people, I love humanity, I love emotions of humanity. Um and I just want to be like in this like solar punk little Utopia, that's my happy </s><s>[INST] place. My [/INST] happy place is like people, I love thinking about cool problems surrounded by a lush, beautiful dynamic nature and secretly high tech in places that count </s><s>[INST] places that use technology to empower that love for other humans and nature. Yeah, [/INST] I think the technology used very sparingly. I don't love when it sort of gets in the way of humanity in many ways. I like just people being humans in the way we sort of like slightly evolved and prefer I think just by default </s><s>[INST] people kept asking me because they know you love reading, are there particular books that you enjoyed that had an impact on you for silly or for profound reasons that you would recommend. You mentioned the vital question [/INST] Many of course. I think in biology as an example. The vital question is a good one. Anything by Nick Lane really uh life ascending I would say is like a bit more potentially representative is like a summary of a lot of the things he's been talking about. I was very impacted by the selfish gene. I thought that was a really good book that helped me understand altruism as an example and where it comes from. And just realizing that, you know, the selection is on the level of jeans was a huge insight for me at the time and it's sort of like cleared up a lot of things for me. </s><s>[INST] What do you think about the the idea that ideas of the organisms [/INST] Love It? 100%. </s><s>[INST] Yeah. [/INST] Are you able to walk </s><s>[INST] around with that notion for a while that that there is an evolutionary kind of process with ideas as well? [/INST] There absolutely is. There's memes just like jeans and they compete and they live in our brains. It's beautiful. </s><s>[INST] Are we silly humans thinking that we're the organisms? Is it possible that the primary organisms are the ideas? Yeah. [/INST] I would say like the the idea is kind of live in the software of like our civilization in the in the minds and so on. </s><s>[INST] We think as humans that the hardware is the fundamental thing I human is a hardware entity. But it could be the software. [/INST] Right? Yeah. Yeah. I would say like there needs to be some grounding at some point to like a physical reality. The </s><s>[INST] software is the thing. Like is this thing that makes that thing special? Right. [/INST] Yeah. I guess you're right. </s><s>[INST] But then cloning might be exceptionally difficult. Like there might be a deep integration between the software and the hardware in ways we don't quite understand [/INST] from the ocean point of view. Like what makes me special is more like the gang of genes that are writing in my chromosomes I suppose. Right. Like they're the replicating unit, I suppose. And </s><s>[INST] no, but that's just the thing that makes you special. Sure. Well, the reality is what makes you special is your ability to survive based on the software that runs on the hardware that was built by the genes. [/INST] Um, so </s><s>[INST] the software is the thing that makes you survive, not the hardware. It's a [/INST] little bit of both. You know, it's just like a second layer. It's a new second layer that hasn't been there before the brain. They both, they both co exist. </s><s>[INST] But there's also layers of the software. I mean it's, it's not, it's a so abstraction and uh, on top of abstractions, [/INST] but so selfish gene um Nick Lane, I would say sometimes books are like not sufficient. I like to reach for textbooks sometimes. Um, I kind of feel like books are for too much of a general consumption sometime and they just kind of like they're too high up in the level of abstraction and it's not good enough. So I like textbooks. I like the cell. I think the cell was pretty cool. That's why also I like the writing of nick Lane is because he's pretty willing to step one level down and he doesn't, </s><s>[INST] uh, yeah, he's [/INST] sort of he's willing to go there, but he's also willing to sort of be throughout the stack, so he'll go down to a lot of detail. But then he will come back up and I think he has a basically I really appreciate that. That's </s><s>[INST] why I love college, early college, even high school, just textbooks on the basics of computer science and mathematics, of biology and chemistry. Those are they condensed down like uh it's sufficiently general that you can understand both the philosophy and the details, but also like you get homework problems and you get to play with it as much as you would if you weren't Yeah, programming stuff. And [/INST] then I'm also suspicious of textbooks honestly, because as an example in deep learning, there's no like amazing textbooks and the field is changing very quickly. I imagine the same is true. And say uh synthetic biology and so on. These books, like the cell are kind of outdated, there's still high level like what is the actual real source of truth is people in wet labs working with cells, you know, sequencing genomes and yeah, actually working with working with it and uh I don't have that much exposure to that or what that looks like. So I still don't fully, I'm reading through the cell and it's kind of interesting and I'm learning but it's still not sufficient, I would say in terms of understanding what </s><s>[INST] it's a clean summarization of the mainstream narrative. Yeah. But you have to learn that before you break [/INST] out </s><s>[INST] towards the cutting edge. [/INST] Yeah. What is the actual process of working with these cells and growing them and incubating them And you know, it's kind of like a massive cooking recipes of making sure yourself slaves and proliferate and then you're sequencing them running experiments and just how that works. I think it's kind of like the source of truth of at the end of the day what's really useful in terms of creating therapies and so on. </s><s>[INST] Yeah. I wonder what in the future ai textbooks will be because you know, there's an artificial intelligence, a modern approach actually haven't read if it's come out the recent version. The recent there's been a recent addition. I also saw there's a science of deep learning book. I'm waiting for textbooks that worth recommending worth reading. [/INST] It's </s><s>[INST] tricky because it's like papers and code. Code code. [/INST] Honestly I find papers are quite good especially like the appendix appendix of any paper as well. It's like it's like the most detail you can </s><s>[INST] have. It doesn't have to be cohesive connected to anything else. You just described a very specific way you saw the particular thing. Yeah. [/INST] Many times papers can be actually quite readable. Not always but sometimes the introduction in the abstract is readable even for someone outside of the field not this is not always true. And sometimes I think unfortunately scientists use complex terms, even when it's not necessary. I think that's harmful? I think there's no reason for </s><s>[INST] that. And papers sometimes are longer than they need to be in the in the parts that don't matter. Yeah, appendix would be long. But then the paper itself, you know, look at Einstein, make it simple. Yeah, [/INST] but certainly I've come across papers I would say and say like synthetic biology or something that I thought were quite readable for the abstract and introduction and then you're reading the rest of it and you don't fully understand, but you kind of are getting a gist and I think it's cool </s><s>[INST] what advice you give advice to folks interested in machine learning and research, but in general life, advice to a young person, high school, early college about how to have a career they can be proud of or a life they can be proud of. [/INST] Yeah, I think I'm very hesitant to give general advice. I think it's really hard. I've mentioned, like some of the stuff I've mentioned is fairly general. I think like, focus on just the amount of work you're spending on like a thing, compare yourself only to yourself not </s><s>[INST] to others. I [/INST] think those are fairly general. </s><s>[INST] How do you pick the thing, [/INST] you just have like a deep interest in something uh or like try to like find the art max over, Like the things that you're interested in </s><s>[INST] art max at that moment and stick with it, how do you not get distracted and switch to another thing [/INST] you can, if you like um </s><s>[INST] if you do an arg max repeatedly every week, [/INST] doesn't converge, </s><s>[INST] doesn't you [/INST] can like low pass filter yourself in terms of like what has consistently been true for you? Um But yeah, I definitely see how it can be hard, but I would say like you're going to work the hardest on the thing that you care about the most so low pass filter yourself and really introspect in your past, what are the things that give you energy and what are the things that take energy away from you? Concrete examples and usually from those concrete examples, sometimes parents can emerge, I like I like it when things look like this one of these positions, so </s><s>[INST] that's not necessarily the field, but the kind of stuff you're doing in a particular field. So for you, it seems like you were energized by implementing stuff, building actual things. [/INST] Yeah. Being low level learning, and then also uh communicating so that others can go through the same realizations and shortening that gap um because I usually have to do way too much work to understand the thing and then I'm like, okay, this is actually like okay, I think I get it and like why was it so much </s><s>[INST] work, [/INST] it should have been much less work and that gives me a lot of frustration and that's why I sometimes go teach. </s><s>[INST] So aside from the teaching you're doing now, putting out videos aside from a potential uh Godfather part two, uh would the A. G. I. At Tesla and beyond. What does the future of apathy hold? Have you figured that out yet or? No? I mean uh as you see through the fog of war, that is all of our future. Um do you, do you start seeing silhouettes of what that possible future could look like? [/INST] The consistent thing I've been always interested in for me at least is is A. I. And um that's probably what I'm spending my rest of my life on because I just care about it a lot and I actually care about like many other problems as well, like say aging, which I basically view as disease and I care about that as well. But I don't think it's a good idea to go after it specifically. I don't actually think that humans will be able to come up with the answer. I think the correct thing to do is to ignore those problems and you solve AI and then use that to solve everything else. And I think there's a chance that this will work. I think it's a very high chance and uh that's kind of like the the way I'm betting at </s><s>[INST] least. So when you think about ai are you interested in all kinds of applications, all kinds of domains and any domain you focus on will allow you to get insights to. The big problem with a guy [/INST] for me is the ultimate mental problem. I don't want to work on any one specific problem. There's too many problems. So how can you work on all problems simultaneously? You solve the meta problem, which to me is just intelligence and how do you automated </s><s>[INST] is their cool small projects like archive sanity and and so on that you're thinking about the the world, the ml world can anticipate. [/INST] There's someone always like some fun side projects. Archive sanity is one basically like there's way too many archive papers. How can I organize it and recommend papers and so on. I transcribed all of your podcasts. </s><s>[INST] What did you learn from that experience? Uh from transcribing the process of like you like consuming audio books and podcasts and so on. And here's a process that achieves um closer to human level performance and annotation. [/INST] Yeah, well I definitely was like surprised that transcription with opening eyes, Whisper was working so well compared to what I'm familiar with from Siri and like a few other systems, I guess it works so well. And uh that's what gave me some energy to like try it out. And I thought it could be fun to random podcasts. It's kind of not obvious to me why whisper is so much better compared to anything else because I feel like there should be a lot of incentive for a lot of companies to produce transcription systems and that they've done so over a long time. Whisper is not a super exotic model. It's a transformer, It takes mel spectrograms and you know, just outputs tokens of texts, not crazy uh model and everything has been around for a long time. I'm not 100% sure why. </s><s>[INST] It's not obvious to me either. It makes me feel like I'm missing something. [/INST] Something. Yeah, </s><s>[INST] because there's a huge even google and so on youtube transcription. Yeah, it's unclear but some of it is also integrating into a bigger system that so the user interface, how is deployed and all that kind of stuff. Maybe running it as an independent thing is much easier, like an order of magnitude easier than deploying to a large integrated system like Youtube transcription or um anything like meetings like zoom has trans transcription that's kind of crappy. But creating interface where detects the different individual speakers, it's able to um display it in compelling ways running in real time, all that kind of stuff. Maybe that's difficult. But that's the only explanation I have because like um I'm currently paying quite a bit for human transcription, human captions annotation and like it seems like uh there's a huge incentive to automate that it's very confusing [/INST] and I think, I mean I don't know if you look at some of the whisper transcripts but they're quite good. They're good, </s><s>[INST] especially in tricky cases. I've seen uh whispers performance on like super tricky cases and it does incredibly well. So I don't know, a podcast is pretty simple. It's like high quality audio and you're speaking usually pretty clearly. And so I don't know it uh I don't know what open they eyes plans are either, [/INST] but there's always like fun, fun projects basically. And stable diffusion also is opening up a huge amount of experimentation, I would say in the visual realm and generating images and videos and movies. And so that's going to be pretty crazy. That's going to, that's going to almost certainly work. And it's going to be really interesting when the cost of content creation is going to fall to zero. Used to need a painter for a few months to paint a thing and now it's going to be speak to your phone to get your video. </s><s>[INST] Hollywood will start using that to generate scenes Um which completely opens up. Yeah. So you can make a like a movie like Avatar eventually for under $1 million, much [/INST] less. Maybe just by talking to your phone? I mean I know it sounds kind of </s><s>[INST] crazy and then there'd be some voting mechanism like how do you have, like, would there be a show on netflix that's generated completely automated lee. So [/INST] potentially. Yeah. And what does it look like? Also when you can generate on demand and it's uh and there's infinity of </s><s>[INST] it. Yeah. Oh man all the synthetic content. I mean it's humbling because we treat ourselves as special for being able to generate art and ideas and all that kind of stuff if that can be done in an automated way by Ai. Yeah. [/INST] I think it's fascinating to me how these uh the predictions of Ai and what is going to look like and it's going to be capable of are completely inverted and wrong and sci fi of 50s and 60s was just like totally not right. They imagined a i is like super calculating here improvers and we're getting things that can talk to you about emotions. They can do art. It's just like weird. </s><s>[INST] Are you excited about that future? Just a eyes like hybrid systems, heterogeneous systems of humans. And Ai is talking about emotions. Netflix and Children ai system. That word. The netflix thing you watch is also generated by ai [/INST] I think it's it's going to be interesting for sure. And I think I'm cautiously optimistic but it's not it's not obvious. Well </s><s>[INST] the sad thing is your brain and mine developed in a time where before twitter before the before the internet. So I wonder people that are born inside of it might have a different experience. Um Like I maybe you can will still resisted. Uh and the people born now will not. [/INST] Well I do feel like humans are extremely malleable. </s><s>[INST] Yeah [/INST] and </s><s>[INST] you're probably right. What is the meaning of life andre [/INST] we we talked about sort </s><s>[INST] of the universe having a conversation with us humans or with the systems we create to try to answer for the university for the creator of the universe to notice us, we're trying to create systems that are allowed enough. Just answer back. [/INST] I don't know if that's the meaning of life, that's like meaning of life for some people, the first level answer I would say is anyone can choose their own meaning of life because we are conscious entity and it's beautiful number one. But I do think that like a deeper meaning of life if someone is interested is along the lines of like what the hell is all this? And like why? And if you look at the into fundamental physics and quantum field theory and the standard model, they're like very complicated and There's this like you know, 19 free parameters, parameters of our universe and like what's going on with all this stuff and why is it here? And can I hack it, can I work with it? Is there a message for me? AM I supposed to create a message? And so I think there's some fundamental answers there but I think there's actually even like you can't actually really make dent in dose without more time. And so to me also there's a big question around just getting more time honestly. Yeah, that's kinda like what I think about quite a bit as well. </s><s>[INST] So kind of the ultimate or at least first way to sneak up to the why question is to try to escape. Uh, the system, the universe and then for that, you sort of backtrack and say, okay for that, that's gonna be take a very long time. So the white question boils down from an engineering perspective to how do we extend? [/INST] Yeah, I think that's the question, number one, practically speaking, because you can't, you're not going to calculate the answer to the deeper questions in time you have and </s><s>[INST] that could be extending your own lifetime or extending just a lifetime of human civilization [/INST] of whoever wants to. Not many people might not want that, but I think people who do want that, I think, I think it's probably possible. Uh, and I don't think, I don't know that people fully realize this. I kind of feel like people think of death as an inevitability, but at the end of the day, this is a physical system, Some things go wrong. Uh, it makes sense why things like this happen evolutionarily speaking. And there's most certainly interventions that mitigate it. </s><s>[INST] That would be interesting if death is eventually looked at as as a fascinating thing that used to happen to humans. I [/INST] don't think it's unlikely. I think it's, I think it's likely </s><s>[INST] and it's up to our imagination to try to predict what the world without death looks like. It's hard to I think the values will completely change could [/INST] be, I don't I don't really buy all these ideas that without that there's no meaning, there's nothing is I don't, intuitively by all those arguments. I think there's plenty of meaning, plenty of things to learn. They're interesting, exciting. I want to know, I want to calculate, I want to improve the condition of all the humans and organisms that are alive. </s><s>[INST] The way we find meaning might change. There is a lot of humans probably including myself that finds meaning in the finite nous of things, but that doesn't mean that's the only source of meaning. [/INST] Yeah, I do think many people will will go with that, which I think is great. I love the idea that people can just choose their own adventure. Like you, you are born as a conscious free entity by default. I like to think and you have your inalienable rights for life </s><s>[INST] in the pursuit of happiness. [/INST] I don't know </s><s>[INST] that in the nature of the landscape of happiness and [/INST] you can choose your own adventure mostly. And that's not fully true, but </s><s>[INST] I still pretty sure I'm an NBC. But um and NBC can't know it's an NBC. There could be different degrees and levels of consciousness. I don't think there's a more beautiful way tended andre you're an incredible person. I'm really honored. You would talk with me everything you've done for the machine learning world for the ai world [/INST] to </s><s>[INST] just inspire people to educate millions of people. It's been it's been great and I can't wait to see what you do next. It's been an honor, man. Thank you so much for talking today. [/INST] Thank you. </s><s>[INST] Thanks for listening to this conversation with Andre Carpathia. To support this podcast. Please check out our sponsors in the description and now let me leave you with some words from Samuel Karlin. The purpose of models is not to fit the data, but to sharpen the questions. Thanks for listening and hope to see you next time. [/INST] </s> |
<s>[INST] cannot change all [/INST] of the selling season. So the phones are going to launch on time and every technology is to be there. The engineer needs to work as </s><s>[INST] one and we do [/INST] have that qualcomm. The other thing you have to have incredible uh, discipline because um, you know, those are very complex systems. So in one way you have to design with </s><s>[INST] quality [/INST] because in many cases we're going to be ramping production and even before we have the silicon back and you have to rely on on </s><s>[INST] our simulation models [/INST] and you have to rely on on the fact that you design uh, for commercial applications that takes, that takes a while </s><s>[INST] to build [/INST] and uh, and you know, it's probably being the history of a semiconductor business at Qualcomm. So you mean </s><s>[INST] like the framework of how [/INST] many people can use </s><s>[INST] simulation salt and all that [/INST] kind of stuff to </s><s>[INST] build the thing with [/INST] a hard deadline that you might not even get back </s><s>[INST] from like manufacture [/INST] before you're not allowed to have any mistakes. No wonder our name is quality communications. Oh, I never even thought about the quality part </s><s>[INST] quality. It's a [/INST] quality and there's a bar </s><s>[INST] that's high and [/INST] you're not allowed to mess up. I mean to me as an engineer that's </s><s>[INST] exciting. Hard [/INST] deadlines. Um, no room for </s><s>[INST] mistakes. I love, [/INST] it's super stressful, but </s><s>[INST] I love it. [/INST] Um so there's a couple of other small companies </s><s>[INST] um [/INST] called google and Apple. So google </s><s>[INST] is not using its own [/INST] Chips for the Pixel six. Apple using is its </s><s>[INST] own, how [/INST] does qualcomm, I'll </s><s>[INST] compete google [/INST] and </s><s>[INST] Apple, how does it beat them? [/INST] We don't have to out compete google. Actually, if </s><s>[INST] you look at our mobile [/INST] strategy today and one thing I was very clear when I became </s><s>[INST] Ceo, I think there's all the [/INST] confusion in the market are mobile strategy is very clear, we're focused of making snapdragon synonymous with premium </s><s>[INST] android experience, [/INST] that's what snapdragon is, android phone of the people. Yes. Uh I </s><s>[INST] just have a love for android and I'm constantly talking trash to iphone people, sorry, go ahead [/INST] premium android </s><s>[INST] experiences. So we do [/INST] produce, we do produce snapdragon and multiple tier for every price point, but you know, every year you mentioned the HN one and every year will provide, you know, the flagship product and then, and then the other </s><s>[INST] series that is trying to [/INST] get the best of every </s><s>[INST] possible technology [/INST] at that time and it's really focused on enabling the android ecosystem. So I'll give an example. So you asked me the question how to compete with google, it's not about competing with google, we're the number one enabler of the google android ecosystem and the the largest, largest </s><s>[INST] and number one [/INST] customer, there is actually a Samsung and if you look what happened to Samsung Samsung um always had </s><s>[INST] since [/INST] I began my, began my relationship with them because they always had their own chip, they </s><s>[INST] always had their own chip. [/INST] And </s><s>[INST] um, and if you'll [/INST] Just look at what happened right now with the galaxy has 22 to just launch, you know, they used to balance their, their business about 50% qualcomm. They will get the most advanced markets like the United States and China and Japan and Korea. They will sign a qualcomm and then they have their own chip for the markets that they would, would be like more emerging markets, open markets markets that they have a control on the channel because they sell a lot of appliances and other things. If you look what happened right now, Gs 22 75% is qualcomm and uh, and then the </s><s>[INST] next [/INST] large O E M S and android IQ system or the chinese ones, Companies like Xiaomi, uh, one of the fastest growing, uh, it was number one in europe at some point </s><s>[INST] last year [/INST] Then followed by opal and one plus and Vivo. So those are some of the largest quicken customers and they actually drive the android ecosystem and that's your mobile strategy and fully aligned with google and uh, and it's working and I </s><s>[INST] was, you know, I'm not [/INST] to get into a lot of the investor conversation but we're also happy we became uh, in a beneficiary of the shifts that we </s><s>[INST] shot in the marketplace as [/INST] huawei became a smaller O. E. M. As a result of the sanctions. Um </s><s>[INST] we saw the rise [/INST] of a lot of the other </s><s>[INST] ohms from [/INST] china, especially for china domestic market. </s><s>[INST] Xiaomi oppo [/INST] vivo. They </s><s>[INST] moved to the premium category [/INST] and they're all doing that with qualcomm. So, </s><s>[INST] uh we're actually [/INST] very fortunate and happy with the position we are </s><s>[INST] in mobile business. [/INST] We </s><s>[INST] do have an Apple relationship, We [/INST] provide modem </s><s>[INST] technology to Apple. Um it's a multiyear [/INST] relationship. Apple has been very public that they are </s><s>[INST] uh [/INST] investing to develop </s><s>[INST] their own modem. [/INST] Uh, but the qualcomm </s><s>[INST] strategy has, [/INST] has been clear, you know, we're </s><s>[INST] really focused on snapdragon. [/INST] You know not our mobile strategy </s><s>[INST] is not defined [/INST] by providing a cellular modem </s><s>[INST] to apple or mobile [/INST] strategies that we just talked </s><s>[INST] about. It is [/INST] about the unique thing of snapdragon that has every single technology integrated into a single S. O. C. And uh and that's what we're doing and uh focusing on the android ecosystem. I don't know if I can </s><s>[INST] ask you this kind of question. [/INST] It's like picking your Children or something like this, but </s><s>[INST] what [/INST] smartphone with a snapdragon you mentioned. Samsung Galaxy? S 22 1 plus those phones. I personally really enjoy. What phone do you currently use or do you have multiple phones. You just, I do have multiple phones, but I do use Galaxy S. 22 That's your favorite one. Alright, well you heard </s><s>[INST] It here 1st folks. [/INST] Okay, so excellent. Can qualcomm </s><s>[INST] also let's take a brief step [/INST] away from mobile and take on intel and </s> |
<s>[INST] What kind of data augmentation is useful for that non contrasted learning method? Are we talking about? Does that not matter that much or it seems like a very important part of the step? Yeah. How you generate the images that are similar but sufficiently different. [/INST] Yeah, that's right. It's an important step and it's also an annoying step because you need to have that knowledge of what the documentation you can do that do not change the nature of the of the object. </s><s>[INST] And so [/INST] the standard scenario which, you know, a lot of people working in this area are using is you use the type of distortion. So so basically you do geometry distortion. So one basically just shifts the image a little bit is called crapping. Another one kind of changes the scale a little bit. Another one kind of rotate it. Another one changes the colors. You know, you can do a shift in color balance or something like that, um saturation. Another one sort of blurs it. Another one as noise. So you have like a catalog of kind of standard things and people try to use the same ones for different algorithms so that they can compare. Um but some algorithms, some self supervised algorithm actually can deal with much bigger, like more aggressive data augmentation and some don't. So that kinda makes the whole thing difficult, but but that's the kind of distortions we're talking about. And and so you you you you trained with those distortions and then uh you you chop off the last layer, a couple layers of the of the network and you use the representation as input to a classifier. Are you trying to classifier um on the internet or whatever and measure the performance? And interestingly enough, the methods that are really good at eliminating the information that is irrelevant, which is the distortions between those images. Do a good job at eliminating it. And as a consequence you cannot use the representations in those systems for things like object detection and localization because that information is gone. So the type of the documentation you need to do depends on the task you want eventually the system to to solve and the type of the documentation standard documentation that we use today are only appropriate for object recognition or image classification. They're not appropriate for things like, </s><s>[INST] can you help me out understand what, why the localization? So you're saying it's just not good at the negative uh like classifying the negative. So that's why it can't be used for the localization. [/INST] No, it's just that you trained the system. You know, you you you give it an image and then you give it the same image shifted and scaled. You tell you that's the same image. So the system basically is trained to eliminate the information about position and size. So now now you want to use </s><s>[INST] that, oh, [/INST] like literally where an object is and what </s><s>[INST] it's like a bounding box that they'd be able to actually okay, it can still find, it can still find the object in the image is just not very good at finding the exact boundaries of that object. Interesting, interesting. Which you know that that's an interesting sort of philosophical question. How important how important is object localization? Anyway, we're like obsessed by measuring like image segmentation, obsessed by measuring perfectly knowing the boundaries of objects when arguably that's not that essential to an understanding what are the contents of the scene? [/INST] On the other hand, I think evolutionarily the first vision systems in animals were basically all about localization, very little about recognition. And in the human brain you have to separate pathways for recognizing the nature of the scene of an object and localizing objects. So, you use the first pathway called the ventral pathway for you. No telling what you're looking at the other path with the dorsal pathway is used for navigation for grasping for everything else. And you know, basically a lot of the things you need for survival are localization and detection </s><s>[INST] is similarity learning or contrast of learning are these non contrast of methods the same as understanding something just because you know, distorted cat is the same as a non distorted cat. Does that mean you understand what it means to be a cat [/INST] to some extent? I mean it's a superficial understanding obviously, </s><s>[INST] but like what is the ceiling of this method do you think? Is this just one trick on the path to doing self supervised learning? Can we go really, really [/INST] far? I think we can go really far. So if we figure out how to use techniques of that type perhaps very different. But you know, the signature to training system from from video to do video prediction essentially. Um I think we'll have a path um you know, towards uh you know, I wouldn't say unlimited but but a path towards some level of um you know, physical common sense in machines. And I also think that that ability to learn how the world works from a sort of high throughput channel, like like vision is a necessary step towards uh sort of real artificial intelligence. In other words, I believe in ground intelligence. I don't think we can train a machine to be intelligent purely from text because I think the amount of information about the world that's contained in text is tiny compared to what we need to know. So for example, let's and you know, people have attempted to do this for for 30 years, right? The side project and things like that. Right of basically kind of writing down all the facts that are known and hoping that some some sort of common sense will emerge. I think it's basically hopeless. But let me take an example. You take an object. I described the situation to you. I take an object, I put it on the table and I pushed the table. It's completely obvious to you that the object will be pushed with the table, right? Because it's sitting on it. Um there's no text in the world, I believe that explains this. And so she trained a machine as powerful as it could be. You know, your gpt 5000 uh or whatever it is, it's never going to learn about this. Um that information is just is not present in any text. </s> |
<s>[INST] it's a little too abstract. There's useful elements to it in that it helps us understand our own ethical codes, humans. So even just as a thought experiment, if you imagine that an A. G. I. System is here today, how would we program? It is a kind of nice thought experiment of constructing how should we have a [/INST] law, </s><s>[INST] have a system of laws for us humans. It's just a nice practical tool and I think there's echoes of that idea too in the ai systems we have today, they don't have to be that intelligent like autonomous vehicles. These things start creeping in that there were thinking about but certainly they shouldn't be framed as as hell. Yeah. Looking back, what is the most, I'm sorry if it's a silly question, but what is the most beautiful or surprising idea in deep learning or ai in general that you've ever come across sort of personally when you said back and I just had this kind of uh that's pretty cool moment. That's nice surprising. [/INST] I don't know if it's an idea rather than sort of empirical fact. The fact that you can build gigantic neural nets, trained, trained them on relatively small amounts of data relatively with two guest agreed in the center that it actually works breaks everything you read in every textbook right? Every pre deployment text book that told you you need to have fewer parameters and you have data samples, um, you know, if you have a non convex objective function, you have no guarantee of convergence, you know, all those things that you read in textbooks and they tell you stay away from this and they're all wrong. </s><s>[INST] The number of parameters non convex and somehow, which is very relative to the number of parameters, data. It's able to learn anything. Does that still surprise you today? [/INST] Well, it was kind of obvious to me before I knew anything that this is a good idea. And then it became surprising that it worked because I started reading those textbooks. Okay, </s><s>[INST] so, okay, you know, we talked to the intuition of why I was obviously if you remember. [/INST] Well, okay, so the intuition was it's it's sort of like, you know those people in the late 19th century who proved that heavier than air flight was impossible, right? And of course you have birds, right? They do fly. And so on the face of it, it's obviously wrong as an empirical question. Right? So we have the same kind of thing that you know, the we know that the brain works, we don't know how, but we know it works and we know it's a large network of neurons and interaction. And learning takes place by changing the connection. So kind of getting this level of inspiration without covering the details, but sort of trying to derive basic principles, you know, that kind of gives you a clue as to which direction to go. There's also the idea somehow that I've been convinced of since I was an undergrad, that even before that intelligence is inseparable from learning. So you the idea somehow that you can create an intelligent machine by basically programming for me, it was a non starter for, you know, from the start, every intelligent entity that we know about arrives at this intelligence through learning. So learning, you know, machine money was completely obvious path. Uh also because I'm lazy, so you know, </s><s>[INST] the automated, basically everything and learning is the automation of intelligence? So do you think? So what is learning then? What falls under learning? Because do you think of reasoning as learning [/INST] what our reasoning is? Certainly a consequence of learning as well, just like other functions of the brain. The big question about reasoning is how do you make reason incomparable with gradient based learning? </s><s>[INST] Do you think neural networks can be made to reason? [/INST] Yes, there's no question about that. Again, we have a good example, right? The question is how So the question is how much prior structure you have to put in the neural net so that something like human reasoning would emerge from it, you know, from running. Another question is all of our kind of model of what reasoning is that are based on logic are discrete and and are therefore incompatible with great and based learning. And I was very strong believer in this idea of granting based learning. I don't believe that other types of learning that don't use kind of gradient information if you want. </s><s>[INST] So you don't like discrete mathematics. You don't like anything discrete. [/INST] Well that's, it's not that I don't like it, it's just that it's invaluable. We're learning and I'm a big fan of learning. Right? So in fact that's perhaps one reason why deep learning has been kind of looked at with suspicion by a lot of computer scientists because the matter is very different the matter you use for deep learning, you know, we kind of has more to do with, you know, cybernetics, uh kind of Matthew doing electrical engineering than the kind of Matthew doing computer science and and you know nothing in in machine learning is exact right. Computer science is all about sort of you know, obviously compulsive attention to details of like you know every index has to be right and you can prove that an algorithm is correct, right? Machine learning is the science of sloppiness really, </s><s>[INST] That's beautiful. So okay, maybe let's feel around in the dark of what is a neural network that reasons or a system that his works with continuous functions that's able to do ah build knowledge. However we think about reasoning, build on previous knowledge, build on extra knowledge, create new knowledge, generalist outside of any training set ever built. What does that look like if Yeah, maybe you have inklings of thoughts of what that might look like? [/INST] Well, yeah, I mean yes or no, if I had precise ideas about this, I think, you know, we'll be building it right now but there are people working on this or whose main research interest is actually exactly that. Right? So what you need to have is a working memory. So you need to have some device, if you want some subsystem they can store a relatively large number of factual episodic information for, you know, a reasonable amount of time. So you, you know, in the in the brain for example the kind of three main types of memory. One is the sort of memory of the state of your cortex And that sort of disappears within 20 seconds. You can't remember things for more than about 20 seconds or a minute if you don't have any other form of memory. The second type of memory, which is longer term is still short term is the hippocampus. So you can, you know, you came into this building, you remember whether where the exit is where the elevators are? Um you have some map of that building that's stored in your hippocampus, you might remember something about what I said, you know, a few minutes ago already has been raised but you know, but that would be in your hippocampus uh and then the longer term memory is in the synapse the synapses, right? Um so what you need, if you want a system that's capable of reasoning is that you want the hippocampus like thing. Right? And that's what people have tried to do with memory networks and you know military machines and stuff like that. Right? And now with transformers which have sort of a memory in there kind of self attention system, you can think of it this way. So so that's one element you need, another thing you need is some sort of network that can access his memory, get an information back and then kind of crunch on it and then do this creatively multiple times because a chain of reasoning is the process by which you you you can you update your knowledge about the state of the world about you know what's going to happen etcetera. And that that has to be this sort of recurrent operation basically. </s> |
<s>[INST] it's not just like a philosophy degree that asked the big questions and it's not uh like biology and the path to be going to medical school or something like that, which is overly pragmatic, not overly is um, is very [/INST] pragmatic. </s><s>[INST] Physics is a good combination of the two. [/INST] Yeah. At least for me it made sense and I was good at it. I liked it. Um, yeah, I mean it wasn't like I did an immense amount of soul searching to choose it or anything. It just was like this. It made the most sense. I mean you have to make this decision in the UK age 17, which is crazy because you know in us you go the first year you do a bunch of stuff right? And then you choose your major. Um, </s><s>[INST] I think the first few years of college you focus on the drugs and only as you get closer to the end. Do you start to think, oh ship, this wasn't about that. And I'm uh, I owe the government a lot of money. Um How many alien civilizations are out there when you, when you looked up at the stars with your [/INST] mom </s><s>[INST] and you were counting them. What's your mom think about the number of alien civilizations? [/INST] I don't know, I would imagine she would take the viewpoint of, you know, she's pretty humble and she knows how many, you know, there's a huge number of potential spawn sites out there. So she would </s><s>[INST] spawn sites, [/INST] sites. Yeah. You know, </s><s>[INST] Yeah, [/INST] polito P. A. We spawned on earth, you know, it's </s><s>[INST] yeah, spawn sites. What does that feel weird to say spawn? Because it makes me feel like it's um there's only one source of life and spawning in different locations. That's why the word spawn because like it feels like life that originate on Earth really originated here, [/INST] right? It is it is in a unique to this particular. Yeah, I mean, but I don't, in my mind that doesn't exclude, you know, the completely different forms of life and different biochemical soups can't also spawn, but I guess it implies that there's some spark that is, </s><s>[INST] you know, [/INST] I kind of like the idea of, you know, </s><s>[INST] and then I get to think about re spawning like after it dies? Like what happens if life on earth ends? Is it is it going to restart again? [/INST] Probably not. It depends </s><s>[INST] depends [/INST] on the type of, you know what, what's the thing that kills, kills it off, right? If it's a paperclip maximizing not, you know, for the for the example, but you know, some kind of very self replicating, you know, high on the capability is very low on the wisdom type thing. So whether that's you know, gray goo, green goo, you know, like nanobots or just a shitty misaligned ai that thinks it needs to turn everything into paperclips. Um You know, if it's something like that then it's gonna be very hard for life, you know, complex life because by definition, you know, a paperclip maximizes the ultimate instance creation of monarchy, deeply low complexity over optimization on a single thing, sacrificing everything else turning the whole world into. </s><s>[INST] Although something tells me like if we actually take a paperclip maximizing it destroys everything, it's a really dumb system that just envelopes the whole of Earth. [/INST] Yeah, </s><s>[INST] I didn't I didn't know that part but Okay, great. [/INST] That's it </s><s>[INST] becomes a multi planetary paper club maximize. Er [/INST] Well it just it just propagates. I mean it depends whether it figures out how to jump the vacuum gap. Um But again, I mean this is also because it's a hypothetical thought experiment which I think doesn't actually have much practical application to the ai safety problem, but it's just a fun thing to play around with but if by definition it is maximally intelligent, which means it is maximally good at navigating the environment around it in order to achieve its goal but extremely bad at choosing goals in the first place. So again we're talking on this orthogonal itty thing, right? It's very low on wisdom but very high on capability, then it will figure out how to jump the vacuum gap between planets and stars and so on and thus just turn every atom it gets its hands on into paperclips. </s><s>[INST] Yeah, by the way, for people who [/INST] which is maximum morality by the way, that's what morality is. </s><s>[INST] But does not mean that morality is necessarily all about maximizing paperclips. In that case it is. So for people who don't know, that's just a thought experiment example of an Ai system. That's very that has a goal and is willing to do anything to accomplish that goal. Including destroying all life on Earth and all human life and all of consciousness in the universe. You know, for the goal of producing a maximum number of paperclips, okay? [/INST] Or whatever its optimization function was that it was set up. But </s><s>[INST] don't you think could be [/INST] making recreating Lexus, maybe I'll tile the universe and lex </s><s>[INST] go on Like this idea [/INST] that's better. That's more interesting than paper </s><s>[INST] that could be infinitely optimal for [/INST] it's still a bad thing because it's permanently capping what the universe could ever be. It's like that's that's it's </s><s>[INST] or achieving the optimal that the universe could ever achieve. But that's different people have different perspectives. But don't you think within the paperclip world that would emerge just like in the zeros and ones that make up a computer that would emerge beautiful complexities like it won't suppress, you know, as you scale to multiple planets and throughout they will they will emerge these little worlds that on top of the fabric of maximizing paperclips, there will be that would emerge like little societies of of paperclips. [/INST] Well then we're not with them, we're not describing a paperclip maximizing anymore because by the like if you think of what a paperclip is, it is literally just a piece of bent iron, right? So if it's maximizing that throughout the universe, it's taking every atom, it gets its hands on into somehow turning it into iron or steel and then bending it into that shape and then done and on by definition like paper clips. There is no there is no way for well, okay, so you're saying that paperclips somehow will just emerge and create gravity or something? </s> |
<s>[INST] Don't you think that's a good thing for the world that ai could potentially [/INST] I do. Absolutely. And furthermore, I want, you know, I'm pushing 80 now by the time I need help like that. I also wanted to roll itself in a corner and shut the funk </s><s>[INST] up. Let me let me linger on that point. Do you really [/INST] though? Yeah, I do. Here's </s><s>[INST] what you wanted to push back a little bit [/INST] a little but I have watched my friends go through the whole issue around having help in the house and some of them have been very lucky and had fabulous help and some of them have had people in the house who want to keep the television going on all day, who want to talk on their phones all day? No, </s><s>[INST] so basically [/INST] just roll yourself in the corner. </s><s>[INST] Unfortunately as humans, when we're assistance, we we care, we're still, even when we're assisting others, we care about ourselves more. And so you create more frustration and robot ai assistant can really optimize the experience for you. I was just speaking to the point, you actually bring up a very, very, very good point. But I was speaking to the fact that as humans are a little complicated, that we don't necessarily want a perfect servant. I don't maybe you disagree with that, but there's a, I think uh there's a push and pull with humans a little attention. A little mystery that of course that's really difficult for you to get right. But I do sense especially in today with social media, that people are getting more and more lonely, even young folks and sometimes especially young folks that loneliness, there's a longing for connection and a I can help alleviate some of that loneliness. Some just somebody who listens like in person, [/INST] um, that's so we </s><s>[INST] speak, so, so, so, so to speak. Yeah, so to speak. Yeah, that to me is really exciting. But so if we look at that, that level of intelligence, which is exceptionally difficult to achieve actually as the singularity or whatever. That's the human level bar that people have dreamt of that to touring dreamt of it. Uh He had a date timeline. Do you have how of your own timeline evolved on? [/INST] I don't I don't even think about it. You don't even think just this field has been so full of surprises for me. You're </s><s>[INST] just taking in and see fundamental. Yeah. [/INST] I just can't. Maybe that's because I've been around the field long enough to think, you know, don't don't go that way. Herb Simon was terrible about making these predictions of when this and that would happen. And he was a sensible guy. </s><s>[INST] His quotes are he often used right? As uh Yeah. Do you have concerns about ai the existential threats that many people like Elon musk and uh Sam Harris and others that [/INST] are thinking about that takes up a half a chapter in my book. I call it the male gaze. Well, you hear me out, The male gaze is actually a term from film criticism and I'm blocking on the woman's who dreamed this up. But she pointed out how most movies were made from the male point of view that women were objects, not subjects they didn't have any uh agency and so on and so forth. So when Elon and his pals Hawking and so on. Ok. Ai is going to eat our lunch and our dinner and our midnight snack too. I thought what? And I said to edify confound. Ah this is the first guy first. These guys have always been the smartest guy on the block and here comes something that might be smarter. Ooh let's stamp it out before it takes over. And Ed laughed. He said I didn't think about it that way. But I did, I did. And it is the male gaze, you know, okay, suppose these things do have agency. Well let's wait and see what happens. Can we imbue them with ethics? Can we imbue them with a sense of empathy or are they just going to be? I know we've had centuries of guys like </s><s>[INST] that. That's interesting that uh the the ego the male gaze is immediately threatened. So you can't think in a patient calm way of of how the tech could evolved. And he's speaking of which, You're 96 book The Future of Women I think at the time and now certainly now. I mean I'm sorry maybe at the time but I'm more cognizant of now is extremely relevant. You and Nancy Ramsey talk about four possible futures of women in science and Tech. So if we look at the decades before and after the book was released, can you tell a history, sorry, of women in science and tech and how it has evolved? How have things changed? Where do we stand? [/INST] Not enough? They have not changed enough the way that women are ground down in computing is simply unbelievable. </s><s>[INST] But What are the four possible futures for women in tech from the book? [/INST] What you're really looking at our various aspects of the present? So for each of those you could say, oh yeah, we do have backlash, look at what's happening with abortion and so on and so forth. Uh We have one step forward. One step back. The Golden Age of Equality was the hardest chapter to write. And I used something from the santa fe institute, which is the sand pile effect that you drop sand very slowly onto a pile and it grows and it grows and it grows until suddenly it just breaks apart. And in a way, me too has done that. That was the last drop of sand that broke everything apart. That was a perfect example of the sand pile effect. And that made me feel good. It didn't change all of society, but it really woke a lot of people up. </s> |
<s>[INST] lots of fiber, [/INST] get lots of fiber, be nice to everybody. And uh yeah, don't take things too seriously. It's a good, it's a good one. Our minds are fucking big weird big it's a big weird shitty fucking bucket of shit that's trying to get you to think horrible shit about yourself all </s><s>[INST] the time. Shitty bucket of ship. [/INST] It </s><s>[INST] is a book I never read but I read the title and it's good Words to live by which is uh don't sweat the small stuff and it's all small stuff. That's another way [/INST] that dr Phil wasn't dr Phil. I </s><s>[INST] don't know. But I think the conclusion is also has fiber as part of it. I think that all ties it together and in the end of course just put love out there in the world. I think that's a pretty good way. [/INST] What would you say is the meaning of life? I </s><s>[INST] would say love. Yeah. Yeah. It's a it's a long conversation on what that really means. But I'm sure robots are involved. Yeah. [/INST] Well let me tell you, I feel a little safer knowing that someone who has a hand in bringing these robots to the masses as you do has that opinion of of love and how important it is. I think that's great because otherwise it's going to be that fucking scene from T. Two where uh linda Hamilton's holding onto the fence and getting all of her flesh blown off of her skeleton before the rest of her is is wiped away because the skynet shit. Anyway. I'm just terrified of dudes all the time. That's why I think that they will </s><s>[INST] dudes in the wrong hands can do a lot of damage. [/INST] That's why chad and I need to do our best to control </s><s>[INST] it. Need to travel back in time and murder chad [/INST] I </s><s>[INST] think. [/INST] Yeah, that's yeah, </s><s>[INST] that's the only way [/INST] it's been said. I </s><s>[INST] don't know why I need to travel back in time but it just [/INST] uh today but I'm </s><s>[INST] very suspicious. [/INST] My nefarious plans for chad involve going back to tomorrow and planning for yesterday and then and hopefully do Dzi will give me the answer there with what, what it is to do with chad's frozen body. If I got to drive it out too. If I got to take my, you know, if I got to get, get a hold of it like one of those Tesla Mom vans and shove my garage freezer in it and plug it in and shove chad in there and drive out to Arizona and deliver him under a mountain or wherever the fun this place is. And so here's this dog tag. What does this get me Like? That's going to be 300 bucks. Do you have, Do you take Amex? No. And I'd be like sh </s><s>[INST] it [/INST] and I'll just dump </s><s>[INST] them somewhere. Breaking [/INST] bad stuff. </s><s>[INST] Well I would like to thank you and the, what is it? The Canadian international agency apparel [/INST] for. I </s><s>[INST] can't wait for the sneakers from dudes. I can't wait for all the, all the podcasts. Uh that Ai can and all the trouble I can get you in. So I'm a huge fan of yours. It's a huge honor that you would talk with me today. Well this has been amazing, [/INST] cheers pal likewise and I'm happy to be here. Man, cheers Holy! What? </s><s>[INST] Thanks for listening to this conversation with Will Sasso to support this podcast. Please check out our sponsors in the description and now let me leave you with some words from john Candy, one of Will's favorite actors. I think I may have become an actor to hide from myself. You can escape into a character. Thank you for listening and hope to see you next time. [/INST] </s> |
<s>[INST] Please check out our sponsors in the description and now with your friends, here's Chris Lattner. It's been, I think, two years since we last talked and in that time, you somehow went and co-created a new programming language called Mojo. So it's optimized for AI, it's a super set of Python. Let's look at the big picture. [/INST] What is the vision for Mojo? For Mojo, well, so I mean, I think you have to zoom out. So I've been working on a lot of related technologies for many, many years. So I've worked on LLVM and a lot of things in mobile and servers and things like this, but the world's changing. And what's happened with AI is we have new GPUs and new machine learning accelerators and other ASICs and things like that that make AI go real fast. At Google, I worked on TPUs. That's one of the biggest, larger scale deployed systems that exist for AI. And really what you see is if you look across all of the things that are happening in the industry, there's this new compute platform coming. And it's not just about CPUs or GPUs or TPUs or NPUs or IPUs or whatever, all the PUs, right? It's about how do we program these things, right? And so for software folks like us, right, it doesn't do us any good if there's this amazing hardware that we can't use. And one of the things you find out really quick is that having the theoretical capability of programming something and then having the world's power and the innovation of all the smart people in the world get unleashed on something can be quite different. And so really where Mojo came from was starting from a problem of we need to be able to take machine learning, take the infrastructure underneath it and make it way more accessible, way more usable, way more understandable by normal people and researchers and other folks that are not themselves like experts in GPUs and things like this. And then through that journey, we realized, hey, we need syntax for this. </s><s>[INST] We need to do program language, right? So one of the main features of the language, I say so fully in jest is that it allows you to have the file extension to be an emoji or the fire emoji, which is one of the first emojis used as a file extension I've ever seen in my life. And then you ask yourself the question, why in the 21st century are we not using Unicode for file extensions? This means it's an epic decision. I think clearly the most important decision made in most, but you could also just use Mojo as the file extension. [/INST] Well, so, okay, so take a step back. I mean, come on, Lex. Do you think that the world's ready for this? This is a big moment in the world, right? </s><s>[INST] This is, we're releasing this onto the world. This is innovation. I mean, it really is kind of brilliant. Emojis are such a big part of our daily lives. [/INST] Why is it not in programming? This is innovation. Well, and like you take a step back and look at what file extensions are, right? They're basically metadata, right? And so why are we spending all the screen space on them and all this stuff? Also, you know, you have them stacked up next to text files and PDF files and whatever else. Like if you're gonna do something cool, you want it to stand out, right? And emojis are colorful. </s><s>[INST] They're visual, they're beautiful, right? What's been the response so far from, is there support on like Windows on the operating system in displaying like file explorer? [/INST] Yeah, yeah, yeah, yeah, yeah. The one problem I've seen is that Git doesn't escape it right. And so it thinks that the fire emoji is unprintable and so it like prints out weird hex things if you use the command line Git tool. But everything else, as far as I'm aware, works fine. And I have faith that Git can be improved. So I'm not sure. And so GitHub is fine. GitHub is fine, yeah. GitHub is fine, Visual Studio Code, Windows, like all this stuff, totally ready because people have internationalization in their normal part of their path. So this is just like taking the next step, right? Somewhere between, oh wow, that makes sense. Cool, I like new things too. Oh my God, you're killing my baby. Like what are you talking about? This can never be, like I can never handle this. How am I gonna type this, like all these things. And so this is something where I think that the world will get there. We don't have to bet the whole farm on this. I think we can provide both paths, </s><s>[INST] but I think it'll be great. And so GitHub, when can we have emojis [/INST] as part of the code, I wonder? Yeah, so I mean, lots of languages provide that. So I think that we have partial support for that. It's probably not fully done yet, but yeah, you can do that. For example, in Swift, you can do that for sure. So an example we gave at Apple was the dog cow. So that's a classical Mac heritage thing. And so you use the dog and the cow emoji together, and that could be your variable name. But of course the internet went and made pile of poop for everything. So if you wanna name your function pile of poop, then you can totally go to town and see how that gets through code review. </s><s>[INST] Okay, so let me just ask a bunch of random questions. So is Mojo primarily designed for AIs, or is it a general purpose programming? [/INST] Yeah, good question, yeah, good question. So it's AI first. And so AI is driving a lot of the requirements. And so Modular is building and designing and driving Mojo forward. And it's not because it's an interesting project theoretically to build, it's because we need it. And so at Modular, we're really tackling the AI infrastructure landscape and the big problems in AI. And the reasons that it is so difficult to use and scale and adopt and deploy and all these big problems in AI. And so we're coming at it from that perspective. Now, when you do that, when you start tackling these problems, you realize that the solution to these problems isn't actually an AI specific solution. And so while we're doing this, we're building Mojo to be a fully general programming language. And that means that you can obviously tackle GPUs and CPUs and these AI things, but it's also a really great way to build NumPy and other things like that. Or just if you look at what many Python libraries are today, often they're a layer of Python for the API and they end up being C and C++ code underneath them. That's very true in AI, that's true in lots of other domains as well. And so anytime you see this pattern, that's an opportunity for Mojo to help simplify the world </s><s>[INST] and help people have one thing. To optimize through simplification by having one thing. So you mentioned Modular. Mojo is the programming language, [/INST] Modular is the whole software stack. So just over a year ago, we started this company called Modular. Okay, what Modular is about is, it's about taking AI and up leveling it into the next generation, right? And so if you take a step back, what's gone on in the last five, six, seven, eight years is that we've had things like TensorFlow and PyTorch and these other systems come in, you've used them, you know this. And what's happened is these things have grown like crazy and they get tons of users. It's in production deployment scenarios, it's being used to power so many systems. I mean, AI is all around us now. It used to be controversial years ago, but now it's a thing. But the challenge with these systems is that they haven't always been thought out with current demands in mind. And so you think about it, where were LLMs eight years ago? Well, they didn't exist, right? AI has changed so much and a lot of what people are doing today are very different than when these systems were built. And meanwhile, the hardware side of this has gotten into a huge mess. There's tons of new chips and accelerators and every big company's announcing a new chip every day, it feels like. And so between that, you have like moving system on one side, moving system on the other side and it just turns into this gigantic mess, which makes it very difficult for people to actually use AI, particularly in production deployment scenarios. And it's what modular is doing is we're helping build out that software stack to help solve some of those problems so then people can be more productive and get more AI research into production. Now, what Mojo does is it's a really, really, really important piece of that. And so that is part of that engine and part of the technology </s><s>[INST] that allows us to solve these problems. So Mojo is a programming language that allows you to do a higher level programming, the low level programming, they do all kinds of programming in that spectrum [/INST] that gets you closer and closer to the hardware. So take a step back. </s><s>[INST] So Lex, what do you love about Python? Oh boy, where do I begin? [/INST] What is love, what do I love about Python? Your guy who knows love, I know this. </s><s>[INST] Yes, yes. How intuitive it is, how it feels like I'm writing natural language English, how when I can not just write, but read other people's code, somehow I can understand it faster. It's more condensed than other languages, like ones I'm really familiar with like C++ and C. There's a bunch of sexy little features. Yeah. We'll probably talk about some of them, [/INST] but list comprehensions and stuff like this. Yeah. Also, and don't forget the entire ecosystem of all the packages. Oh yeah, there's probably huge. Because there's always something, </s><s>[INST] if you wanna do anything, there's always a package. Yeah, so it's not just the ecosystem of the packages and the ecosystem of the humans that do it. That's a really, that's an interesting dynamic. That's good. I think something about the usability and the ecosystem makes the thing viral, [/INST] it grows and then it's a virtuous cycle, I think. Well, and there's many things that went into that. So I think that ML was very good for Python. And so I think that TensorFlow and PyTorch and these systems embracing Python really took and helped Python grow. But I think that the major thing underlying it is that Python's like the universal connector, right? It really helps bring together lots of different systems so you can compose them and build out larger systems without having to understand how it works. </s><s>[INST] But then what is the problem with Python? Well, I guess you could say several things, [/INST] but probably that it's slow. I think that's usually what people complain about, right? And so slow, I mean, other people complain about tabs and spaces versus curly braces or whatever, but I mean, those people are just wrong because it is actually just better to use indentation. </s><s>[INST] Wow, strong words. So actually, I'm a small tangent. Let's actually take that, let's take all kinds of tangents. Oh, come on, Lex, you can push me on. I can take it. Design, listen, I've recently left Emacs for VS Code. Okay. The kind of hate mail I had to receive because on the way to doing that, I also said I've considered Vim. Yep. And chose not to and went with VS Code. You're touching on deep religions, right? Anyway, tabs is an interesting design decision and so you've really written a new programming language here. Yes, it is a super set of Python, but you can make a bunch of different interesting decisions here. Totally, yeah. And you chose actually to stick with Python as in terms of some of the syntax. [/INST] Yeah, well, so let me explain why, right? So I mean, you can explain this in many rational ways. I think that the indentation is beautiful, but that's not a rational explanation, right? But I can defend it rationally, right? So first of all, Python 1 has millions of programmers. Yeah. It is huge, it's everywhere, it owns machine learning, right? And so factually, it is the thing, right? Second of all, if you look at it, C code, C++ code, Java, whatever, Swift, curly brace languages, also run through formatting tools and get indented. And so if they're not indented correctly, first of all, we'll twist your brain around, it can lead to bugs. There's notorious bugs that have happened across time where the indentation was wrong or misleading and it wasn't formatted, right? And so it turned into an issue, right? And so what ends up happening in modern large scale code bases is people run automatic formatters. So now what you end up with is indentation and curly braces. Well, if you're gonna have the notion of grouping, why not have one thing, right? And get rid of all the clutter and have a more beautiful thing, right? Also you look at many of these languages, it's like, okay, well you can have curly braces or you can omit them if there's one statement or you just like enter this entire world of complicated design space that objectively you don't need </s><s>[INST] if you have Python style indentation, so. Yeah, I would love to actually see statistics on errors made because of indentation. Like how many errors are made in Python versus in C++ that have to do with basic formatting, all that kind of stuff. [/INST] I would love to see. I think it's probably pretty minor because once you get, like you use VS code, I do too. So if you get VS code set up, it does the indentation for you generally, right? And so you don't, you know, it's actually really nice to not have to fight it. And then what you can see is the editor is telling you how your code will work by indenting it, </s><s>[INST] which I think is pretty cool. I honestly don't think I've ever, I don't remember having an error in Python [/INST] because I indented stuff wrong. So, I mean, I think that there's, again, this is a religious thing. And so I can joke about it and I love to kind of, you know, I realize that this is such a polarizing thing and everybody wants to argue about it. And so I like poking at the bear a little bit, right? But frankly, right, come back to the first point, Python one, like it's huge, it's an AI, it's the right thing. For us, like we see Mojo as being an incredible part of the Python ecosystem. We're not looking to break Python or change it or quote unquote fix it. We love Python for what it is. Our view is that Python is just not done yet. And so if you look at, you know, you mentioned Python being slow. Well, there's a couple of different things that go into that, which we can talk about if you want. But one of them is that it just doesn't have those features that you would use to do C-like programming. And so if you say, okay, well, I'm forced out of Python into C for certain use cases, well, then what we're doing is we're saying, okay, well, why is that? Can we just add those features that are missing from Python back up to Mojo? And then you can have everything that's great about Python, all the things you're talking about that you love, plus not be forced out of it when you do something a little bit more computationally intense </s><s>[INST] or weird or hardware-y or whatever it is that you're doing. Well, a million questions I want to ask, but high level again, is it compiled or is it an interpreted language? [/INST] So Python is just in time compilation, what's Mojo? So Mojo is a complicated answer, it does all the things. So it's interpreted, it's shit compiled, and it's statically compiled. And so this is for a variety of reasons. So one of the things that makes Python beautiful is that it's very dynamic. And because it's dynamic, one of the things they added is that it has this powerful metaprogramming feature. And so if you look at something like PyTorch or TensorFlow or I mean, even a simple use case, like you'd find a class that has the plus method, right? You can overload the dunder methods like dunder add, for example, and then the plus method works on your class. And so it has very nice and very expressive dynamic metaprogramming features. In Mojo, we want all those features come in. Like we don't want to break Python, we want all to work. But the problem is, is you can't run those super dynamic features on an embedded processor or on a GPU, right? Or if you could, you probably don't want to just because of the performance. And so we entered this question of saying, okay, how do you get the power of this dynamic metaprogramming into a language that has to be super efficient in specific cases? And so what we did was we said, okay, we'll take that interpreter, Python has an interpreter in it. Right? Take that interpreter and allow it to run at compile time. And so now what you get is you get compile time metaprogramming. And so this is super interesting and super powerful because one of the big advantages you get is you get Python style, expressive APIs. you get the ability to have overloaded operators. If you look at what happens inside of PyTorch, for example, with automatic differentiation and eager mode and all these things, they're using these really dynamic and powerful features at runtime. But we can take those features and lift them, so that they run at compile time. </s><s>[INST] So your C++ is metaprogramming with templates, [/INST] but it's really messy. It's super messy. It was accidentally, I mean, different people have different interpretations. My interpretation is that it was made accidentally powerful. It was not designed to be Turing complete, for example, but that was discovered along the way accidentally. So there have been a number of languages in the space, and so they usually have templates or code instantiation, code copying features of various sorts. Some more modern languages or some more newer languages, let's say, they're fairly unknown, like Zig, for example, says, okay, well, let's take all of those types that you can run it, all those things you can do at runtime, and allow them to happen at compile time. So one of the problems with C++, I mean, which is one of the problems with C++. Here we go. Wrong words. We're going to send everybody to it. That's okay. I mean, everybody hates me for a variety of reasons anyways, I'm sure. I've written enough. Just the way they show love. I've written enough C++ code to earn a little bit of grumpiness with C++. But one of the problems with it is that the meta programming system templates is just a completely different universe from the normal runtime programming world. So if you do a meta programming and programming, it's just like a different universe. Different syntax, different concepts, different stuff going on. So again, one of our goals with Mojo's to make things really easy to use, easy to learn. </s><s>[INST] So there's a natural stepping stone, for example. You're going to send everybody to it. [/INST] I freaking. As you do this, you say, okay, well, I have to do programming at runtime, after you're programming at compile time, why are these different things? </s><s>[INST] How hard is that to pull it out, because that sounds, to me, as a fan of meta programming in C++, even, how hard is it to pull that off? That sounds really, really exciting, because you can do the same style programming at compile time and at runtime. [/INST] That's really, really exciting. Yep, yep. And so, in terms of the compiler implementation details, it's hard. I won't be shy about that, it's super hard. It requires what Mojo has underneath the covers is a completely new approach to the design of the compiler itself. And so this builds on these technologies like MLIR that you mentioned, but also includes other caching and other interpreters and JIT compilers, and other stuff. Show you a duplicate of the inside the compiler. Within the compiler, yes. </s><s>[INST] Show you a duplicate of the interpreter. [/INST] Within in the compiler, yeah. And so it really takes the standard model of programming languages and kind of twists it and unifies it the runtime model, which I think is really cool. To me, the value of that is that, again, many of these languages have metaprogramming features. They grow macros or something. Lisp, right? I know your roots, right? This is a powerful thing. If you go back to Lisp, one of the most powerful things about it is that it said that the metaprogramming and the programming are the same. That made it way simpler, way more consistent, way easier to understand, reason about, and it made it more composable. If you build a library, you can use it both at runtime and compile time, which is pretty </s><s>[INST] cool. Cool. For machine learning, I think metaprogramming, I think we could generally say, is extremely useful. I'll jump around, but the feature of autotuning and adaptive compilation just blows my mind. [/INST] Yeah. Let's come back to that. All right. What is machine learning? What is a machine learning model? You take a PyTorch model off the internet, right? It's really interesting to me because what PyTorch and what TensorFlow and all these frameworks are pushing compute into is they're pushing into this abstract specification of a compute problem, which then gets mapped in a whole bunch of different ways. This is why it became a metaprogramming problem. You want to be able to say, cool, I have this neural net. Now run it with batch size 1,000. Do a mapping across batch. Or okay, I want to take this problem, now run it across 1,000 GPUs. This problem of describe the compute and then map it and do things and transform it, actually it's very profound, and that's one of the things that makes machine learning systems really special. </s><s>[INST] Maybe keep checking. All right. Can you describe autotuning and how do you pull off... I mean, I guess adaptive compilation is what we're talking about as metaprogramming. How do you pull off autotuning? Is that as profound as I think it is? It seems like a really... We'll mention list comprehensions. To me, from a quick glance at Mojo, which by the way, I have to absolutely dive in. As I realize how amazing this is, I absolutely must dive in. [/INST] That looks like just an incredible feature for machine learning people. Yeah. Well, so what is autotuning? Take a step back. Autotuning is a feature in Mojo. Very little of what we're doing is actually research. Many of these ideas have existed in other systems and other places, and so what we're doing is we're pulling together good ideas, remixing them, and making them into hopefully a beautiful system. So autotuning, the observation is that it turns out hardware systems, algorithms are really complicated. It turns out maybe you don't actually want to know how the hardware works. A lot of people don't. There are lots of really smart hardware people, I know a lot of them, where they know everything about, okay, the cache size is this, and the number of registers is that, and if you use this length of vector, it's going to be super efficient because it maps directly onto what it can do. All this kind of stuff. The GPU has SMs, and it has a warp size of whatever. All the stuff that goes into these things, or the tile size of a TPU is 128, these factoids. My belief is that most normal people, and I love hardware people also, I'm not trying to fen literally everybody on the internet, but most programmers actually don't want to know this stuff. So if you come at it from perspective of how do we allow people to build both more abstracted but also more portable code, because it could be that the vector length changes, or the cache size changes, or it could be that the tile size of your matrix changes, or the number in A100 versus an H100 versus a Volta versus a whatever GPU have different characteristics. A lot of the algorithms that you run are actually the same, but the parameters, these magic numbers you have to fill in end up being really fiddly numbers that an expert has to go figure out. So what autotuning does, it says, okay, well, guess what? There's a lot of compute out there. So instead of having humans go randomly try all the things, or do a grid search, or go search some complicated multi-dimensional space, how about we have computers do that? And so what autotuning does is you can say, hey, here's my algorithm. If it's a matrix operation or something like that, you can say, okay, I'm going to carve it up into blocks, I'm going to do those blocks in parallel, and I want this with 128 things that I'm running on, I want to cut it this way or that way or whatever. </s><s>[INST] And you can say, hey, go see which one's actually empirically better on this system. [/INST] And then the result of that, you cash for that system, you save it and save it. And so come back to twisting your compiler brain, right? So not only does the compiler have an interpreter that you use to do programming, the compiler that interpreter that metaprogramming now has to actually take your code and go run it on a target machine, see which one of likes the best and then stitch it in, and then keep </s><s>[INST] going. Part of the compilation is machine specific. [/INST] Yeah. Well, so I mean, this is an optional feature, right? So you don't have to use it for everything. But yeah, if you're so one, one of one of the things that we're in the quest of is ultimate performance. Yes. Right. Ultimate performance is important for a couple of reasons, right? So if you're an enterprise, you're looking to save costs and compute and things like this, ultimate performance translates to fewer servers. Like if you care about the environment, hey, better performance leads to more efficiency, right? I mean, you could joke and say like, you know, Python's bad for the environment. Right. And so if you move to Mojo, it's like at least 10 X better just out of the box and keep going. Right. But but performance is also interesting because it leads to better products. And so in the space of machine learning, if you reduce the latency of a model so that it runs faster, so every time you query the server running the model, it takes less time, well, then the product team can go and make the model bigger. Well, that's actually makes it so you have a better experience as a customer. And so a lot of people care about that. </s><s>[INST] Yeah. Hey, you know. So for auto tuning for like tile size, you mentioned 120 for TPU. You would specify like a bunch of options to try just in the code. Just simple statement. And then you just set and forget and know, depending [/INST] what wherever it compiles, it'll actually be the fastest. And yeah, exactly. And the beauty of this is that it helps you in a whole bunch of different ways. Right. So if you're building so often what will happen is that, you know, you've written a bunch of software yourself, right? You you wake up one day, you say, I have an idea. I'm going to go cut up some code. I get to work. I forget about it and move on with life. I come back six months or a year or two years or three years later, you dust it off and you go use it again in a new environment. And maybe your GP is different. Maybe you're running on a server instead of a laptop, maybe whatever. Right. And so the problem now is you say, OK, well, I mean, again, not everybody cares about performance, but if you do, you say, OK, well, I want to take advantage of all these new features. I don't want to break the old thing, though. Right. And so the typical way of handling this kind of stuff before is, you know, if you're talking about people's templates, or you're talking about C with macros, you end up with if deaths, you get like all these weird things get layered in, make the code super complicated. And then how do you test it becomes this crazy complexity, multidimensional space you have to worry about? And, you know, that just doesn't scale very well. </s><s>[INST] Actually, let me just jump around before I go to some specific features like the increase in performance here that we're talking about can be just insane. You write that Moja can provide a 35000 X speed up over Python. [/INST] How does it do that? How does it do that? Yeah, so I can even do more, but we'll get to that. So so first of all, when we say that we're talking about what's called C Python, it's the default Python that everybody uses when you type Python three. That's like typically the one you use. Right. So C Python is an interpreter. And so interpreters, they have an extra layer of bike codes and things like this that they have to go read, parse, interpret, and it makes them kind of slow from that perspective. And so one of the first things we do is we move to a compiler. And so I'm just moving to a compiler, getting the interpreter out of the loop is two to five to 10 X speed up depending on the code. So just out of the gate, like she's in more modern techniques. Now, if you do that, one of the things you can do is you can start to look at how CPython started to lay out data. And so one of the things that CPython did, and this isn't part of the Python spec necessarily, but this is just sets of decisions, is that if you take an integer, for example, it'll put it in an object because in Python, everything's an object. And so they do the very logical thing of keeping the memory representation of all objects the same. So all objects have a header, they have payload data, and what this means is that every time you pass around an object, you're passing around a pointer to the data. Well, this has overhead. It turns out that modern computers don't like chasing pointers very much and things like this. It means you have to allocate the data. It means you have to reference count it, which is another way that Python uses to keep track of memory. And so this has a lot of overhead. And so if you say, okay, pointer, let's try to get that out of the heap, out of a box, of the new direction, and into the registers. </s><s>[INST] That's another 10x. So it adds up, if you reference counting every single thing you create, that adds up. [/INST] That adds up. Yeah, and if you look at people complain about the Python GIL, this is one of the things that hurts the parallelism, that's because of the reference counting, right, so the GIL and reference counting are very tightly intertwined in Python, it's not the only thing, but it's very tightly intertwined. And so then you lean into this and you say, OK, cool, well, modern computers, they can do more than one operation at a time. So they have vectors. What is a vector? Well, a vector allows you to take one, instead of taking one piece of data, doing an add or a multiply and then picking up the next one. You can now do 4, or 8, or 16, or 32 at a time. Well, Python doesn't expose that because of reasons. And so now you can say, OK, well, you can adopt that. And now you have threads. Now you have additional things, like you can control memory hierarchy. And so Moda allows you to do, is it allows you to start taking advantage of all these powerful things that have been built into the hardware over time and it gives, the library gives very nice features so you can say just parallelize this, do this in parallel, right? So it's very powerful weapons against slowness which is why people have been I think having fun like just taking code and making it go fast because it's just kind of an adrenaline </s><s>[INST] rush to see like how fast you can get things. Before I talk about some of the interesting stuff with parallelization and all that, let's first talk about like the basics. We talked to indentation, right? So this thing looks like Python. It's sexy and beautiful like Python as I mentioned. Is it a typed language? So what's the role of types? Yeah, good question. [/INST] So yeah, good question. So Python has types, it has strings, it has integers, it has dictionaries and like all that stuff but they all live at runtime, right? And so because all those types live at runtime in Python you never, you don't have to spell them. Python also has like this whole typing thing going on now and a lot of people use it. I'm not talking about that. That's kind of a different thing. We can go back to that if you want. But typically the, you know, you just say I take, I have a def and my def takes two parameters and I'm gonna call them a and b and I don't have to write a type, okay? So that is great but what that does is that forces what's called a consistent representation. So these things have to be a pointer to an object with the object header and they all have to look the same. And then when you dispatch a method, when you go through all the same different paths no matter what the receiver, whatever that type is. So what Mojo does is it allows you to have more than one kind of type. And so what it does is allows you to say okay, cool. I have an object and objects behave like Python does, and so it's fully dynamic and that's all great. And for many things, classes like that's all very powerful and very important. But if you want to say hey it's an integer and it's 32 bits, or 64 bits, or whatever it is. Or it's a floating point value. It's 64 bits. Well then the compiler can take that and it can use that to do way better optimization. It turns out, again, getting rid of the indirection, that's huge, means you can get better code completion because compiler knows what the type is and so it knows what operations work on it. That's actually pretty huge. What Mojo does is it allows you to progressively adopt types into your program. You can start, again, it's compatible with Python. Then you can add however many types you want wherever you want them. And if you don't wanna deal with it you don't have to deal with it, right? And so one of our opinions on this is that it's not that types are the right thing </s><s>[INST] or the wrong thing, it's that they're a useful thing. Which was kind of optional, it's not strict typing and you don't have to specify type. Exactly. Okay, so starting from the thing that Python is kind of reaching towards right now with trying to inject types into it, [/INST] what it's doing. Yeah, with a very different approach but yes. </s><s>[INST] Yeah, so. What's the different approach, I'm actually one of the people that have not been using types very much in Python. So I haven't. Why did you say? It just, well, because I know the importance. It's like adults use strict typing. And so I refuse to grow up in that sense. It's a kind of rebellion. But I just know that. [/INST] That's okay. Why did you say? </s><s>[INST] But it probably reduces the amount of errors, even just for forget about performance improvements. [/INST] It probably reduces errors when you do strict typing. Yeah. So I mean, I think it's interesting if you look at that, right? The reason I'm giving you a hard time, is that there's this cultural norm, this pressure, there has to be a right way to do things. Growing up is only do it one way. And if you don't do that, you should feel bad. Some people feel like Python's a guilty pleasure or something. And that's like, when it gets serious, I need to go rewrite it. I mean, cool, I understand history and I understand kind of where this comes from, but I don't think it has to be a guilty pleasure. So if you look at that, you say, why do you have to rewrite it? Well, we have to rewrite to deploy. Well, why do you want to deploy? Well, you care about performance, or you care about perfectability or you want, a tiny thing on the server that has no dependencies or you know, you have objectives that you're trying to attain. So what if Python can achieve those objectives? So if you want types, well, maybe you want types because you want to make sure you're passing the right thing. Sure, you can add a type. If you don't care, you're prototyping some stuff, you're hacking some things out, you're like pulling some ram code off the internet, it should just work. And you shouldn't be pressured, you shouldn't feel bad about doing the right thing or the thing that feels good. Now, if you're in a team, you're working at some massive internet company and you have 400 million lines of Python code, well, they may have a house rule that you use types because it makes it easier for different humans to talk to each other and understand what's going on and bugs at scale. And so there are lots of good reasons why you might want to use types, but that doesn't mean that everybody should use them all the time. So what Mojo does is it says, cool, well, allow people to use types. And if you use types, you get nice things out of it, right? You get better performance and things like this, right? But Mojo is a full, compatible super set of Python, right? And so that means that has to work without types. It has to support all the dynamic things, has to support all the packages, it has to support list comprehensions and things like this, right? And so that that starting point, I think, is really important. And I think that, again, you can look at why I care so much about this, and there's many different aspects of that, one of which is the world went through a very challenging migration from Python 2 to Python 3, right? And this migration took many years, and it was very painful for many teams, right? And there's a lot of things that went on in that. I'm not an expert in all the details, and I honestly don't want to be. I don't want the world to have to go through that, right? And people can ignore Mojo, and if it's not their thing, that's cool. But if they want to use Mojo, I don't want them to have to rewrite all their code. </s><s>[INST] Yeah, I mean, okay, the superset part is just, I mean, there's so much brilliant stuff here. That definitely is incredible. We'll talk about that. First of all, how's the typing implemented differently in Python versus Mojo? So this heterogeneous flexibility, you said, is definitely implemented. [/INST] First of all, how is it definitely implemented? Yeah. So I'm not a full expert in the whole backstory on types in Python. So I'll give you that. I can give you my understanding. My understanding is basically, like many dynamic languages, the ecosystem went through a phase where people went from writing scripts to writing large scale, huge code bases in Python, and at scale, it kind of helps to have types. People want to be able to reason about interfaces. What do you expect to string or an int? These basic things, right? And so what the Python community started doing is it started saying, okay, let's have tools on the side, checker tools. Right? To go and enforce some variants, check for bugs, try to identify things. These are called static analysis tools generally, and so these tools run over your code and try to look for bugs. What ended up happening is there's so many of these things, so many different weird patterns and different approaches on specifying the types and different things going on that the Python community realize and recognize, hey, hey, hey, there's a thing here. And so what they started to do is they started to standardize the syntax for adding types to Python. Now, one of the challenges that they had is if they're coming from kind of this fragmented world where there's lots of different tools, they have different trade-offs and interpretations and the types mean different things. And so, if you look at types in Python, according to the Python spec, the types are ignored, right? So, according to the Python spec, you can write pretty much anything in a type position, okay? And technically, you can write any expression, okay. Now, that's beautiful because you can extend it. You can do cool things. build your own tools, you can build your own house, linter or something like that, right? But it's also a problem because any existing Python program may be using different tools and they have different interpretations. And so if you adopt somebody's package into your ecosystem, try around the tool you prefer, it may throw out tons of weird errors and warnings and problems just because it's incompatible with how these things work. Also because they're added late and they're not checked by the Python interpreter, it's always kind of more of a hint than it is a requirement. Also the CPython implementation can't use them for performance and so it's really- </s><s>[INST] That's a big one, right? So you can't utilize for the compilation, [/INST] for the just in time compilation. Okay. Exactly. And this all comes back to the design principle of it's, they're kind of hints, they're kind of the definitions a little bit murky. It's unclear exactly the interpretation in a bunch of cases. And so because of that, you can't actually, even if you want to, it's really difficult to use them to say like, it is going to be an int and if it's not, it's a problem, right? A lot of code would break if you did that. So in Mojo, right. So you can still use those kinds of type annotations, it's fine, but in Mojo if you declare a type and you use it, then it means it is going to be that type and the compiler helps you check that enforce it and it's safe. </s><s>[INST] And it's not, it's not a best effort hint kind of a thing. So if you try to shove a string type thing into an integer- You get an error. And you get another. [/INST] From the compiler at compile time. You get an error. </s><s>[INST] And you get another. Nice. [/INST] Okay, what kind of basic types are there? Yeah, so Mojo is pretty hardcore in terms of what it tries to do in the language, which is the philosophy there is that we, again, if you look at Python, Python's a beautiful language because it's so extensible. And so all of the different things in Python, like for loops and plus, and all these things can be accessed through these underbar and bar methods. Okay, so you have to say, okay, if I make something that is super fast, I can go all the way down to the metal. Why do I need to have integers built into the language? And so what Mojo does is it says, okay, well, we can have this notion of structs. So you have classes in Python, now you can have structs. Classes are dynamic, structs are static. Cool, we can get high performance, we can write C++ kind of code with structs if you want. These things mix and work beautifully together. But what that means is that you can go and implement strings and ints and floats and arrays and all that kind of stuff in the language, right? And so that's really cool because, you know, to me as a idealizing compiler language type of person, what I wanna do is I wanna get magic out of the compiler and put it in the libraries. Because if somebody can, you know, if we can build an integer that's beautiful and it has an amazing API and it does all the things you'd expect an integer to do, but you don't like it, maybe you want a big integer, maybe you wanna like sideways integer, I don't know, like what all the space of integers are, then you can do that and it's not a second-class citizen. And so if you look at certain other languages like C++, one I also love and use a lot, int is hard-coded in the language, but complex is not. And so it's kind of weird that, you know, you have this STD complex class, but you have int. And complex tries to look like a natural numeric type and things like this, but integers in floating point have these like special promotion rules and other things like that, they're magic and they're hacked into the compiler. And because of that, you can't actually make something </s><s>[INST] that works like the built in types. Is there something provided as a standard, because you know, because it's AI first, you know, numerical types is so important here, so is there something come like a nice standard implementation of integer inflow? [/INST] Yeah, yeah. So we're still building all that stuff out, so we provide entries and floats and all that kind of stuff. We also provide buffers and tensors and things like that that you'd expect in an ML context. Honestly, we need to keep designing and redesigning and working with the community to build that out and make that better. That's not our strength right now. Give us six months or a year, and I think it'll be way better. But the power of putting in the library means that we can have teams of experts that aren't </s><s>[INST] compiler engineers that can help us design and refine and drive us forward. So one of the exciting things we should mention here is that this is new and fresh. This cake is unbaked. It's almost baked. You can tell it's delicious, but it's not fully ready to be consumed. [/INST] Yep. That's very fair. It is very useful, but it's very useful if you're a super low-level programmer right now. And what we're doing is we're working our way up the stack. And so the way I would look at Mojo today in May in 2023 is that it's like a 0.1. So I think that a year from now, it's going to be way more interesting to a variety of people. But what we're doing is we decided to release it early so that people can get access to and play with it, and we can build it with the community. We have a big roadmap, fully published, being transparent about this. And a lot of people are involved in this stuff. And so what we're doing is we're really optimizing for building this thing the right way and building it the right way. It's interesting working with the community, because everybody wants it yesterday. </s><s>[INST] And so sometimes it's kind of, you know, there's some dynamics there, but I think it's the right thing. So there's a discord also. So the dynamics is pretty interesting. Sometimes the community probably can be very chaotic and introduce a lot of stress. Guido famously quit over the stress of the wall of his operator. I mean, it broke, maybe that would be, yeah, exactly. And so like, it could be very stressful to develop, but can you just tangent upon a tangent? Is it stressful to work through the design of various features here, given that the community is so richly involved? [/INST] Yeah. Yeah. I mean, it's, you know, it broke. It probably broke the camel's back. Exactly. So I've been doing open development and community stuff for decades now. Somehow this has happened to me. So I've learned some tricks, but the thing that always gets me is I want to make people happy. Right? And so this is maybe not all people all happy all the time, but generally I want people to be happy. Right? And so the challenge is that, again, we're tapping into some long, some deep seated long tensions and pressures, both in the Python world, but also in the AI world and the hardware world and things like this. And so people just want us to move faster, right? And so again, our decision was let's release this early. Let's get people used to it or access to it and play with it and like, let's build in the open, which we could have had the language monk sitting in the cloister up on the hilltop, like beavering away trying to build something. But in my experience, you get something that's way better if you work with the community. And so yes, it can be frustrating and can be challenging for lots of people involved. And, you know, if you, I mean, if you mentioned our Discord, we have over 10,000 people on the Discord, 11,000 people or something. Keep in mind we released Mojo like two weeks ago. So it's very active. So it's very cool. But what that means is that, you know, 10, 11,000 people all will want something different. And so what we've done is we've tried to say, okay, cool, here's our roadmap. The roadmap isn't completely arbitrary. It's based on here's the logical order in which to build these features or add these capabilities and things like that. And what we've done is we've spun really fast on bug fixes. And so we actually have very few bugs, which is cool. I mean, actually for a project in the state, but then what we're doing is we're dropping in features very deliberately. I mean, this is cool. </s><s>[INST] I mean, this is fun to watch because you got the two gigantic communities of like hardware like systems engineers, and then you have the machine learning Python people that are like higher level. Yeah. [/INST] And it's just too like for like army like they've been at war. Yeah. Yeah. Sure. Right. And so here's a test. It's like a Tolkien novel or something. Okay. So here's the test. And again, like it's super funny for something that's only been out for two weeks, right? People are so impatient, right? But okay, cool. Let's fast forward a year. Like in a year's time, Mojo will be actually quite amazing and solve tons of problems and be very good. People still have these problems, right? And so you look at this and you say, and the way I look at this at least is to say, okay, </s><s>[INST] well, we're solving big, longstanding problems, right? And so here's a test. It's like a Tolkien novel or something. [/INST] Okay. To me, I, again, working on many different problems, I want to make sure we do it right. Right? There's like a responsibility feel because if you mess it up, right, there's very few opportunities to do projects like this and have them really have impact on the world. If we do it right, then maybe we can take those feuding armies and actually heal some of those wounds. </s><s>[INST] Yeah. Right? Yeah. I mean, this is like a speech by George Washington or Abraham Lincoln or something. [/INST] And you look at this and it's like, okay, well, how different are we? Yeah. We all want beautiful things. We all want something that's nice. We all want to be able to work together. We all want our stuff to be used, right? And so if we can help heal that, now I'm not optimistic that all people will use Mojo and they'll stop using C++. Like, that's not my goal. Right? But if we can heal some of that, I think that'd be pretty cool. </s><s>[INST] Yeah. Yeah. And we start by putting the people who like braces into the gulag. No. [/INST] So, so there are, so there are proposals for adding braces to Mojo and we just tell them no. Okay. </s><s>[INST] Delightly. Yeah. Anyway, so there's a lot of amazing features on the roadmap and those ready to implement it. It'd be awesome. I can just ask you a few things. Yeah. Go for it. So, the, the other performance improvement comes from immutability. So what's the, what's this var and this let thing that we got going on? [/INST] What's immutability? Yeah. So, one of the things that is useful and it's not always required, but it's useful is knowing whether something can change out from underneath you. Right? And so in Python you have a pointer to an array, right? And so you pass that pointer to an array around to things. If you pass into a function, they may take that and scroll away in some other data structure. And so you get your array back and you go to use it and now somebody else is like putting stuff in your array. Yeah. How do you reason about that? It can be very complicated and at least lots of bugs, right? And so one of the things that, you know, again, this is not something Mojo forces on you, but something that Mojo enables is a thing called value semantics. And what value semantics do is they take collections like arrays, like dictionaries, also tensors and strings and things like this that are much higher level and make them behave like proper values. And so it makes it look like if you pass these things around, you get a logical copy of all the data. And so if I pass you an array, it's your array. You can go do what you want to it. You're not going to hurt my array. Now that is an interesting and very powerful design principle, defines away a ton of bugs. You have to be careful to implement it in an efficient way. </s><s>[INST] Yeah. [/INST] Is there a performance hit that's significant? Generally not. If you implement it the right way, but it requires a lot of very low level getting the language right bits. </s><s>[INST] I assume there'd be a huge performance hit because it's a really, the benefit is really [/INST] nice because you don't get into the complex. Absolutely. Well, the trick is you can't do copies. So you have to provide the behavior of copying without doing the copy. Yeah. </s><s>[INST] How do you do that? [/INST] How do you do that? It's not magic. It's just, it's actually pretty cool. Well, so first, before we talk about how that works, let's talk about how it works in Python. So in Python, you define a person class or maybe a person class is a bad idea. You define a database class, right? The database class has an array of records, something like that. And so the problem is, is that if you pass in a record or a class instance into database, it'll take a hold of that object and then it assumes it has it. And if you're passing an object in, you have to know that that database is going to take it. And therefore you shouldn't change it after you put in the database. This is the problem. You just kind of have to know that. You just have to kind of know that, right? And so you roll out version 1 of the database, you just kind of have to know that. Of course, Lex uses its own database, right? Yeah. Right? Because you built it. You understand how this works, right? Somebody else joins the team. They don't know this. Yes. Right? And so now they suddenly get bugs. You're having to maintain the database. You shake your fist. Mm-hmm. You argue the tenth time this happens. You're like, okay, we have to do something different, right? And so what you do is you go change your Python code and you change your database class to copy the record every time you add it. And so what ends up happening is you say, okay, I will do what's called a defensive copy inside the database. And then that way, if somebody passes something in, I will have my own copy of it. And they can go do whatever, and they're not going to break my thing. Okay. This is usually the two design patterns. If you look in PyTorch, for example, this is cloning a tensor. There's a specific thing, and you have to know where to call it. And if you don't call in the right place, you get these bugs, and this is state of the art, right? Right? Yeah. So a different approach. So it's used in many languages. So I work with it in Swift. As you say, okay, well, let's provide value semantics. And so we want to provide the view that you get a logically independent copy. But we want to do that lazily. And so what we do is we say, okay, if you pass something into a function, it doesn't actually make a copy. What it actually does is it just increments the reference to it. And if you pass it around, you stick in your database, they can go into the database. You own it. And then you come back out of the stack, nobody has copied anything. You come back out of the stack and then the caller lets go of it. Well, then you've just handed it off to the database. you've transferred it, and there's no copies made. Now on the other hand, if your coworker goes and hands you a record and you pass it in, you stick it in the database, and then you go to town and you start modifying it, what happens is you get a copy lazily on demand. And so what this does is gives you copies only when you need them, so it defines away </s><s>[INST] the bugs, but also generally reduces the number of copies in practice. But the implementation details are tricky here? Yeah. So this is, yes. Sometimes it was reference counting, but to make it performant across a number of different kinds of objects. [/INST] Yeah. Well, so you need a couple of things, and so there's many, so this concept has existed in many different worlds, and again, it's not novel research at all, right? The magic is getting the design right so that you can do this in a reasonable way, right? And so there's a number of components that go into this. One is when you're passing around, so we're talking about Python and reference counting and the expense of doing that, when you're passing values around, you don't want to do extra reference counting for no good reason, and so you have to make sure that you're efficient and you transfer ownership instead of duplicating references and things like that, which is a very low-level problem. You also have to adopt this, and you have to build these data structures. And so if you say, you know, Mojo has to be compatible with Python, so of course the default list is a reference semantic list that works the way you'd expect in Python, but then you have to design a value semantic list, and so you just have to implement that and then you implement the logic within. And so the role of the language here is to provide all the low-level hooks that allow the author of the type to be able to get and express this behavior without forcing it into all cases or hardcoding this into the language itself. But there's ownership so you're constantly tracking who owns the thing, who owns the thing? Yes, and so there's a whole system called ownership, and so this is related to work done in the Rust community, also the Swift community has done a bunch of work and there's a bunch of different other languages that have all kind of C++ actually has copy constructors and destructors and things like that. And so, and, I mean, C++ has everything so it has move constructors, and has like this whole world of things and so this is, this is a body of work that's kind of been developing for many years now. And so Mojo takes some of the best ideas out of all these systems and remixes it in a nice of ways, so that you get the power of something like the Rust programming language. But you don't have to deal with it when you don't want to, which is a major thing in terms </s><s>[INST] of teaching and learning and being able to use and scale these systems. How does that play with argument – conventions? What are they? Why are they important? How does the Bayley semantics, how does the transfer ownership, work with the arguments [/INST] when they're pastened? Yeah, so if you go deep into a systems programming land – this isn't, again, this is not something for everybody, but if you go deep into systems-programming land, what you encounter is you encounter these types that get weird. So if you're used to Python, you think about everything, I can just copy it around, I can go change it and mutate it, and do these things, and it's all cool. If you get into systems-programming land, you get into these things like I have an atomic number, or I have a mutex, or I have a uniquely-owned database handle, things like this, right? these types you can't necessarily copy. Sometimes you can't necessarily even move them to a different address. And so what Mojo allows you to do is it allows you to express, hey, I don't want to get a copy of this thing. I want to actually just get a reference to it. And by doing that, well, you can say, you can say, okay, if I'm defining something weird, like a atomic number or something, it's like it has to be, so an atomic number is a, an area in memory that multiple threads can access at a time without synchronous, without locks. And so and so like the definition of atomic numbers, multiple, different things have to be poking up that therefore they have to agree on where it is. And he can't just like move it out from underneath one because it kind of breaks what it means. And so that's an example of a type that you can't even you can't copy. You can't move it like once you created it has to be where it was. Now if you look at many other examples like a database handle, right? So, okay, well, what happens? How do you copy a database handle? Do you copy the whole database? That's not something you necessarily want to do. There's a lot of types like that where you want to be able to say that they are uniquely owned. So, there's always one of this thing and or if I create a thing, I don't copy it. And so, what Mojo allows you to do is it allows you to say, hey, I want to pass around a reference to this thing without copying it. And so, it has borrowed conventions. So, you can say you can use it, but you don't get to change it. You can pass it by mutable reference. So, if you do that, then you can get a reference to it, but you can change it. And so, it manages all that kind of stuff. </s><s>[INST] So, this is just a really nice implementation of make a C++ has, you know, the- Yes, reference has- The categories. Yes. Smart, smart- different kinds of implementations and smart pointers that you can explicitly define this a lot, but you're saying that's more like [/INST] the weird case versus the common case. The different kinds of pointers. Yeah, it has pointers. Smarts. Well, it depends on where, I mean, I don't think I'm a normal person, so I mean, I'm not one to call other people weird, but if you talk to a typical Python programmer, you're typically not thinking about this, right? This is a lower level of abstraction. Now, if you talk to a C++ programmer, certainly if you talk to a Rust programmer, again, they're not weird, they're delightful. Like these are all good people, right? Those folks will think about all the time, right? And so I look at this as there's a spectrum between very deep low level systems. I'm gonna go poke the bits and care about how they're laid out in memory, all the way up to application and scripting and other things like this. And so it's not that anybody's right or wrong, </s><s>[INST] it's about how do we build one system that scales. [/INST] I don't, so. </s><s>[INST] Yes, yeah. By the way, the idea of an atomic number has been something that always brought me deep happiness because the flip side of that, the idea that threads can just modify stuff asynchronously, just the whole idea of concurrent programming [/INST] is a source of infinite stress for me. Well, so this is where you jump into, again, you zoom out and get out of programming languages or compilers and you just look at what the industry has done. My mind is constantly blown by this, right? And you look at Moore's law. Moore's law has this idea that computers, for a long time, single thread performance just got faster and faster and faster and faster for free. But then physics and other things intervened and power consumption, other things started to matter. And so what ended up happening is we went from single core computers to multi-core, then we went to accelerators. This trend towards specialization of hardware is only gonna continue. And so for years, us programming language nerds and compiler people have been saying, okay, well, how do we tackle multi-core, right? For a while, it was like, multi-core is the future. We have to get on top of this thing. And then it was multi-core is the default. What are we doing with this thing? And then it's like, there's chips with hundreds of cores in them, what happened, right? And so I'm super inspired by the fact that in the face of this, those machine learning people invented this idea of a tensor, right, and what is a tensor? A tensor is an, like an arithmetic and algebraic concept. It's like an abstraction around a gigantic, parallelizable data set, right? And because of that, and because of things like TensorFlow and PyTorch, we're able to say, okay, well, express the math of the system. This enables you to do automatic differentiations, enables you to do like all these cool things. And it's an abstract representation. Because you have that abstract representation, you can now map it onto these parallel machines without having to control, okay, put that right here, put that right there, put that right there. And this has enabled an explosion in terms of AI, compute, accelerators, like all the stuff. And so that's super, super exciting. </s><s>[INST] What happened, right? What about the deployment, the execution across multiple machines? So you write that the modular compute platform dynamically partitions models with billions of parameters and distributes their execution across multiple machines, enabling unparalleled efficiency. By the way, the use of unparalleled in that sentence, anyway, enabling unparalleled efficiency scale and the reliability for the largest workloads. So how do you do this abstraction [/INST] of distributed deployment of large? Yeah, so one of the really interesting tensions. So there's a whole bunch of stuff that goes into that. I'll pick a random walk through it. If you go back and replay the history of machine learning, right, I mean, the brief, the brief, the most recent history of machine learning, because this is, as you know, very deep. I knew Lex when he had an AI podcast. Yes. Right, yep. So if you look at just TensorFlow and PyTorch, which is pretty recent history in the big picture, right? But TensorFlow is all about graphs. PyTorch, I think pretty unarguably ended up winning and why did it win? Mostly because of usability, right? And the usability of PyTorch is, I think, huge. And I think, again, that's a huge testament to the power of taking abstract, theoretical, technical concepts and bringing it to the masses, right? Now, the challenge with what the TensorFlow versus the PyTorch design points was that TensorFlow is kind of difficult to use for researchers, but it was actually pretty good for deployment. PyTorch is really good for researchers. It's kind of not super great for deployment, right? And so I think that we, as an industry, have been struggling. And if you look at what deploying a machine learning model today means is that you'll have researchers who are, I mean, wicked smart, of course, but they're wicked smart at model architecture and data and calculus. And they call it, they're wicked smart in various domains. They don't want to know anything about hardware or deployment or C++ or things like this, right? And so what's happened is you get people who train the model, they throw it over the fence and then you have people that tried to deploy the model. Well every time you have a team A does X, they throw it over the fence, and team B does Y, you have a problem. Because, of course, it never works the first time. And so, you throw it over the fence, they figure out, okay, it's too slow, won't fit, doesn't use the right operator, the tool crashes, whatever the problem is, then they have to throw it back over the fence. And every time you throw a thing over a fence, It takes three weeks of project managers and meetings and things like this. And so, what we've seen today is that getting models in production can take weeks or months. Like, it's not atypical, right? I talk to lots of people and you talk about like, VPS software at some internet company trying to deploy a model, and they're like, why do I need a team of 45 people? It's so easy to train a model. Why can't I deploy it, right? And if you dig into this, every layer is problematic. So if you look at the language piece, I mean, this is the tip of the ice bird, it's a very exciting tip of the iceberg for folks, but you've got Python on one side and C++ on the other side. Python doesn't really deploy. I mean, it can theoretically, technically in some cases, but often a lot of production teams will want to get things out of Python because they get their performance and control and whatever else. So Mojo can help with that. If you look at serving, so you talk about gigantic models. Well, a gigantic model won't fit on one machine, right? And so now you have this model, it's written in Python, it has to be rewritten in C++. Now it also has to be carved up so that half of it runs on one machine, half of it runs on another machine, or maybe it runs on 10 machines. Well, so now suddenly the complexity is exploding, right? And the reason for this is that if you look into TensorFlow, PyTorch, these systems, they weren't really designed for this world, right? They were designed for, you know, back in the day when we were starting and doing things where it was a different, much simpler world, like you want to run ResNet-50 or some ancient model architecture like this, and it was just a completely different world. Train on one GPU, doing something on one GPU. Yeah, AlexNet, right? In the major breakthrough. And the world has changed, right? And so now the challenge is that TensorFlow, PyTorch, these systems, they weren't actually designed for LLMs, like that was not a thing. And so where TensorFlow actually has amazing power in terms of scale and deployment, and things like that, and I think Google is, I mean, maybe not unmatched, but they're like incredible in terms of their capabilities and gigantic scale. Many researchers using PyTorch, right? And so PyTorch doesn't have those same capabilities. And so Modular can do is it can help with that. Now, if you take a step back and say, like, what is Modular doing, right? So Modular has like a AlexNet, a bitter enemy that we're fighting against in the industry. And it's one of these things where everybody knows it, but nobody is usually willing to talk about it. The bitter enemy. The bitter thing that we have to destroy, that we're all struggling with, and it's like fish can't see water, is complexity. Yes, it's complexity. That was very close off. And so if you look at it, yes, it is on the hardware side. Yes. All these accelerators, all these software stacks that go with the accelerator, all these, there's just massive complexity over there. You look at what's happening on the modeling side, massive amount of complexity. Like things are changing all the time. People are inventing... Turns out the research is not done, right? And so people want to be able to move fast. Transformers are amazing, but there's a ton of diversity even within Transformers. And what's the next Transformer, right? And you look into serving. Also huge amounts of complexity. It turns out that all the cloud providers have all their very weird, but very cool hardware. But very cool hardware for networking and all this kind of stuff and it's all very complicated people aren't using that you look at Classical serving right there There's this whole world of people who know how to write high-performance servers with zero copy networking and like all all this fancy Asynchronous IO and like all these fancy things in the in the serving community Very little that has pervaded into the machine learning world Right, and why is that? Well, it's because again these systems have been built up over many years They they haven't been rethought there hasn't been a first principles approach to this. And so modular is doing as we're saying, okay </s><s>[INST] Complexly [/INST] We've built many of these things like so I've worked on TensorFlow and TPUs and things like that other folks on our team like worked on PyTorch core We've worked on onyx one time we've worked on many of these other systems and so built The systems like the Apple accelerators and all that kind of stuff like our team is quite amazing. And so one of the things that Roughly every modular is grumpy about is that when you're working on one of these projects you have a first-order goal Get the hardware to work get the system to enable one more model get this product out the door enable the specific workload or make it Solve this problem for this this product team, right and nobody's been given a chance to actually do that step back And so we as an industry we didn't take two steps forward we took like 18 steps forward in terms of all this really cool technology across compilers and systems and run times and heterogeneous computing like all this kind of stuff and like all this technology has been You know, I wouldn't say Beautifully designed but it's been proven in different quadrants like, you know, you look at Google with TPUs Massive huge exa flops of compute strapped together and into machines that researchers are programming in Python in a notebook That's huge. It's amazing. That's incredible, right? It's incredible. And so you look at the technology that goes into that and The the algorithms are actually quite general and so Lots of other hardware out there and lots of other teams out there don't have the sophistication or that maybe the the years working on it Or the the budget or whatever that Google does, right? And so they should be getting access to same algorithms, but they just don't have that and it's what modular is doing So saying it's just amazing cool. This is not research anymore. Like we've we've built auto tuning in many systems We've built programming languages, right? And so I kept have you know implemented c++ have implemented Swift have implemented many of these things And so, you know this it's hard, but it's not research and you look at accelerators Well, we know there's a bunch of different weird kind of accelerators, but they actually cluster together Right and you look at GPUs. Well, there's a couple of major vendors of GPUs and they maybe don't always get along But their architectures are very similar you look at CPUs CPUs are still super important for the deployment side of things You see new new architectures coming out from all the cloud providers and things like this and they're all super important to the world All right But they don't have the 30 years of development that the entrenched people do right and so what modular can do is we're saying Okay, all this complexity. Like it's not it's not bad complexity. It's actually innovation Right. And so it's innovation that's happening and it's for good reasons But I have sympathy for the poor software people, right? I mean again, i'm a generally a software person, too I love hardware but software people want to build applications and products and solutions that scale over many years They don't want to build a solution for one generation of hardware with one vendors tools And because of this they need something that scales with them. They need, something that works on cloud and mobile, and so I like it because their product manager said, hey, I want it to be, to have lower latency, and it's better for personalization, or whatever they decide, right. Products evolve and so the challenge with the machine learning technology and the infrastructure that we have today in the industry is that it's all these point solutions. And because there are all these point solutions, it means that as your product evolves you have to, to like switch different technology stacks or switch to different vendor. </s><s>[INST] And what that does is that slows down progress. So basically, a lot of the things we've developed in those little silos for machine learning tasks, you want to make that the first-class citizen of a general purpose programming language that can then be compiled [/INST] across all these kinds of hardware. Well, so it's not really about a programming language. I mean, the programming language is a component of the mission, right? And the mission is, or not literal, but our joking mission is to save the world from terrible AI software. Excellent, I love this. So, you know, if you look at this mission, you need a syntax. So that's, so yeah, she needed programming language, right? And like we wouldn't have to build the programming language if one existed, right? So if Python was already good enough, then cool. We would just used it, right? We're not just doing very large scale, expensive engineering projects for the sake of it. Like it's to solve a problem, right? It's also about accelerators. It's also about exotic numerics and B float 16 and matrix multiplications and convolutions and like this kind of stuff. Within the stack, there are things like kernel fusion. That's an esoteric, but really important thing that leads to much better performance and much more general research hackability together, right? </s><s>[INST] And that's enabled by the ASICs. That's enabled by certain hardware. Well, it's like, where's the dance between, I mean, there's several questions here. Like, how do you add a piece of hardware to this stack? If I have this genius invention of a specialized accelerator, how do I add that to the modular framework? And also how does modular as a standard start to define the kinds of hardware [/INST] that should be developed? Yeah, so let me take a step back and talk about status quo, okay? And so if you go back to TensorFlow One, PyTorch One, this kind of timeframe, and these have all evolved and gotten way more complicated. So let's go back to the glorious simple days, right? These things basically were CPUs and CUDA. And so what you do is you say, go do a dense layer and a dense layer has a matrix multiplication in it, right? And so when you say that, you say, go do this big operation of matrix multiplication. And if it's on a GPU, kick off a CUDA kernel. If it's on a CPU, go do like an Intel algorithm or something like that with the Intel MKO, okay? Now, that's really cool if you're either video or Intel, right, but then more hardware comes in, right? And on one axis, you have more hardware coming in. On the other hand, you have an explosion of innovation in AI. And so what happened with both TensorFlow and PyTorch is that the explosion of innovation in AI has led to, it's not just about matrix multiplication and convolution, these things have now like 2000 different operators. And on the other hand, you have, I don't know how many pieces of hardware there are out there, it's a lot, okay? It's not even hundreds, it's probably thousands, okay? And across all of Edge and across like all the different things. That are used at scale. </s><s>[INST] Yeah, exactly, I mean, AI's everywhere, yeah. Yeah, exactly, I mean, AI's everywhere, yeah. [/INST] It's not a handful of TPU alternatives. Correct. It's every phone often with many different chips inside of it from different vendors, right? </s><s>[INST] Like it's, AI is everywhere, it's a thing, right? Why are they all making their own chips? Like why is everybody making their own thing? Well, so because- [/INST] Is that a good thing, for sure? So Chris's philosophy on hardware, right? So my philosophy is that there isn't one right solution, right, and so I think that, again, we're at the end of Moore's law, specialization happens. If you're building, if you're training GPT-5, he wants some crazy supercomputer data center thingy. If you're making a smart camera that runs on batteries, you want something that looks very different. If you're building a phone, you want something that looks very different. If you have something like a laptop, you want something that looks, maybe similar but a different scale, right? And so AI ends up touching all of our lives, robotics, right, and like lots of different things. And so as you look into this, these have different power envelopes, There's different trade-offs in terms of the algorithms. There's new innovations in sparsity and other data formats and things like that. And so hardware innovation, I think is a really good thing, right? And what I'm interested in is unlocking that innovation. There's also like analog and quantum and like all the really weird stuff, right? And so if somebody can come up with a chip that uses analog computing and it's a hundred X more power efficient, think what that would mean in terms of the daily impact on the products we use. That'd be huge. Now, if you're building an analog computer, you may not be a compiler specialist. The really weird stuff, right? These are different skill sets, right? And so you can hire some compiler people if you're running a big company maybe. But it turns out these are really like exotic new generation of compilers. Like this is a different thing, right? And so if you take a step back out and come back to what is the status quo? Status quo is that if you're Intel or you're Nvidia, you continue to keep up with the industry and you chase and okay, there's 1900, now there's 2000, now there's 2100 and you have a huge team of people that are trying to keep up in tune and optimize. And even when one of the big guys comes out with a new generation of their chip, they have to go back and rewrite all these things, right? So really it's only powered by having hundreds of people that are all like frantically trying to keep up. And what that does is that keeps out the little guys and sometimes they're not so little guys, the big guys are also just not in those dominant positions. And so what has been happening, and so a lot of, you talk about the rise of new exotic crazy accelerators, is people have been trying to turn this from a, let's go write lots of special kernels problem into a compiler problem. And so we and I contributed to this as well. We as an industry went into it like, let's go make this compiler problem phase, let's call it. And much of the industry is still in this phase, by the way, so I wouldn't say this phase is over. And so the idea is to say, look, okay, what a compiler does is it provides a much more general, extensible, hackable interface for dealing with the general case, right? And so within machine learning algorithms, for example, people figured out that, hey, if I do a matrix multiplication and I do a ReLU, right? The classic activation function, it is way faster to do one pass over the data and then do the ReLU on the output where I'm writing out the data, because ReLU is just a maximum operation, right? Maximum zero. And so it's an amazing optimization, take matmul ReLU squished together in one operation, now I have matmul ReLU. Well, wait a second. If I do that now, I just went from having, two operators to three, but now I figure out, okay, well, there's a lot of activation functions. What about leaky ReLU? What about like a million things that are out there, right? And so as I start fusing these in, now I get permutations of all these algorithms, right? And so with the compiler people said, is they said, hey, cool. Well, I will go enumerate all the algorithms and I will enumerate all the pairs and I will actually generate a kernel for you. And I think that this has been very, very useful for the industry. This is one of the things that powers Google TPUs, pyTorch2s, like rolling out really cool compiler stuff with Triton and this other technology and things like this. And so the compiler people are kind of coming into their fore and saying like awesome, It's awesome! This is a compiler problem,REL compiler it." Right? Here's the problem, not everybody's a compiler person. I love compiler people, trust me. Right? But not everybody can or should be a compiler person. It turns out that there are people that know analog computers really well, or they know some GPU internal architecture thing really well, or they know some crazy sparse numeric interesting algorithm that is the cusp of research. But they're not compiler people. And so one of the challenges with this new wave of technology trying to turn everything into a compiler, because again, it has excluded a ton of people. And so you look at what does Mojo do? What does the modular stack do? This brings programmability back into this world. Like it enables, I wouldn't say normal people, but a different kind of delightful nerd that cares about numerics or cares about hardware or cares about things like this, to be able to express that in the stack and extend the stack without having to actually go </s><s>[INST] hack the compiler itself. So extend the stack on the algorithm side, [/INST] and then on the hardware side. Yeah, so again, go back to the simplest example of int. And so what both Swift and Mojo and other things like this did is we said, OK, pull magic out of the compiler and put it in the standard library. And so what modular is doing with the engine that we're providing and this very deep technology stack, which goes into heterogeneous runtimes and a whole bunch of really cool things, this whole stack allows that stack to be extended and hacked and changed by researchers and by hardware innovators and by people who know things that we don't know because modular has some smart people, </s><s>[INST] but we don't have all the smart people, it turns out. [/INST] What are heterogeneous runtimes? Yeah, so what is heterogeneous, right? So heterogeneous just means many different kinds of things together. And so the simplest example you might come up with is a CPU and a GPU. And so it's a simple heterogeneous computer to say, I will run my data loading and preprocessing and other algorithms on the CPU. And then once I get it into the right shape, I shove it into the GPU. I do a lot of matrix multiplications and convolutions and things like this. And I get it back out and I do some reductions and summaries and I shove it across the network to another machine. And so you've got now what are effectively two computers, a CPU and a GPU talking to each other, working together in a heterogeneous system. But that was 10 years ago. You look at a modern cell phone. You've got CPUs. And they're not just CPUs. There's big dot little CPUs. And so there's multiple different kinds of CPUs that are, again, working together. They're multi-core. You've got GPUs. You've got neural network accelerators. You've got dedicated hardware blocks for media, so for video decode and JPEG decode and things like this. And so you've got this massively complicated system. And this isn't just cell phones. Every laptop these days is doing the same thing. And all these blocks can run at the same time and need to be choreographed, right? And so, again, one of the cool things about machine learning is it's moving things to like data flow graphs and higher level abstractions and tensors and these things that it doesn't specify. Here's how to do the algorithm. It gives the system a lot more flexibility in terms of how to translate or map or compile it onto the system that you have. And so what you need, you know, at the bottom-est part of the layer there is a way for all these devices to talk to each other. And so this is one thing that, you know, I'm very passionate about, I mean, you know, I'm a nerd. But all these machines and all these systems are effectively parallel computers running at the same time, sending messages to each other. And so they're all fully asynchronous. Well, this is actually a small version of the same problem you have in a data center, right? In a data center, you now have multiple different machines, sometimes very specialized, sometimes with GPUs or TPs in one node and sometimes with disks in another node. So you get a much larger scale heterogeneous computer. And so what ends up happening is you have this like multi-layer abstraction of hierarchical parallelism, hierarchical asynchronous communication. And making that, again, my enemy is complexity by getting that away from being different specialized systems at every different part of the stack and having more consistency and uniformity. I think we can help lift the world </s><s>[INST] and make it much simpler and actually get used. But how do you leverage like the strengths of the different specialized systems? So looking inside the smartphone, like there's just, what, like, I don't know, five, six computers, essentially, inside the smartphone? How do you, without trying to minimize the explicit, making it explicit which computer is supposed to be for which operation? [/INST] For which operation? Yeah, so there's a pretty well-known algorithm. And what you're doing is you're looking at two factors. You're looking at the factor of sending data from one thing to another, right? It takes time to get it from that side of the chip to that side of the chip and things like this. And then you're looking at what is the time it takes to do an operation on a particular block. So, take CPUs. CPUs are fully general. They can do anything. But then you have a neural net accelerator that's really good at matrix multiplications. And so, you say, okay, well, if my workload is all matrix multiplications, I start up, I send the data over the neural net thing, it goes and does matrix multiplications. when it's done, it sends me back the result. All is good. The simplest thing is just saying, do matrix operations over there. Then you realize you get a little bit more complicated, because you can do matrix multiplications on a GPU, you can do it on a neural net accelerator, you can do it on CPU, and they'll have different trade-offs and costs, and it's not just matrix multiplication. What you actually look at is you look at, I have generally a graph of compute, I want to do a partitioning, I want to look at the communication, the bisection bandwidth, and the overhead and the sending of all these different things, and build a model for this, and then decide, okay, it's an optimization problem, where do I want to place this compute? </s><s>[INST] The old school theoretical computer science problem of scheduling. Then how does presumably [/INST] it's possible to somehow magically include autotune into this? Absolutely. In my opinion, so this is an opinion, not everybody would agree with this, but in my opinion, the world benefits from simple and predictable systems at the bottom that you can control. But then, once you have a predictable execution layer, you can build lots of different policies on top of it. One policy can be that the human programmer says, do that here, do that here, do that here, do that here, and fully manually controls everything, and the system should just just do it. Then, you quickly get in the mode of, I don't want to have to tell it to do it. The next logical step that people typically take is they write some terrible heuristic. If it's a major location, do it over there. If it's floating point, do it on the GPU. If it's integer, do it on CPU, something like that. Then you then get into this mode of people care more and more and more, and you say, okay, well, let's actually make the heuristic better. Let's get into auto tuning. is actually do search of the space, to decide, well, what is actually better, right? Well, then you get into this problem where you realize this is not a small space, this is a many dimensional, us reasons dimensional space, that you cannot exhaustively search, so, do you know of any other algorithms that are good at searching very complicated spaces for- </s><s>[INST] Don't tell me you're gonna turn this [/INST] into a machine learning problem. So then you turn into a machine learning problem, And then you have a space of genetic algorithms and reinforcement learning and all these cool things. </s><s>[INST] Can you include that into the stack, into the modular stack? Yeah, yeah. Where does it sit? Where does it live? [/INST] Is it a separate thing or is it part of the compilation? Yeah, yeah. So you start from simple and predictable models. And so you can have full control and you can have coarse-grained knobs like nudged systems, so you don't have to do this. But if you really care about getting the best, last ounce out of a problem then you can use additional tools and they're the cool things you don't want to do this every time you run a model you want to figure out the right answer and then cash it and once you do that you can get you can say okay cool I can get up and running very quickly I can get good execution out of my system I can decide if something's important and if it's important I can go throw a bunch of machines at it and do a big expensive search ever the space using whatever technique I feel like its pretty up to to the problem, and then when I get the right answer, cool, I can just start using it. And so you can get out of this trade-off between, okay, am I gonna spend forever doing a thing, or do I get up and running quickly, and as a quality result, these are actually not in contention with each other if the system's designed to scale. </s><s>[INST] You started and did a little bit of a whirlwind overview of how you get 35,000X speed up or more over Python. Jeremy Howard did a really great presentation about sort of the basic, like, look at the code, here's how you get the speed up. Like you said, that's something we could, probably developers can do for their own code to see how you can get this gigantic speed up. But can you maybe speak to the machine learning task in general, how do you make some of this code fast and specific, like, what would you say is the main bottleneck for machine learning tasks? So are we talking about that mall matrix multiplication, how do you make that fast? [/INST] So, I mean, if you just look the Python problem, right, you can say, how do I make Python faster? There's been a lot of people that have been working on the, okay, how do I make Python 2X faster, 10X faster, or something like that, right? And there've been a ton of projects in that vein, right? Mojo started from the, what can the hardware do, like, what is the limit of physics? What is the speed of light, what is it? Like, how fast can this thing go? And then, how do I express that? Yeah, right. And so it wasn't well, it wasn't anchored relatively on make Python a little bit faster. It's saying cool I know what the hardware can do. Let's unlock that right now when you yeah </s><s>[INST] Yeah, just say how how gutsy that is to be in the meeting and as opposed to trying to see how do we get the improvement? [/INST] It's like what can the physics do? I Mean, maybe I'm a special kind of nerd, but you look at that. What is the limit of physics? How fast can these things go, right? When you start looking at that, typically it ends up being a memory problem. Mm-hmm. Right and so today Particularly with these specialized accelerators. The problem is that you can do a lot of math within them But yet you get bottleneck sending data back and forth to memory whether it be local memory or distant memory or disk or whatever It is and and that that bottleneck particularly is the training sizes get large as you start doing tons of inferences all over the place Like that becomes a huge bottleneck for people, right? So again, what happened is we went through a phase of many years where people took the special case and hand tuned it Tweaked it and tricked out and they knew exactly how the hardware worked and they knew the model and they made it they made it fast Didn't generalize and so you can make you know Resident 50 or some or Alex net or something inception v1 like you can you can do that, right because the models are small They fit in your head Right, but as the models get bigger more complicated as the machines get more complicated it stops working Right, and so this is where things like kernel fusion come in. So what is kernel fusion? This is this idea of saying let's avoid going to memory and let's do that by building a new hybrid Kernel and a numerical algorithm that actually keeps things in the accelerator instead of having to write all the way out to memory All right, what's happened with with these accelerators now is you get multiple levels of memory like in a GPU For example, you'll have global memory and local memory and like all these things If you zoom way into how hardware works the register file is actually a memory so the registers are like an L0 cache and so a lot of Taking advantage of the hardware ends up being full utilizing the full power in all of its capability and this has a number of problems, right one of which is again the complexity of disaster, right? There's too much hardware. Even if you just say let's look at The chips from one line of vendor like Apple or Intel or whatever it is Each version of the chip comes out with new features and they change things so that it takes more time or less time to do different Things and you can't rewrite all the software Whenever a new chip comes out, right? And so this is where you need a much more scalable approach and this is what mojo and what the Modular stack provides is it provides this infrastructure and the system for factoring all this complexity and then allowing people to express algorithms You talk about auto tuning for example Express algorithms in a more portable way so that when a new chip comes out you don't have to rewrite it all So to me like, you know, I kind of joke like what is a compiler? Well, there's many ways to explain that you convert thing a and thing B and you convert source code to machine code Like you can talk about many many Things that compilers do but to me it's about a bag of tricks It's about a system and a framework that you can hang Complexity it's a system that can then generalize and it can work on problems that are bigger than fit in one human's head Right. And so what that means what a good stack and what the modular stack provides is The ability to walk up to it with a new problem and it'll generally work quite well And that's something with a lot of machine learning infrastructure and tools and technologies don't have Typical state-of-the-art today as you walk up particularly if you're deploying if you walk up with a new model You try to push it through the converter and the converter crashes. That's crazy The state of ML tooling today it's not anything that a C programmer would ever accept, right? And it's always been this kind of flaky side of tooling that's never been integrated. Well, and it's been Never worked together and because it's not designed together. It's built by different teams It's built by different hardware vendors is built by different systems it's built by different internet companies They're trying to solve their their problems Right and so that means that we get this fragmented terrible mess of complexity </s><s>[INST] So I mean the specifics of enemy Jeremy showed this. There's the vectorized function, which I guess is Built in to the into mojo Vectorized as he showed us built into the library into the library libraries in a library vectorized paralyzed Which vectorized is more low-level paralyzes higher level. There's the tiling thing Which is how he demonstrated the? [/INST] Autotune I think so so think of think about this in like levels Hierarchical levels of abstraction right and so it at the very if you zoom all the way into a compute problem You have one floating point number right and so then you say okay I want to be I could do things one at a time in an interpreter It's pretty slow right so I can get to doing one one at a time in a compiler I can see then I can get to doing four or eight or sixteen at a time with vectors That's called vectorization Then you can say hey, I have a whole bunch of different You know what a multi-core computer is is this basically a bunch of computers? Right so they're all independent computers that they can talk to each other and they share memory And so now what parallelize does it says okay run multiple instances on different computers? And now they call work together our program right and so what you're doing is you're saying keep going out to the next level out And and as you do that. How do I take advantage of this so tiling is a memory optimization right It says okay Let's make sure that we're keeping the data close to the compute Part of the problem instead of sending it all back and forth through memory every every time. I would load a block And the size of the block size is all that's how you get to the auto tune to make right optimized right optimized Yeah, well so all of these the details matter so much to get good performance This is another funny thing about machine learning and high performance computing that is very different than See compilers we all grew up grew up with where you know If you get a new version of GCC or new version of clang or something like that, you know Maybe something will go 1% faster, right? And so compiler ensures will work really really really hard to get half a percent out of your C code something like that but when you're talking about an accelerator or an AI application or you're talking about these kinds of Algorithms and these are things people used to write in Fortran for example, right if you get it wrong It's not 5% or 1% that could be 2x or 10x Right if you think about it You really want to make use of the full memory you have the cache for example But if you use too much space it doesn't fit in the cache Now you're gonna be thrashing all the way back out to main memory and these can be 2x 10x Major performance differences and so this is where getting these magic numbers and these things right is really actually quite important </s><s>[INST] So you mentioned that Mojo is a superset of Python. Can you run? [/INST] Python code as if it's mojo code. Yes. Yes So and so and this has two sides of it. So mojo is not done yet. So I'll give you disclaimer mojo It's not done yet, but already we see people that take small pieces of Python code move it over They don't change it and you can get 12x speedups like somebody was just tweeting about that yesterday, which is pretty cool Right and again interpreters compilers, right? And so without changing a code without Also, this is not with this is not JIT compiling or do anything fancy. This is just Basic stuff move it straight over now mojo will continue to grow out and as it grows out It will have more and more and more features and our nor stars to be a full superset of Python And so you can bring over Basically arbitrary Python code and have it just work and it may not always be 12x faster But but it should be at least as fast and way faster in many cases. This is the goal, right? Now take time to do that and Python is a complicated language. There's not just the obvious things but there's also Non-obvious things that are complicated like we have to be able to talk to see Python Packages to talk to the C API and there's a bunch of there's a bunch of pieces you have not done yet. Cool </s><s>[INST] You have to any just to make explicit the obvious I mean not be so obvious. I don't you think about it. So, you know to run Python code to run Python code, that means you have to run all the Python packages and libraries. Yeah. Yeah. So that means what? What's the relationship between Mojo and CPython, the interpreter that's presumably would be [/INST] tasked with getting those packages to work and libraries to work? Yeah. Yep. So in the fullness of time, Mojo will solve for all the problems and you'll be able to move Python packages over and run them in Mojo. Without the CPython. I'll CPython someday. Yeah. Right. It's not today, but someday. And that'll be a beautiful day because then you'll get a whole bunch of advantages and you'll get massive speedups and things like this. So you can do that one at a time, right? You can move packages one at a time. Exactly. But we're not willing to wait for that. Python is too important. The ecosystem is too broad. We want to both be able to build Mojo out. We also want to do it the right way without time, like without intense time pressure. We're obviously moving fast, but, and so what we do is we say, okay, well, let's make it </s><s>[INST] so you can import an arbitrary existing package. [/INST] You can do that one at a time, right? Carpetory, including like you write your own on your local disk or whatever it's not, it's not like an arbitrary package and import that using CPython, whatever it's not. CPython already runs all the packages. Right. And so what we do is we built an integration layer where we can actually use CPython. Again, I'm practical, and to actually just load and use all the existing packages as they are. The downside of that is you don't get the benefits of Mojo for those packages. They run as fast as they do in the traditional CPython way, but what that does is that gives you an incremental migration path. If you say, hey, cool, the Python ecosystem is vast. I want all of it to just work, but there's certain things that are really important. If I'm doing weather forecasting or something, well, I want to be able to load all the data. I want to be able to work with it, and then I have my own crazy algorithm inside of it. Normally, I'd write that in C++. If I can write in Mojo and have one system that scales, well, that's way easier to work </s><s>[INST] with. Is it hard to do that, to have that layer that's running CPython, because is there some [/INST] communication back and forth? Yes, it's complicated. This is what we do. We make it look easy, but it is complicated. What we do is we use the CPython existing interpreter, so it's running its own byte codes and that's how it provides full compatibility, and then it gives us CPython objects. We use those objects as is. That way, we're fully compatible with all the CPython objects and all the, it's not just the Python part, it's also the C packages, the C libraries underneath them, because they're often hybrid, so we can fully run and we're fully compatible with all that. The way we do that is that we have to play by their rules, so we keep objects in that representation when they're coming from that world. What's the representation that's being used in memory? We'd have to know a lot about how the CPython interpreter works. It has, for example, reference counting, but also different rules on how to pass pointers around and things like this. Super low-level fiddly, and it's not like Python. It's like how the interpreter works. And so that gets all exposed out, and then you have to define wrappers around the low-level C code. And so what this means is you have to know not only C, which is a different world from Python, obviously, not only Python, but the wrappers. But the interpreter, and the wrappers, and the implementation details, and the conventions, and it's just this really complicated mess. And when you do that, now suddenly you have a debugger that debugs Python. They can't step in the C code, so you have this two-world problem. And so by pulling this all into Mojo, what you get is you get one world. You get the ability to say, cool, you have untyped, very dynamic, beautiful, simple code. OK, I care about performance for whatever reason. There's lots of reasons you might care. And so then you add types. You can parallelize things. You can vectorize things. You can use these techniques, which are general techniques, to solve a problem. And then you can do that by staying in the system. And if you have that one Python package that's really important to you, you can move it to Mojo. You get massive performance benefits on that, and other advantages. If you like exact types, it's nice if they're enforced. Some people like that, rather than being hints. So there's other advantages, too. </s><s>[INST] And then you can do that incrementally as you go. So one different perspective on this would be, why Mojo instead of making C Python faster or redesigning C Python? [/INST] Yeah, well, I mean, you could argue Mojo is redesigning C Python. But why not make C Python faster and better and other things like that? There's lots of people working on it. So actually, there's a team at Microsoft that is really improving. I think C Python 3.11 came out in October or something like that, and it was 15% faster, 20% faster across the board, which is pretty huge, given how mature Python is and things like this. And so that's awesome. I love it. Doesn't run on GPU. It doesn't do AI stuff. It doesn't do vectors. It doesn't do things. 20% is good. 35,000 times is better. So I'm a huge fan of that work, by the way, and it composes well with what we're doing. And so it's not like we're fighting or anything like that. It's actually just general goodness for the world. But it's just a different path. And again, we're not working forwards from making Python a little bit better. </s><s>[INST] We're working backwards from what is the limit of physics. What's the process of porting Python code to Mojo? Is there what's involved in that in the process? [/INST] Is there tooling for that? Not yet. So we're missing some basic features right now. And so we're continuing to drop out new features on a weekly basis. But at the fullness of time, give us a year and a half, maybe two years. Is it an automatable process? </s><s>[INST] So when we're ready, it will be very automatable, yes. Is it automatable? Is it possible to automate, in the general case, the Python to Mojo conversion? [/INST] Yeah, well, you're saying it's possible. Yeah, well, you're saying it's possible. Well, so, and this is why, I mean, among other reasons, why we use tabs, right? So first of all, by being a superset, it's like C versus C++. Can you move C code to C++? Yes. Yeah, right. And you can move C code to C++. And then you can adopt classes. You can adopt templates. You can adopt other references or whatever C++ features you want after you move C code to C++. You can't use templates in C. And so if you leave it to C, fine, you can't use the cool features, but it still works. And C and C++ code work together. And so that's the analogy. Now, here, there's not a Python is bad and then Mojo is good. Mojo just gives you superpowers. And so if you want to stay with Python, that's cool. But the tooling should be actually very beautiful and simple because we're doing the hard work of defining </s><s>[INST] a superset. Yes, right. So you're right. So there's several things to say there, but also the conversion tooling should probably [/INST] give you hints as to how you can improve the code. And then exactly once you're in the new world, then you can build all kinds of cool tools to say, hey, should you adopt this feature? And we haven't built those tools yet, but I fully expect those tools will exist. And then you can like, quote unquote, modernize your code or however you want to look at it. So I mean, one of the things that I think is really interesting about Mojo is that there have been a lot of projects to improve Python over the years. Everything from getting Python to run on the Java virtual machine, PyPy, which is a JIT compiler. There's tons of these projects out there that have been working on improving Python in various ways. They fall into one of two camps. So PyPy is a great example of a camp that is trying to be compatible with Python. Even there, not really. It doesn't work with all the C packages and stuff like that. But they're trying to be compatible with Python. There's also another category of these things where they're saying, well, Python is too complicated. And I'm going to cheat on the edges. And integers in Python can be an arbitrary size integer. If you care about it going fast in a register in a computer, that's really annoying. And so you can choose to pass on that. You can say, well, people don't really use big integers that often. Therefore, I'm going to just not do it, and it will be fine. Not a Python superset. Or you can do the hard thing and say, OK, this is Python. You can't be a superset of Python without being a superset of Python. And that's a really hard technical problem. But it's, in my opinion, worth it. And it's worth it because it's not about any one package. It's about this ecosystem. It's about what Python means for the world. And it also means we don't want to repeat the Python 2 to Python 3 transition. We want people to be able to adopt this stuff quickly. </s><s>[INST] And so by doing that work, we can help lift people. Yeah, it's a really interesting technical philosophical challenge of really making a language a superset of another language. [/INST] It's breaking my brain a little bit. Well, it paints you in the corners. So again, I'm very happy with Python. So all joking aside, I think that the indentation thing is not the actual important part of the problem. But the fact that Python has amazing dynamic metaprogramming features, and they translate to beautiful static metaprogramming features, I think is profound. I think that's huge. And so Python, I've talked with Guido about this. It was not designed to do what we're doing. That was not the reason they built it this way. But because they really cared and they were very thoughtful about how they designed the language, it scales very elegantly in this space. But if you look at other languages, for example, C and C++, if you're building a superset, you get stuck with the design decisions of the subset. And so C++ is way more complicated because of C in the legacy than it would have been if they would have theoretically designed a from scratch thing. And there's lots of people right now that are trying to make C++ better. And re-syntax C++ is going to be great. We'll just change all the syntax. But if you do that, now suddenly you </s><s>[INST] have zero packages, you don't have compatibility. So what are the, if you could just linger on that, what are the biggest challenges of keeping that superset status? What are the things you're struggling with? [/INST] Is it all boiled down to having a big integer? No, I mean, it's a long-tail weird things. So let me give you a war story. So war story in this space is you go way back in time. Project I worked on is called Clang. Clang, what it is is a C++ parser. And when I started working on Clang, it must have been like 2006 or something. It was 2007, 2006 when I first started working on it. It's funny how time flies. I started that project, and I'm like, OK, well, I want to build a C++ parser for LLVM. It's going to be the work, GCC is yucky. This is me in earlier times. It's yucky. It's on principle. It has all these weird features, all these bugs. It's yucky, so I'm going to build a standard compliant C and C++ parser. It's going to be beautiful. It'll be amazing, well engineered, all the cool things that NGR wants to do. And so I started implementing and building it out, and building it out, and then I got to include standardio.h. And all of the headers in the world use all the GCC stuff. So again, come back away from theory back to reality. I was at a fork in the road. I could have built an amazingly beautiful academic thing </s><s>[INST] that nobody would ever use. [/INST] What are the other things like? Clang. Or I could say, well, it's yucky in various ways. All these design mistakes, accents of history, the legacy. At that point, GCC was over 20 years old, which by the way, now LLVM's over 20 years old. So it's funny how time catches up to you. And so you say, OK, well, what is easier? As an engineer, it's actually much easier for me to go implement long tail compatibility, weird features, even if they're distasteful, and just do the hard work, and figure it out, reverse engineer it, understand what it is, write a bunch of test cases, try to understand behavior. It's way easier to do all that work as an engineer than it is to go talk to all C programmers and get argue with them, and try to get them to rewrite their code, by the way. Yeah. </s><s>[INST] And so it's funny how engineer it understand. Because that breaks a lot more things. [/INST] Yeah. A lot more things. Yeah. And you have realities. Nobody actually even understands how the code works, because it was written by the person who quit 10 years ago. And so this software is kind of frustrating that way. But that's how the world works. </s><s>[INST] Yeah. [/INST] Unfortunately, it can never be this perfect, beautiful thing. Well, there are occasions in which you get to build, like you invent a new data structure, or something like that. Or there's this beautiful algorithm that just makes you super happy. And I love that moment. But when you're working with people, and you're working with code, and Dusty Deck code bases, and things like this, right? It's not about what's theoretically beautiful. It's about what's practical, what's real, what people will actually use. And I don't meet a lot of people that say, </s><s>[INST] I want to rewrite all my code, just for the sake of it. By the way, there could be interesting possibilities. And we'll probably talk about it where AI can help rewrite some code. That might be farther out future. But it's a really interesting one, how that could create more, be a tool in the battle against this monster of complexity that you mentioned. You mentioned Guido, the benevolent dictator for life of Python. What does he think about Mojo? [/INST] Have you talked to him much about it? I have talked with him about it. He found it very interesting. We actually talked with Guido before it launched. And so he was aware of it before it went public. I have a ton of respect for Guido for a bunch of different reasons. You talk about walrus operator. And Guido is pretty amazing in terms of steering such a huge and diverse community. And driving it forward, and I think Python is what it is, thanks to him. And so to me, it was really important starting to work on Mojo to get his feedback and get his input and get his eyes on this. Now, a lot of what Guido was and is, I think, concerned about is, have we not fragment the community? We don't want to Python 2 to Python 3 thing. That was really painful for everybody involved. And so we spent quite a bit of time talking about that. And some of the tricks I learned from Swift, for example. So in the migration from Swift, we managed to not just convert Objective-C into a slightly prettier Objective-C, which we did. We then converted, not entirely, but almost an entire community to a completely different language. And so there's a bunch of tricks that you learn along the way that are directly relevant to what we do. And so this is where, for example, you leverage C-Python while bringing up the new thing. That approach is, I think, proven and comes from experience. And so Guido was very interested in, OK, cool. I think that Python is really his legacy. It's his baby. I have tons of respect for that. Incidentally, I see Mojo as a member of the Python family. We're not trying to take Python away from Guido and from the Python community. And so to me, it's really important that we're a good member of that community. And so I think that, again, you would have to ask Guido this. But I think that he was very interested in this notion of, cool, Python gets beaten up for being slow community. Maybe there's a path out of that. And if the future is Python, look at the far outside case on this. And I'm not saying this is Guido's perspective. But there's this path of saying, OK, well, Python can suddenly go all the places it's never been able to go before. And that means that Python can go even further and can have even more impact on the world. So in some sense, Mojo could be seen as Python 4.0. I would not say that. </s><s>[INST] I think that would drive a lot of people really crazy. [/INST] Because of the PTSD of the 3.0 2.0. I'm willing to annoy people about Emacs versus VIM or about TAF versus Spaces. But not that one. I don't know. That might be a little bit far even for me. My skin may not be that thick. </s><s>[INST] But not that one. TAF versus Spaces. But not that one. But the point is the step to it being a super set and allowing all of these capabilities, I think, is the evolution of a language. It feels like an evolution of a language. So he's interested by the ideas that you're playing with, but also concerned about the fragmentation. So what are the ideas you've learned? What are you thinking about? How do we avoid fragmenting the community? Where the Pythonistas and the, I don't know what to call the Mojo people. The magicians can coexist happily and share code and basically just have these big code bases that [/INST] are using CPython and more and more moving towards Mojo. The magicians, I like it, are moving towards Mojo. Well, so again, these are lessons I learned from Swift. And here we face very similar problems. In Swift, you have Objective-C, super dynamic. They're very different syntax. But you're talking to people who have large-scale code bases. Apple's got the biggest, largest-scale code base of Objective-C code. And so none of the companies, none of the iOS developers, none of the other developers want to rewrite everything all at once, and so you want to be able to adopt things piece at a time. And so a thing that I found that worked very well in the Swift community was saying, OK, cool. And this is when Swift was very young. As you say, OK, you have a million line of code with Objective-C app. Don't rewrite it all. But when you implement a new feature, go implement that new class using Swift. And so now this, it turns out, is a very wonderful thing for an app developer, but it's a huge challenge for the compiler team and the systems people that are implementing this. And this comes back to, what is this trade-off between doing the hard thing that enables scale versus doing the theoretically pure and ideal thing? And so Swift adopted and built a lot of different machinery to deeply integrate with the Objective-C runtime. And we're doing the same thing with Python. Now, what happened in the case of Swift is that Swift as a language got more and more and more mature over time. And incidentally, Mojo is a much simpler language than Swift in many ways, and so I think that Mojo will develop way faster than Swift for a variety of reasons. But as the language gets more mature, in parallel with that, you have new people starting new projects. And so when the language is mature and somebody's starting a new project, that's when they say, OK, cool. I'm not dealing with a million lines of code. I'll just start and use the new thing for my whole stack. Now, the problem is, again, you come back to where communities and where people that work together, you build a new subsystem or a new feature, a new thing, in Swift, or you build a new thing in Mojo, then you want to end up being used on the other side. And so then you need to work on integration back the other way. And so it's not just Mojo talking to Python. It's also Python talking to Mojo. And so what I would love to see, and I don't want to see this next month, but what I want to see over the course of time is I would love to see people that are building these packages, like NumPy, or TensorFlow, or these packages that are half Python, half C++. And if you say, OK, cool, I want to get out of this Python C++ world into a unified world, and so I can move to Mojo, but I can't give up all my Python clients. Because these libraries get used by everybody, and they're not all going to switch all at once, and maybe never. Well, so the way we should do that is we should bend Python interfaces to the Mojo types. And that's what we did in Swift, and it worked great. I mean, it was a huge implementation challenge for the compiler people, but there's only a dozen of those compiler people, and there are millions of users. And so it's a very expensive, capital-intensive, skill-set intensive problem. But once you solve that problem, it really helps adoption, and it really helps the community progressively adopt technologies. And so I think that this approach will work quite well with the Python and the Mojo world. </s><s>[INST] So for a package, port it to Mojo, and then create a Python interface. So how do you just link on these packages, NumPy, PyTorch, TensorFlow? How do they play nicely together? So is Mojo supposed to be? Let's talk about the machine learning ones. Is Mojo kind of vision to replace PyTorch and TensorFlow to incorporate it? [/INST] What's the relationship in this? All right, so take a step back. So I wear many hats. So you're angling it on the Mojo side. Mojo's a programming language, and so it </s><s>[INST] can help solve the C, C++, Python view that's happening. The FireMojo got me. I'm sorry. We should be talking in modular. [/INST] Yes, yes, yes. OK, so the FireMojo is amazing. I love it. It's a big deal. The other side of this is the FireMojo is in service of solving some big AI problems. And so the big AI problems are, again, this fragmentation, this hardware nightmare, this explosion of new potential, but it's not getting felt by the industry. And so when you look at how does the modular engine help TensorFlow and PyTorch, it's not replacing them. In fact, when I talk to people, again, they don't like to rewrite all their code, you have people that are using a bunch of PyTorch a bunch of TensorFlow. They have models that they've been building over the course of many years. And when I talk to them, there's a few exceptions, but, generally, they don't want to rewrite all their code. And so what we're doing is we're saying, OK, well, you don't have to rewrite all your code. What happens is, the modular engine goes in there and goes underneath TensorFlow and PyTorch. It's fully compatible and just provides better performance, better predictability, better tooling. It's a better experience that helps lift TensorFlow and PyTorch and make them even better. I love Python, I love TensorFlow, I love PyTorch, right? This is about making the world better </s><s>[INST] because we need AI to go further. But if I have a process that trains a model and I have a process that performs inference on that model and I have the model itself, what should I do with that in the long arc of history in terms of if I use PyTorch to train it, should I rewrite stuff in Mojo? [/INST] Would that, if I care about performance? Oh, so I mean, again, it depends. So if you care about performance, then writing in a Mojo is gonna be way better than writing in Python. But if you look at LLM companies, for example, you look at OpenAI rumored and you look at many of the other folks that are working on many of these, many of these LLMs and other like innovative machine learning models, on the one hand, they're innovating in the data collection and the model billions of parameters and the model architecture and the RLEHF and the like all the cool things that people are talking about. But on the other hand, they're spending a lot of time writing CUDA kernels. And so you say, wait a second, how much faster could all this progress go if they were not having to hand write all these CUDA kernels? And so there are a few technologies that are out there and people have been working on this problem for a while and they're trying to solve subsets to the problem, again, kind of fragmenting the space. And so what Mojo provides for these kinds of companies is the ability to say, cool, I can have a unifying theory. Right, and again, better together, the unifying theory, the two-world problem or the three-world problem or the n-world problem, like this is the thing that is slowing people down. And so as we help solve this problem, I think it will be very helpful </s><s>[INST] for making this whole cycle go faster. So obviously we've talked about the transition from Objective-C to Swift, have designed this programming language. And you've also talked quite a bit about the use of Swift for machine learning context. Why have you decided to move away from maybe an intense focus on Swift for the machine learning context versus sort of designing a new programming language [/INST] that happens to be a super-service? You're saying this is an irrational set of life choices I make, or what? </s><s>[INST] Did you go to the desert and did you meditate on it? Okay, all right. No, it was bold and needed. And I think, I mean, it's just bold. And sometimes to take those leaps [/INST] is a difficult leap to take. Yeah, well, so, okay. I mean, I think there's a couple of different things. So, actually I left Apple back in 2017, like January of 2017. So it's been a number of years that I left Apple. And the reason I left Apple was to do AI. Okay, so, and again, I won't comment on Apple and AI, but at the time, right, I wanted to get into and understand and understand the technology, understand the applications, the workloads, and so I was like, okay, I'm gonna go dive deep into applied AI and then the technology underneath it, right? </s><s>[INST] I found myself at Google. [/INST] And that was like when TPUs were waking up. Exactly, and so I found myself at Google and Jeff Dean, who's a rock star, as you know, right? And in 2017, TensorFlow is like really taking off and doing incredible things. And I was attracted to Google to help them with the TPUs, right, and TPUs are an innovative hardware accelerator platform, have now, I mean, I think proven massive scale and like done incredible things, right? And so one of the things that this led into is a bunch of different projects, which I'll skip over, right? One of which was this Swift for TensorFlow project, right? And so that project was a research project. And so the idea of that is say, okay, well, let's look at innovative new programming models where we can get a fast programming language, we can get automatic differentiation into language, let's push the boundaries of these things in a research setting, right? Now that project, I think lasted two, three years. There's some really cool outcomes of that. So one of the things that's really interesting is, I published a talk at an LLVM conference in 2018. Again, this seems like so long ago about graph program abstraction, which is basically the thing that's in PyTorch 2. And so PyTorch 2, with all this Dynamo real thing, it's all about this graph program abstraction thing from Python bytecodes. And so a lot of the research that was done ended up pursuing and going out through the industry and influencing things. And I think it's super exciting and awesome to see that. But the Swift for TensorFlow project itself did not work out super well. And so there's a couple of different problems with that. One of which is that you may have noticed Swift is not Python. There's a few people that write Python code. Yes. And so it turns out that all of ML </s><s>[INST] is pretty happy with Python. It's actually a problem that other programming languages have as well, that they're not Python. We'll probably maybe briefly talk about Julia, who's a very interesting, beautiful programming language, [/INST] but it's not Python. Exactly, well, and so like if you're saying, I'm gonna solve a machine learning problem where all the programmers are Python programmers, and you say, the first thing you have to do is switch to a different language. Well, your new thing may be good or bad or whatever, but if it's a new thing, the adoption barrier is massive. It's still possible. Still possible, yeah, absolutely. The world changes and evolves, and there's definitely room for new and good ideas, but it just makes it so much harder. And so lesson learned, Swift is not Python, and people are not always in search of learning a new thing for the sake of learning a new thing, and if you wanna be compatible with all the world's code, turns out, still possible. Meet the world where it is, right? Second thing is that lesson learned is that Swift as a very fast and efficient language, kind of like Mojo, but a different take on it still, really worked well with eager mode, and so eager mode is something that PyTorch does, and it proved out really well, and it enables really expressive and dynamic and easy to debug programming. TensorFlow at the time was not set up for that, </s><s>[INST] let's say. [/INST] The timing is also important in this world. Yeah, yeah, and TensorFlow is a good thing, and it has many, many strengths, but you could say Swift for TensorFlow is a good idea except for the Swift and except for the TensorFlow part. </s><s>[INST] Swift because it's not Python and TensorFlow [/INST] because it's not eager mode at the time. It wasn't set up for eager mode at the time. as 1.0. Exactly. And so one of the things about that is in the context of it being a research project, I'm very happy with the fact that we built a lot of really cool technology, we learned a lot of things. I think the ideas went on to have influence in other systems like PyTorch, few people use that right here, right? And so I think that's super cool. And for me personally, I learned so much from it, right? And I think a lot of the engineers that worked on it also learned a tremendous amount. And so, you know, I think that that's just really exciting to see and, you know, I'm sorry that the project didn't work out. I wish it did, of course, right? But, you know, it's a research project and so you're there to learn from it. </s><s>[INST] Well, it's interesting to think about as 1.0. Exactly. About the evolution of programming as we come up with these whole new set of algorithms in machine learning and artificial intelligence and what's going to win out. Because it could be a new programming language. Yeah. And I mean, I just mentioned Julia. I think there's a lot of ideas behind Julia that Mojo shares. What are your thoughts about Julia in general? [/INST] So I will have to say that when we launched Mojo, one of the biggest things I didn't predict was the response from the Julia community. And so I was not, I mean, let me take a step back. I've known the Julia folks for a really long time. They were an adopter of LLVM a long time ago. They've been pushing state-of-the-art in a bunch of different ways. Julia is a really cool system. I had always thought of Julia as being mostly a scientific computing focused environment. And I thought that was its focus. I neglected to understand that one of their missions is to help make Python work end to end. And so I think that was my air for not understanding that. And so I could have been maybe more sensitive to that. But there's major differences between what Mojo is doing and what Julia is doing. So as you say, Julia is not Python. And so one of the things that a lot of the Julia people came out and said is like, okay, well, if we put a ton of more energy and a ton more money or engineering or whatever into Julia, maybe that would be better than starting Mojo. I mean, maybe that's true, but it still wouldn't make Julia into Python. So if you've worked backwards from the goal of, let's build something for Python programmers without requiring them to relearn syntax, then Julia just isn't there, right? I mean, that's a different thing, right? And so if you anchor on, I love Julia and I want Julia to go further, then you can look at it from a different lens. But the lens we were coming at was, hey, everybody is using Python. Python isn't, syntax isn't broken. Let's take what's great about Python and make it even better. And so it was just a different starting point. So I think Julia is a great language. The community is a lovely community. They're doing really cool stuff, </s><s>[INST] but it's just a slightly different angle. But it does seem that Python is quite sticky. Is there some philosophical almost thing you could say about why Python by many measures seems to be the most popular programming language [/INST] in the world? Well, I can tell you things I love about it. Maybe that's one way to answer the question, right? So huge package ecosystem, super lightweight and easy to integrate. It has very low startup time, right? So it was startup time, you mean like learning curve or what? Yeah, so if you look at certain other languages, you say like go and it just takes like Java, for example, takes a long time to JIT compile all the things and then the VM starts up and the garbage closure kicks in and then it revs its engines and then it can cloud through a lot of internet stuff or whatever, right? Python is like scripting, like it just goes, right? Python has very low compile time. Like, so you're not sitting there waiting. Python integrates into notebooks in a very elegant way that makes exploration super interactive and it's awesome, right? Python is also, it's like almost the glue of computing because it has such a simple object representation, a lot of things plug into it. That dynamic metaprogramming thing we were talking about also enables really expressive and beautiful APIs, right? So there's lots of reasons that you can look at technical things that Python has done and say like, okay, wow, this is actually a pretty amazing thing and any one of those you can neglect. People will all just talk about indentation and ignore like the fundamental things, but then you also look at the community side, right? </s><s>[INST] So Python owns machine learning. So what's startup time, you mean like learning curve [/INST] or what? Machine learning is pretty big? And it's growing. </s><s>[INST] And it's growing and important, right? And there's a reputation of prestige to machine learning to where like, if you're a new programmer, you're thinking about like which programming, which do I use? Well, I should probably care about machine learning. Therefore, let me try Python. [/INST] And builds and builds and builds and builds. And builds and builds and builds. Go back before that. My kids learn Python. Not because I'm telling them to learn Python. What are they going against you or what? Alright, they also learn Scratch. And things like this too, but it's because Python is taught everywhere, because it's easy to learn. </s><s>[INST] And because it's pervasive. I learned"; back to my day. learn Java and C++, well, uphill both directions. But yes, I guess Python is the main language [/INST] of teaching software engineering in schools now. Yeah, well, if you look at this, there's these growth cycles, right? If you look at what causes things to become popular and then gain in popularity, there's reinforcing feedback loops and things like this. And I think Python has done, again, the whole community has done a really good job of building those growth loops and help propel the ecosystem. And I think that, again, you look at what you can get done with just a few lines of code. </s><s>[INST] It's amazing. So this kind of self-building loop, it's interesting to understand because when you look at Mojo, what it stands for, some of the features, it seems sort of clear that this is a good direction for programming languages to evolve in the machine learning community. But it's still not obvious that it will because of this, whatever the engine of popularity, of virality, is there something you could speak to? [/INST] How do you get people to switch? Yeah, well, I mean, I think that the viral growth loop is to switch people to Unicode. I think the Unicode file extensions are what I'm betting on. I think that's gonna be the thing. </s><s>[INST] Yeah, tell the kids that you can use the fire emoji [/INST] and they'd be like, what? Exactly. Exactly. Well, in all seriousness, I think there's really, I'll give you two opposite answers. One is, I hope if it's useful, if it solves problems, and if people care about those problems being solved, they'll adopt the tech. That's kind of the simple answer. And when you're looking to get tech adopted, the question is, is it solving an important problem people need solved? And is the adoption cost low enough that they're willing to make the switch and cut over and do the pain upfront so that they can actually do it? And so hopefully Mojo will be that for a bunch of people and people building these hybrid packages are suffering. It's really painful. And so I think that we have a good shot of helping people. But the other side is, it's okay if people don't use Mojo. It's not my job to say, everybody should do this. I'm not saying Python is bad. I hope Python, CPython, all these implementations, because Python ecosystem is not just CPython. It's also a bunch of different implementations with different trade-offs. And this ecosystem is really powerful and exciting, as are other programming languages. It's not like TypeScript or something is gonna go away. And so there's not a winner-take-all thing. And so I hope that Mojo is exciting and useful to people, </s><s>[INST] but if it's not, that's also fine. But I also wonder what the use case for why you should try Mojo would be. So practically speaking, it seems like, so there's entertainment. There's a dopamine hit of saying, holy shit, this is 10 times faster. This little piece of code is 10 times faster in Mojo. Out of the box before he gets to 35,000. Exactly. I mean, just even that, I mean, that's the dopamine hit that every programmer sort of dreams of is the optimization. It's also the drug that can pull you in and have you waste way too much of your life optimizing and over-optimizing, right? But so what do you see that would be like comedy? It's very hard to predict, of course, but if you look 10 years from now, Mojo's super successful. What do you think would be the thing where people try and then use it regularly [/INST] and it kind of grows and grows and grows and grows and grows? Well, let's say you talk about dopamine hit. And so again, humans are not one thing. And some people love rewriting their code and learning new things and throwing themselves in the deep end and trying out a new thing. In my experience, most people don't. Like they're too busy, they have other things going on. By number, most people don't like this. I want to rewrite all my code. But even those people, the too busy people, the people that don't actually care about the language, just care about getting stuff done, those people do like learning new things, right? And so you talk about the dopamine rush of 10X faster. Wow, that's cool, I want to do that again. Well, it's also like, here's the thing I've heard about in a different domain. And I don't have to rewrite all my code, I can learn a new trick, right? Well, that's called growth, you know? And so one thing that I think is cool about Mojo, and again, this will take a little bit of time for, for example, the blog posts and the books and like all that kind of stuff to develop and the language needs to get further along. But what we're doing, you talk about types. Like you can say, look, you can start with the world you already know and you can progressively learn new things and adopt them where it makes sense. If you never do that, that's cool, you're not a bad person. If you get really excited about it and want to go all the way in the deep end and rewrite everything and like whatever, that's cool, right? But I think the middle path is actually the more likely one where it's, you know, you come out with a new idea and you discover, wow, that makes my code way simpler, way more beautiful, way faster, way whatever. And I think that's what people like. Now, if you fast forward and you said like 10 years up, right, I can give you a very different answer on that, which is, I mean, if you go back and look at what computers look like 20 years ago, every 18 months, they got faster for free, right? 2x faster every 18 months, it was like clockwork, it was free, right? You go back 10 years ago and we entered in this world where suddenly we had multi-core CPUs and we had GPUs. And if you squint and turn your head, what a GPU is, it's just a many core, very simple CPU thing kind of, right? And so, and 10 years ago, it was CPUs and GPUs and graphics. Today, we have CPUs, GPUs, graphics and AI, because it's so important, because the compute is so demanding, because of the smart cameras and the watches and all the different places that AI needs to work in our lives, it's caused this explosion of hardware. And so part of my thesis, part of my belief of where computing goes, if you look out 10 years from now, is it's not gonna get simpler. Physics isn't going back to where we came from, it's only gonna get weirder from here on out, right? And so to me, the exciting part about what we're building is it's about building that universal platform, which the world can continue to get weird, because again, I don't think it's avoidable, it's physics. But we can help lift people scale, do things with it, and they don't have to rewrite their code every time a new device comes out. And I think that's pretty cool. And so if Mojo can help with that problem, then I think that it will be hopefully quite interesting and quite useful to a wide range of people, because there's so much potential and maybe analog computers will become a thing or something, right? And we need to be able to get into a mode where we can move this programming model forward, but do so in a way where we're lifting people and growing them instead of forcing them </s><s>[INST] to rewrite all their code and exploding them. Do you think there'll be a few major libraries that go Mojo first? [/INST] Well, so I mean, the modular engines all Mojo. So again, come back to, we're not building Mojo because it's fun, we're building Mojo </s><s>[INST] because we had to dissolve these accelerators. That's the origin story. [/INST] But I mean, ones that are currently in Python. Yeah, so I think that a number of these projects will. And so one of the things, again, this is just my best guess, like each of the package maintainers also has, I'm sure plenty of other things going on. People like really don't like rewriting code just for the sake of rewriting code. But sometimes like people are excited about like adopting a new idea. Yeah. It turns out that while rewriting code is generally not people's first thing, it turns out that redesigning something while you rewrite it and using a rewrite as an excuse to redesign can lead to the 2.0 of your thing that's way better than the 1.0, right? And so I have no idea, I can't predict that. But there's a lot of these places where again, if you have a package that is half C and half Python, right, you just solve the pain, make it easier to move things faster, make it easier to debug and evolve your tech, adopting Mojo kind of makes sense to start with. And then it gives you this opportunity </s><s>[INST] to rethink these things. Yeah. So the two big gains are that there's a performance gain and then there's the portability [/INST] to all kinds of different devices. And there's safety, right? So you talk about real types. I mean, not saying this is for everybody, but that's actually a pretty big thing, right? Yeah, types are. And so there's a bunch of different aspects of what value Mojo provides. And so, I mean, it's funny for me, like I've been working on these kinds of technologies and tools for too many years now, but you look at Swift, right? And we talked about Swift for TensorFlow, but Swift as a programming language, right? Swift's now 13 years old from when I started it. So, because I started in 2010, if I remember. And so that project, and I was involved with it for 12 years or something, right? That project has gone through its own really interesting story, right? And it's a mature, successful, used by millions of people system, right? Certainly not dead yet, right? But also going through that story arc, I learned a tremendous amount about building languages, about building compilers, about working with community and things like this. And so that experience, like I'm helping channel and bring directly into Mojo. And other systems, same thing. Like apparently I like building, building and iterating and evolving things. And so you look at this LLVM thing that I worked on 20 years ago, you look at MLIR, right? And so a lot of the lessons learned in LLVM got fed into MLIR. And I think that MLIR is a way better system than LLVM was. And Swift is a really good system and it's amazing. But I hope that Mojo will take the next step forward </s><s>[INST] in terms of design, right? Yeah, types are, in terms of running Mojo, people can play with it. What's a Mojo playground? Yeah. And from the interface perspective, and from the hardware perspective, [/INST] what's this incredible thing running on? Yeah, so right now, so here we are two weeks after launch. We decided that, okay, we have this incredible set of technology that we think might be good, but we have not given it to lots of people yet. And so we were very conservative and said, let's put it in a workbook so that if it crashes, we can do something about it. We can monitor and track that, right? And so again, things are still super early, but we're having like one person a minute sign up </s><s>[INST] with over 70,000 people two weeks in, it's kinda crazy. So you can sign up to Mojo Playground [/INST] and you can use it in the cloud, in your browser, in your browser. And so what that's running on is that's running on cloud VMs. And so you share a machine with a bunch of other people, but it turns out there's a bunch of them now because there's a lot of people. And so what you're doing is you're getting free compute and you're getting to play with this thing in kind of a limited, controlled way so that we can make sure that it doesn't totally crash and be embarrassing. So, not a lot of the feedback we've gotten is people want to download it locally. So, we're working on that right now. And so that's... </s><s>[INST] So that's the goal to be able to download [/INST] locally into it? Yeah, that's what everybody expects and so we're working on that right now. And so we just want to make sure that we do it right. And I think this is one of the lessons I learned from Swift also, by the way, is that when we launched Swift, feels like forever ago, it was 2014. It was super exciting. I and we, the team, had worked on Swift for a number of years in secrecy. Four years into this development, roughly, of working on this thing, at that point, about 250 people at Apple knew about it. It was secret. Apple's good at secrecy, and it was a secret project. We launched this at WWDC, a bunch of hoopla on excitement, and said, developers, you're going to be able to develop and submit apps to the app store in three months. Several interesting things happened. First of all, we learned that, hey, it had a lot of bugs. It was not actually production quality, and it was extremely stressful in terms of trying to get it working for a bunch of people. What happened was we went from zero to, I don't know how many developers Apple had at the time, but a lot of developers overnight, and they ran into a lot of bugs, and it was really embarrassing, and it was very stressful for everybody involved. It was also very exciting, because everybody was excited about that. The other thing I learned is that when that happened, roughly every software engineer who did not know about the project at Apple, their head exploded when it was launched, because they didn't know it was coming. They're like, wait, what is this? I signed up to work for Apple because I love objective C. Why is there a new thing? What that meant, practically, is that the push from launch to, first of all, the fall, but then to 2.0 and 3.0 and all the way forward was super painful for the engineering team and myself. It was very stressful. The developer community was very grumpy about it because they're like, okay, well, wait a second. You're changing and breaking my code, and we have to to fix the bugs, and it was just a lot of tension and friction on all sides. There's a lot of technical debt in the compiler because we have to run really fast and you have to go implement the thing and unblock the use case and do the thing. You know it's not right, but you never have time to go back and do it. I'm very proud of the Swift team because they've come, they're heading. I mean, we, but they came so far and made so much progress over this time since launch. It's pretty incredible, and Swift is a very, very good thing, </s><s>[INST] but I just don't want to do that again. I'll iterate more through the development process. [/INST] Yeah, and so what we're doing is we're not launching it when it's, hopefully, 0.9 with no testers. We're launching it and saying it's 0.1, right? And so we're setting expectations of saying, okay, well, don't use this for production, right? If you're interested in what we're doing, we'll do it in an open way, and we can do it together, but don't use it in production yet. We'll get there, but let's do it the right way. I'm also saying, we're not in a race. The thing that I want to do is build the world's best thing. Yeah, right because if you do it right and it lifts the industry it doesn't matter if it takes an extra two months Yeah, like two months is worth waiting. And so doing it right and not being Overwhelmed with technical debt and things like this is like again war wounds Lessons learned whatever you want to say I think is absolutely the right thing to do even though right now people are very frustrated that you know You can't download it or the it doesn't have feature X or something like this </s><s>[INST] and so what have you learned in the in a little bit of time since it's been Released into the wild that people have been complaining about future X or Y or Z What have they been complaining about whether they have been excited about like, yeah almost like detailed things versus a big day [/INST] I think everyone would be very excited about the big vision. Yeah. Yeah. Well, so, I mean I've been very pleased In fact, I mean, we've been massively overwhelmed with response which is a good problem to have. It's kind of like a success disaster in a sense, right? And so, I mean, if you go back in time, when we started Modular, which was just not yet a year and a half ago. So it's still a pretty new company, new team, small but very good team of people. Like we started with extreme conviction that there's a set of problems that we need to solve. And if we solve it, then people will be interested in what we're doing, right? But again, you're building in basically secret, right? You're trying to figure it out. The creation is a messy process. You're having to go through different paths and understand what you want to do and how to explain it. Often when you're doing disruptive and new kinds of things, just knowing how to explain it is super difficult, right? And so when we launched, we hoped people would be excited. But I'm an optimist, but I also don't want to get ahead of myself. And so when people found out about Mojo, I think their heads exploded a little bit, right? And here's, I think, a pretty credible team that has built some languages and some tools before. And so they have some lessons learned and are tackling some of the deep problems in the Python ecosystem and giving it the love and attention that it should be getting. And I think people got very excited about that. And so if you look at that, I mean, I think people are excited about ownership and taking a step beyond Rust, right? And there's people that are very excited about that. And there's people that are excited about, you know, just, like, I made Game of Life go 400 times faster. Right, and things like that. And that's really cool. There are people that're really excited about, OK. I really hate writing stuff in C++. Save me. Like systems in your, they're, like, stepping up like, yes. So that's me, by the way. Also, I don't really want to stop writing C++. </s><s>[INST] stop writing C++, but the. Like systems in your, they're like stepping up like. Yeah. Yes. I get third person excitement when people tweet, here I made this code, game of life or whatever. [/INST] It's faster and you're like. Yeah, and also like, I would also say that, let me cast blame out to people who deserve it. Sure. These terrible people who convinced me to do some of this. Yes. Jeremy Howard. Yes. </s><s>[INST] That guy. Well, he's been pushing for this kind of thing. He's won this for years. He's wanted this for a long. He's won this for years. For people who don't know a Jeremy Howard. He's like one of the most legit people in the machine learning community. He has a grassroots. He really teaches. He's an incredible educator, he's an incredible teacher, but also legit in terms of a machine learning engineer himself. He's been running the fast.ai and looking I think for exactly what. [/INST] Exactly. He's been pushing for more years. Yeah, he's a long, long time. He's one of those three years. And so, fuck. Thank you. Thank you. Exactly. And so, I mean, the first time, so I met Jeremy pretty early on. But the first time I sat up and I'm like, this guy is ridiculous, is when I was at Google and we were bringing up TPUs and we had a whole team of people. And there was this competition called Dawn Bench of who can train ImageNet fastest, right? And Jeremy and one of his researchers crushed Google by not through sheer force of the amazing amount of compute and the number of TPUs and stuff like that, that he just decided that progressive image resizing was the right way to train the model. And if you're an epoch faster and make the whole thing go vroom, right? And I'm like, this guy is incredible. And so you can say, anyways, come back to, you know, where's Mojo coming from? Let's finally listen to Jeremy. </s><s>[INST] It's all his fault. It's all his fault. Well, there's a kind of very refreshing, pragmatic view that he has about machine learning that I don't know if it's like this mix of desire for efficiency, but ultimately grounded in desire to make machine learning more accessible to a lot of people. I don't know what that is. I guess that's coupled with efficiency and performance, but it's not just obsessed about [/INST] performance. Well, so a lot of AI and AI research ends up being that it has to go fast enough to get scale. So a lot of people don't actually care about performance, particularly on the research side, until it allows them to have more, a bigger data set, right? And so suddenly now you care about distributed compute and like all these exotic HPC, like you don't actually want to know about that. You just want to be able to do more experiments faster and do so with bigger data sets, right? And so Jeremy has been really pushing the limits. And one of the things I'll say about Jeremy, and there's many things I could say about Jeremy because I'm a fanboy of his, but he fits in his head. And Jeremy actually takes the time where many people don't to really dive deep into why is the beta parameter of the atom optimizer equal to this, right? And he'll go survey and understand what are all the activation functions in the trade offs and why is it that everybody that does this model pick that thing. So the why, not just trying different values, like really what is going on here? Right. And so as a consequence of that, like he's always, again, he makes time, but he spends time to understand things at a depth that a lot of people don't. And as you say, he then brings it and teaches people and his mission is to help lift his website says, making AI uncool again. Like it's about, like, forget about the hype. It's actually practical and useful. Let's teach people how to do this, right? Now the problem Jeremy struggled with is that he's pushing the envelope, right? Research isn't about doing the thing that is staying on the happy path or the well-paved road, right? And so a lot of the systems today have been these really fragile fragmented things or special case in this happy path. </s><s>[INST] And if you fall off the happy path, you get eaten by an alligator. So what about, uh, so Python has this giant ecosystem of packages, uh, and there's a package repository. Do you have ideas of how to do that? Well, for Mojo, yeah, how to do a repository packages. [/INST] Yeah. Well, so that's another really interesting problem that I knew about, but I didn't understand how big of a problem it was. Uh, Python packaging, a lot of people have very big pain points and a lot of scars with Python packaging. Oh, you mean, uh, so there's several things to say building and distributing and several things to say. Building and distributing and managing dependencies and versioning and all this stuff. </s><s>[INST] So from the, so from the perspective of if you want to create your own package. [/INST] Yes. Yes. And then, or you want to build on top of a bunch of other people's packages and then they get updated and things like this. Now I'm not an expert in this, so I don't know the answer. I think this is one of the reasons why it's great that we work as a team and there's other really good and smart people involved. Um, the, uh, but one of my, one of the things I've heard from smart people who've done a lot of this is that the packaging becomes a huge disaster when you get the Python and C together. And so if you have this problem where you have code split between Python and C now, not only do you have to package the C code, you have to build the C code. C doesn't have a package manager, right? C doesn't have a dependency versioning management system, right? And so I'm not experiencing the state of the art and, uh, all the different Python package managers, but my understanding is that's a massive part of the problem. And I think Mojo solves that part of the problem directly heads on. Now, one of the things I think we'll do with the community, and this isn't, again, we're not solving all the world's problems at once. We have to be kind of focused to start with is that I think that we will have an opportunity to reevaluate packaging, right? And so I think that we can come back and say, okay, well, given the new tools and technologies and the cool things we have that we've built up, because we have not just syntax, we have an entirely new compiler stack that works in a new way. Maybe there's other innovations we can bring together and maybe we can help solve that </s><s>[INST] problem community. So almost a tangent to that question from the user perspective of packages, it was always surprising to me that it was not easier to sort of explore and find packages that, you know, with pip install, it feels, it's an incredible ecosystem, it's just interesting that it wasn't made, it's still, I think, not made easier to discover packages to do, [/INST] like search and discovery, as YouTube calls it. Yeah. Well, I mean, it's kind of funny because this is one of the challenges of these like intentionally decentralized communities. And so I don't know what the right answer is for Python, I mean, there are many people that didn't, I mean, I don't even know the right answer from Mojo. So there are many people that would have much more informed opinions than I do. But it's interesting if you look at this, right, open source communities, you know, there's Git, Git is a fully decentralized, anybody can do it any way they want, but then there's GitHub, right, and GitHub centralized, commercial in that case, right, thing, really help pull together and help solve some of the discovery problems and help build a more consistent community. </s><s>[INST] And so maybe there's opportunities for something like a GitHub for the community. Yeah. Although even GitHub, I might be wrong on this, but the search and discovery for GitHub is not that great. Like I still use Google search. [/INST] Yeah, well, I mean, yeah, well, I mean, maybe that's because GitHub doesn't want to replace Google search. Right. And I think there is room for specialized solutions to specific problems. Sure. I don't know. I don't know the right answer for GitHub either. I just think they can go figure that out. </s><s>[INST] Right. But the point is to have an interface that's usable, that's accessible to people of all different skill levels. [/INST] Well, well, and again, like what are the benefits of standards, right? Standards allow you to build these next level up ecosystem, the next level up infrastructure or next level up things. And so again, come back to hate complexity, C, C plus Python is complicated. It makes everything more difficult to deal with. It makes it difficult to port, move code around, work with, all these things get more complicated. And so, I mean, I'm not an expert, but maybe Mojo can help a little bit by helping reduce the amount of C in this ecosystem and make it therefore scale better. </s><s>[INST] So, I mean, you kind of pack out of packages that are hybrid in nature would be a natural fit to move to Mojo. Which is a lot of them, by the way. Yeah. A lot of them, especially that are doing some interesting stuff, computationalized. Let me ask you about some features. Yeah. So, we talked about, obviously, indentation, that it's a typed language or optionally typed. [/INST] Is that the right way to say it? Which is a lot of them. Yeah. So, either optionally or progressively. Progressively. I think, so people have very strong opinions on the right word to use. Yeah. </s><s>[INST] I don't know. I look forward progressively. Forward to your letters. So, there's the var versus let, but let is for constants, var is an optional. Yeah, var makes it mutable, so you can reassign. Yeah. Okay. Then there's function overloading. Oh, okay. Yeah. I mean, there's a lot of source of happiness for me, but function overloading, that's, I guess, is that for performance or is that, why does Python not have function overloading? [/INST] Oh, okay. Yeah. So, I can speculate. So, Python is a dynamic language. The way it works is that Python and Objective C are actually very similar worlds if you ignore syntax. And so, Objective C is straight line derived from Smalltalk, a really venerable, interesting language that much of the world has forgotten about, but the people that remember it love it generally. And the way that Smalltalk works is that every object has a dictionary in it, and the dictionary maps from the name of a function or the name of a value within an object to its implementation. And so, the way you call a method in Objective C is you say, go look up the way I call foo as I go, look up foo. I get a pointer to the function back, and then I call it. That's how Python works, right? And so, now, the problem with that is that the dictionary within a Python object, all the keys are strings, and it's a dictionary, so you can only have one entry per name. Do you think it's as simple as that? I think it's as simple as that. and so now why do they never fix this? Like why do they not change it to not be a dictionary? Why do they not change, do other things? Well you don't really have to in Python because it's dynamic and so you can say I get into the function, now if I got past an integer, do some dynamic test for it if it's a string, go do another thing. There's another additional challenge which is even if you did support overloading, you're saying, okay well here's a version of a function for integers and a function for strings. So even if you could put it in that dictionary, you'd have to have the caller do the dispatch. And so every time you call the function, you'd have to say is it an integer, is it a string? And so you'd have to figure out where to do that test. And so in a dynamic language, overloading is something you don't have to have. So, but now you get into a type language. And in Python, if you subscript with an integer then you get typically one element out of a collection. subscript with a range, you get a different thing out. And so often in typed languages, you'll want to be able to express the fact that, cool, I have different behavior depending on what I actually pass into this thing. If you can model that, it can make it safer and more predictable and faster and all these </s><s>[INST] things. You think it's as simple as that? It somehow feels safer, yes, but it also feels empowering. Make it in terms of clarity. [/INST] You don't have to design whole different functions. Make it in terms of clarity. Yeah, well, this is also one of the challenges with the existing Python typing systems is that in practice, like you take subscript, like in practice, a lot of these functions, they don't have one signature. They actually have different behavior in different cases. And so this is why it's difficult to retrofit this into existing Python code and make it play well with typing. </s><s>[INST] You kind of have to design for that. Okay, so there's an interesting distinction that people that program Python might be interested in is def versus fn. So it's two different ways to define a function. And fn is a stricter version of def. [/INST] What's the coolness that comes from the strictness? So here you get into what is the trade-off with the superset? Yes. Okay, so superset, you have to or you really want to be compatible. If you're doing a superset, you've decided compatibility with existing code is the important thing, even if some of the decisions they made were maybe not what you choose. So that means you put a lot of time in compatibility, and it means that you get locked into decisions of the past, even if they may not have been a good thing. Now, systems programmers typically like to control things. And they want to make sure that, not in all cases, of course, and even systems programmers are not one thing. But often, you want predictability. And so one of the things that Python has, for example, as you know, is that if you define a variable, you just say x equals 4. I have a variable named x. Now I say some long name equals 17, print out some long name, oops, but I typoed it. Well, the compiler, the Python compiler doesn't know, in all cases, what you're defining and what you're using. And did you typo the use of it or the definition? And so for people coming from typed languages, again, I'm not saying they're right or wrong, but that drives them crazy because they want the compiler to tell them, you typoed the name of this thing, right? And so what FN does is it turns on, as you say, it's a strict mode. And so it says, okay, well, you have to actually intentionally declare your variables before you use them. That gives you more predictability, more error checking and things like this, but you don't have to use it. And this is a way that Mojo is both compatible because defs work the same way that defs have always worked. But it provides a new alternative that gives you more control and allows certain kinds </s><s>[INST] of people that have a different philosophy to be able to express that and get that. [/INST] But usually, if you're writing Mojo code from scratch, you'll be using FN. It depends, again, it depends on your mentality, right? It's not that def is Python and FN is Mojo, Mojo has both and it loves both, right? And it really depends on... Time is just straight. Yeah, exactly. Are you playing around and scripting something out? Is it a one-off throwaway script? Cool. </s><s>[INST] Yeah, Python is great at that. I will still be using FN, but yeah. [/INST] I love strictness. Okay, well, so... Control. Power. You also like suffering, right? Yes. Go hand in hand. How many pull-ups? Yes. </s><s>[INST] Go hand in hand. I've lost count at this point. Exactly. [/INST] So, yeah, exactly. Exactly. And that's cool, I love you for that. And I love other people who like strict things, right? But I don't want to say that that's the right thing because Python's also very beautiful for hacking around and doing stuff and research and these other cases where you may not want that. </s><s>[INST] See, I just... Yeah. I feel like... Maybe I'm wrong on that, but it feels like strictness leads to faster debugging. So, in terms of going from even on a small project from zero to completion, it just... I guess it depends how many bugs you generate usually. [/INST] Yeah. You know, it's, again, lessons learned in looking at the ecosystem. It's really... I mean, I think if you study some of these languages over time, like the Ruby community, for example. Now, Ruby is a pretty well-developed, pretty established community, but along their path, they really invested in unit testing. So, I think that the Ruby community has really pushed forward the state-of-the-art of testing because they didn't have a type system that caught a lot of bugs in compile time. Right? And so, you can have the best of both worlds. You can have good testing and good types and things like this. but I thought that it was really interesting to see how certain challenges get solved. And in Python, for example, the interactive notebook kind of experiences and stuff like this are really amazing. And if you typo something, it doesn't matter, it just tells you, that's fine, right? And so I think that the trade-offs are very different if you're building a large-scale production system </s><s>[INST] versus you're building and exploring in a notebook. The speaking of control, the hilarious thing, if you look at code I write just for myself for fun, it's like littered with asserts everywhere. Okay, okay. It's a kind of. [/INST] Yeah, you'd like types. </s><s>[INST] It's basically saying in a dictatorial way, [/INST] this should be true now, otherwise everything stops. Well, and that is the sign, again, I love you, man, but that is a sign of somebody who likes control. Yeah. And so, yes, I think that you'll like, I think I'll like Mojo. </s><s>[INST] Yeah, I think I'll like Mojo. Therapy session, yes, I definitely will. Speaking of asserts, exceptions are called errors. [/INST] Why is it called errors? So, I mean, we use the same, we're the same as Python, but we implement in a very different way. And so, if you look at other languages, like we'll pick on C++, our favorite, right? C++ has this thing called zero cost exception handling. Okay, and this is, in my opinion, something to learn lessons from. It's a nice polite way of saying it. And so, zero cost exception handling, the way it works is that it's called zero cost because if you don't throw an exception, there's supposed to be no overhead for the non-error code. And so it takes the error path out of the common path. It does this by making throwing an error extremely expensive. And so if you actually throw an error with a C++ compiler using exceptions, it has to go look up in tables on the side and do all this stuff. And so throwing an error could be like 10,000 times more expensive than referring for a function, right? Also, it's called zero cost exceptions, but it's not zero cost by any stretch of the imagination because it massively blows out your code, your binary. It also adds a whole bunch of different paths because of destructors and other things like that that exist in C++. And it reduces the number of optimizations. It adds like all these effects. And so this thing that was called zero cost exceptions, it really ain't, okay? Now, if you fast forward to newer languages and this includes Swift and Rust and Go and now Mojo, well, and Python's a little bit different because it's interpreted. And so like, it's got a little bit of a different thing going on, but if you look at compiled languages, many newer languages say, okay, well, let's not do that zero cost exception handling thing. Let's actually treat and throwing an error the same as returning a variant, returning either the normal result or an error. Now, programmers generally don't want to deal with all the typing machinery and like pushing around a variant. And so you use all the syntax that Python gives us, for example, try and catch, you know, functions that raise and things like this. You can put a raises decorator on your functions, stuff like this, and if you want to control that, and then the language can provide syntax for it. But under the hood, the way the computer executes it, throwing an error is basically as fast as returning something. </s><s>[INST] It's a nice polite way of saying it, nice polite. Oh, interesting. So it's exactly the same way. [/INST] It's from a compiled perspective. And so this is actually, I mean, it's a fairly nerdy thing, right? Which is why I love it. But this has a huge impact on the way you design your APIs, right? So in C++, huge communities turn off exceptions because the cost is just so high, right? And so the zero cost cost is so high, right? And so that means you can't actually use exceptions in many libraries, right? And even for the people that do use it, well, okay, how and when do you want to pay the cost? If I try to open a file, should I throw an error? Well, what if I'm probing around looking for something, right, and I'm looking it up in many different paths? Well, if it's really slow to do that, maybe I'll add another function that doesn't throw an error or turns an error code instead. And I have two different versions of the same thing. And so it causes you to fork your APIs. And so, you know, one of the things I learned from Apple and I so love is the art of API design is actually really profound. I think this is something that Python's also done a pretty good job at in terms of building out this large scale package ecosystem. It's about having standards and things like this. And so, you know, we wouldn't want to enter a mode where, you know, there's this theoretical feature that exists in language, but people don't use it in practice. Now, I'll also say one of the other really cool things about this implementation approach is that it can run on GPUs and it can run on accelerators and things like this. And that standard zero cost exception thing would never work on an accelerator. And so this is also part of how Mojo can scale all the way down to like little embedded systems </s><s>[INST] and to running on GPUs and things like that. Can you actually say about the, maybe, is there some high level way to describe the challenge of exceptions and how they work in code during compilation? [/INST] So it's just this idea of percolating up a thing, an error. Yeah, yeah, so the way to think about it is think about a function that doesn't return anything, just as a simple case, right? And so you have function one calls function two calls function three calls function four along that call stack that are tri-blocks. And so if you have function one calls function two, function two has a tri-block and then within it, it calls function three, right? Well, what happens to function three throws? Well actually, start simpler, what happens if it returns? Well for returns, is it supposed to go back out and continue executing and then fall off the bottom of the tri-block and keep going and it all's good. If the function throws, you're supposed to exit the current function and then get into the accept clause. And then do whatever codes there and then keep following on and going on. And so the way that a compiler like Mojo works is that the call to that function, which happens in the accept block, calls the function. And then, instead of returning nothing, it actually returns a variant between nothing and an error. And so if you return normally, fall off the bottom and do a return, you return nothing. And if you throw an error, you return the variant that is I'm an error. So when you get to the call, you say, okay, cool. I called a function. Hey, I know locally I'm in a tri-block, right? And so I, I call the function and then I check to see what it returns. </s><s>[INST] Aha, if it's that error thing, jump to the accept block. [/INST] And that's all done for you behind the scenes. Exactly. And so the competitors, all this for you. And I mean, one of the things, if you dig into how this stuff works in Python, it gets a little bit more complicated because you have finally blocks, which now you need to go into, do some stuff, and then those can also throw and return. Wait, what? Yeah, in like the stuff matters for compatibility. Um, like there's really, there's, there's with clauses. And so with clauses are kind of like finally blocks with some special stuff going on. </s><s>[INST] And so there's nesting in general, nesting of anything, nesting of functions should be illegal. It just feels like it adds a level of complexity. [/INST] Lex, I'm merely an implementer. And so this is again, one of the, one of the, one of the trade-offs you get when you decide to build a super set is you get to implement a full fidelity implementation of the thing that you decided is good. And so, yeah, I mean, we can, we can complain about the reality of </s><s>[INST] the world and shake our fist, but it always feels like you shouldn't be allowed to do that, like to declare functions and certain functions inside functions. [/INST] Oh, wait, wait, wait, what happened to Lex, the LISP guy? </s><s>[INST] No. I understand that, but LISP is what I used to do in college. So now you've grown up. You know, we've all done things in college, we're not part of... Wait, wait a sec. I love LISP! I love... [/INST] I love LISP. I love LISP! Yeah. I was going to say, you're afraid of me. You're taking a whole internet away. </s><s>[INST] You're taking the whole internet away. I love LISP. [/INST] It worked, it worked as a joke on my head, and to come out, right. So nested functions are joking aside, actually really great. For certain things right and so these are also called closures Closures are pretty cool, and you can pass callbacks. There's a lot of good patterns and so </s><s>[INST] so speaking of which I don't think you have [/INST] Nested functions implemented yet in mojo. Oh, we don't have lambda syntax, but we do have </s><s>[INST] There's a few things on the roadmap they have that it'd be cool to sort of just fly through because it's interesting to see You know how many features there are in a language small and big yep. They have to implement yeah So first of all there's tuple support and that has to do with some very specific aspect of it like the parentheses are not parentheses [/INST] That yep, they have to implement. Yeah. Yeah, this is just a totally a syntactic thing </s><s>[INST] I think that's the thing I think that's the thing okay. There's but it's cool still [/INST] So keyword arguments and functions yeah, so this is where in Python you can say call a function x equals 4 Yeah, and x is the name of the yeah argument. That's a nice sort of documenting self-documenting feature feature Yeah, I mean again. This isn't rocket science to implement. That's just the longer. It's just on the list The bigger features are things like traits So traits are when you want to define abstract so when you get into typed languages You need the ability to write generics and so you want to say I won't write this function and now I want to work on all things that are arithmetic like well, what does arithmetic like mean well arithmetic like is a categorization of a bunch of types and so Again, you can define many different ways and I'm not going to go into ring theory or something But the you know you can say it's arithmetic like if you can add subtract multi divide it for example Right and so what you're saying is you're saying There's a set of traits that apply to a broad variety of types, and so there are all these types of arithmetic-like, all these tensors and floating-point integer, and there's this category of types, and then I can define on an orthogonal axis algorithms that then work against types that have those properties. And so this is a, again, it's a widely known thing. It's been implemented in Swift and Rust and many languages, so it's not Haskell, which is where everybody learns their tricks from, but we need to implement that, and that'll </s><s>[INST] enable a new level of expressivity, a new. So classes. Yeah, classes are a big deal. It's a big deal still to be implemented. Like you said, Lambda syntax, and there's like detailed stuff like whole module import. Yeah, classes are a big deal. Support for top-level code and file scope, and then global variables also, so being able [/INST] to have variables outside of a top-level code. And so this comes back to the where Mojo came from and the fact that this is 0.1, right? And so we're building, so modular is building an AI stack, right? And an AI stack has a bunch of problems working with hardware and writing high-performance kernels and doing this kernel fusion thing I was talking about and getting the most out of the hardware. And so we've really prioritized and built Mojo to solve modular's problem, right? Now, our North Star is build out and support all the things, and so we're making incredible progress. By the way, Mojo's only like seven months old. </s><s>[INST] So that's another interesting thing. I mean, part of the reason I wanted to mention some of these things is like, there's a lot to do, and it's pretty cool how you just kind of, sometimes you take for granted how much there is in a programming language, how many cool features you kind of rely on. [/INST] And this is kind of a nice reminder when you lay it as a to-do list. Yeah, and so, I mean, but also you look into, it's amazing how much is also there, and you take it for granted that a value, if you define it, it will get destroyed automatically. Like that little feature itself is actually really complicated, given the way the ownership system has to work, and the way that works within Mojo is a huge step forward from what Rust and Swift have done. But can you say that again, when a value, when you define it, gets destroyed automatically? Yeah, so like, say you have a string, right? So you just find a string on the stack, whatever that means, like in your local function, right? And so you say, like, whether it be in a def, and so you just say x equals hello world, right? Well, if your string type requires you to allocate memory, then when it's destroyed, you have to deallocate it. So in Python and in Mojo, you define that with the dell method, right? Where does that get run? Well, it gets run sometime between the last use of the value and the end of the program. Like in this, you now get into garbage collection, you get into like all these long debated, you talk about religions and trade offs and things like this, this is a hugely hotly contested world. If you look at C++, the way this works is that if you define a variable, or a set of variables within a function, they get destroyed in a last in first out order. So it's like nesting, okay? This has a huge problem because if you define you have a big scope, and you define a whole bunch of values at the top, and then you use them, and then you do a whole bunch of code that doesn't use them, they don't get destroyed until the very end of that scope, right? And so this also destroys tail calls, so good functional programming, right? This has a bunch of different impacts on, you know, you talk about reference counting optimizations and things like this, a bunch of very low level things. And so what Mojo does is it has a different approach on that from any language I'm familiar where it destroys them as soon as possible. And by doing that you get better memory use, you get better predictability, you get tail calls that work. You get a bunch of other things, you get better ownership tracking. There's a bunch of these very simple things that are very fundamental, that are already built in there in MOJO today, that are the things that nobody talks about generally but when they don't work right you find out and you have to complain about. </s><s>[INST] Is it trivial to know what's the soon as possible [/INST] to delete a thing that's not gonna be used again? Yeah, I mean its generally trivial, it's after the last use of it. So if you find X as a string and then you have some use of X somewhere in your code. </s><s>[INST] What doesqt mean within scope? What do you mean within scope this accessible? [/INST] Yeah, exactly. So you can only use something within its scope. And so then it doesn't wait until the end of the scope to delete it, </s><s>[INST] it destroys it after the last use. So there's kind of some very eager machine [/INST] that's just sitting there and deleting. Yeah, it's all on the compiler so it's not at runtime, which is also cool. And so yeah, and so what and this is actually non trivial because you have control flow, and so it gets complicated pretty quickly and so like getting this right. </s><s>[INST] Was not, and so it gets Christina, it's got this like in a lot of places [/INST] reason potentially compiler has reason about this and this is where again, it's experience building languages and not getting this, right? So, again you get another chance to do it and you get, basic things like this, right? But it's, it's extremely powerful when you do that. Right. And so there's a bunch of things like that, that kind of combine together. And this comes back to the, you get a chance to do it the right way, do it the right way, and make sure that every brick you put down is really good. So that when you put more bricks on top of it, they stack up to something that's beautiful. Listen, this is where... </s><s>[INST] Well, there's also, like how many design discussions do there have to be about particular details, like implementation of particular small features? Because the features that seem small, I bet some of them might be like really require really big design decisions. [/INST] Yeah, yeah. Well, so I mean, let me give you another example of this. Python has a feature called async await. So it's, it's a new feature. I mean, in the long arc of history, it's a relatively new feature, right? That allows way more expressive asynchronous programming. Okay. Again, this is, this is a, Python is a beautiful thing, and they did things that are great for mojo for completely different reasons. The reason the async await got added to Python, as far as I know, is because Python doesn't support threads. Okay. And so Python doesn't support threads, but you want to work with networking and other things like that that can block. I mean, Python does support threads, it's just not its strength. And so, and so they added this feature called async await. It's also seen in other languages like Swift and JavaScript and many other places as well. async await and mojo is amazing, because we have a high performance header, gs, compute, runtime underneath the covers, that then allows non-blocking IO so you get full use of your accelerator. That's huge. Turns out, it's actually really an important part of fully utilizing the machine. You talk about design discussions. That took a lot of discussions, right? And it probably will require more iteration. And so my philosophy with mojo is that, you know, we have a small team of really good people that are pushing forward, and they're very good at the extremely deep knowing how the compiler and runtime and like all the low level stuff works together. But they're not perfect. Same thing as the Swift team, right? And this is where one of the reasons we released mojo much earlier is so we can get feedback. And we've already like renamed a keyword, did a community feedback and we use an ampersand and now it's named in out. We're not we're not renaming existing Python keywords because that breaks compatibility, right? We're naming things we're adding and making sure that they are designed well, we get usage experience, we iterate and work with the community because, again, if you scale something really fast and everybody writes all their code and they start using it in production, then it's impossible to change. And so you want to learn from people, you want to iterate and work on that early on. And this is where design discussions, it's actually quite </s><s>[INST] important. Could you incorporate an emoji like into the language, into the main [/INST] language? Like a good, like a good, like a, like a, do you have a, do you have a </s><s>[INST] favorite one? Well, I really like in terms of humor, like raw, full, whatever, rolling on the floor, laughing. So that could be like, what would that be the use case for that? Like an exception, throw an exception of some sort? I don't know. You should totally file feature requests. Or maybe a hard one. It has to [/INST] be a hard one. People have told me that I'm insane. So this is, this is, I'm </s><s>[INST] liking this. I'm going to, I'm going to use the viral nature of the internet to [/INST] actually get this, to get this past. I mean, it's funny, you come back to the flame emoji file extension, right? The, you know, we have the option to use the flame emoji, which just even that concept cause, for example, the people that get </s><s>[INST] up to say, now I've seen everything. Yeah. There's something, uh, it kind of, it's reinvigorating. It's like, uh, it's like, Oh, that's possible. That's really cool. [/INST] That for some reason that makes everything else like, I think the world is ready for this stuff. And so, you know, when we have a package manager, we'll clearly have to innovate by having the compiled package thing be the little box </s><s>[INST] with the bow on it. Right. I mean, it has to be done. It has to be done. It has to be done. Is there some stuff on the roadmap that you particularly stressed [/INST] about or excited about that you're thinking about a lot? I mean, as a today snapshot, which will be absolutely tomorrow, uh, the lifetime stuff is really exciting. And so lifetimes give you safe references to memory without dangling pointers. And so this has been done in languages like Rust before. And so we have a new approach, which is really cool. I'm very excited about that. That'll be out to the community very soon. Um, the traits feature is really a big deal. And so that's blocking a lot of API design. And so there's that. I think that's really exciting. Um, a lot of it is these kind of table stakes features. Um, one of the things that is, uh, again, also lessons learned with Swift, uh, is that, uh, programmers in general like to add syntactic sugar. And so it's like, Oh, well this annoying thing, like, like in Python, you have to spell unbar, unbar add. Why can't I just use plus deaf plus? Come on. Why can't I just do that? Right. And so trigger a bit of syntactic sugar. It makes sense. It's beautiful. It's obvious. We're trying not to do that. And so, um, for two different reasons, one of which is that again, lesson learned with Swift. Swift has a lot of syntactic sugar, um, which may or may be a good thing. Maybe not. I don't know. But, um, but because it's such an easy and addictive thing to do sugar, like make sure blood get crazy, right? Um, like the community will really dig into that and want to do a lot of that. And I think it's very distracting from building the core abstractions. The second is we want to be a good member of the Python community, right? And so we want to work with the broader Python community and yeah, we're pushing forward a bunch of systems programming features and we need to build them out to understand them. But once we get a long ways forward, I want to make sure that we go back to the Python community and say, okay, let's do some design reviews. Let's actually talk about this stuff. Let's figure out how we want this stuff all to work together. And syntactic sugar just makes all that more complicated. So, </s><s>[INST] and, uh, yeah, list comprehensions are yet to be implemented. And my favorite did, I mean, dictionaries. [/INST] Yeah. There's some, there's some basic, there's some, there's some basic, </s><s>[INST] 0.1, 0.1. [/INST] Yeah. But nonetheless, it's actually so quite interesting and useful. </s><s>[INST] As you mentioned, modular is very new. Mojo is very new. It's a relatively small team. Yeah. There's building up this gigantic stack, this incredible stack that's going to perhaps define the future of development of our AI overlords. We just hope it will be useful. As do all of us. So what, uh, what have you learned from this process of building up a team? Maybe one question is, how do you hire great programmers, great people that operate in this compiler hardware, machine learning, software interface design space? Yeah. And maybe are a little bit fluid in what they can do. So, okay. So language [/INST] design, yeah, we're just, so, okay. So, so building a company is just as interesting in different ways as building a language, like different skill sets, different things, but super interesting. And I've built a lot of teams, a lot of different places. Um, if you zoom in from the big problem into recruiting, well, so here's our problem. Okay. I'll just, I'll be very straightforward about this. We started modular with a lot of conviction about, we understand the problems. We understand the customer pain points. We need to work backwards from the suffering in the industry. And if we solve those problems, we think it'll be useful for people. But the problem is, is that the people we need to hire, as you say, are all these super specialized people that have jobs at big tech, big tech worlds, right? And, you know, we, I don't think we have, um, product market fit in the way that a normal startup does. And we don't have product market fit challenges because right now everybody's using AI and so many of them are suffering and they want help. And so again, we started with strong conviction. Now, again, you have to hire and recruit the best and the best all have jobs. And so what we've done is we said, okay, well let's build an amazing culture. Start with that. That's usually not something a company starts with. Usually you hire a bunch of people and then people start fighting and it turns into a gigantic mess. And then you try to figure out how to improve your culture later. My co-founder, Tim, in particular, is super passionate about making sure that that's right. And we've spent a lot of time early on to make sure that we can scale. </s><s>[INST] Can you comment, sorry, before we get to the second, what makes for a good culture? [/INST] Yeah. Um, so, I mean, there's many different cultures and I have learned many things from several very unique, almost famously unique cultures. And some of them I learned what to do and some of them I learned what not to do. Okay. And so, um, we want an inclusive culture. Uh, I believe in to do. Like amazing people working together. And so I've seen cultures where you have amazing people and they're fighting each other. I see amazing people and they're told what to do like doubt shalt line up and do what I say it doesn't matter if it's the right thing do it right and neither of these is the and I've seen people that have no direction they're just kind of floating in different places and they want to be amazing they just don't know how and so a lot of it starts with have a clear vision right and so we have a clear vision of what we're doing and so I kind of grew up at Apple in my engineering life right and so a lot of the Apple DNA rubbed off on me my co-founder Tim also is like a strong product guy and so what we learned is you know I decided Apple that you don't work from building cool technology you don't work from like come up with a cool product and think about the features you'll have in the big check boxes and stuff like this because if you go talk to customers they don't actually care about your product they don't care about your technology what they care about is their problems right and if your product can help solve their problems well hey they might be interested in that right and so if you speak to them about their problems if you understand you have compassion you understand what people are working with, then you can work backwards to building an amazing product. So the vision starts by defining the problem. And then you can work backwards in solving technology. And at Apple, I think pretty famously said that there's 100 no's for every yes. I would refine that to say that there's 100 not yes for every yes. But famously, if you go back to the iPhone, for example, iPhone 1, many people laughed at it because it didn't have 3G, it didn't have copy and paste. And then a year later, okay, finally, it has 3G, but it still doesn't have copy and paste. It's a joke. Nobody will ever use this product, blah, blah, blah, blah, blah, blah, blah. Right? Well, year three, it had copy and paste and people stop talking about it. And so being laser focused and having conviction, understanding what the core problems are and giving the team the space to be able to build the right tech is really important. And also, I mean, you come back to recruiting. You have to pay well, right? So we have to pay industry leading salaries and have good benefits and things like this. That's a big piece. We're a remote first company. And so we have to... So remote first has a very strong set of pros and cons. On the one hand, you can hire people from wherever they are. And you can attract amazing talent, even if they live in strange places or unusual places. On the other hand, you have time zones. On the other hand, you have everybody on the internet will fight if they don't understand each other. And so we've had to learn how to have a system where we actually fly people in and we get the whole company together periodically and then we get work groups together and we plan and execute together. </s><s>[INST] And there's like an intimacy to the in-person brainstorming that I guess you lose, but maybe you don't, maybe if you get to know each other well and you trust each other, maybe you can do that. [/INST] Yeah. Well, so when the pandemic first hit, I mean, I'm curious about your experience too. The first thing I missed was having whiteboards, right? Those design discussions where like I can high intensity work through things, get things done, work through the problem of the day, understand where you're on, figure out and solve the problem and move forward. But we figured out ways to work around that now with all these screen sharing and other things like that that we do. The thing I miss now is sitting down at a lunch table with the team, the spontaneous things, like the coffee bar things and the bumping into each other and getting to know people outside of the transactional solve a problem over Zoom. </s><s>[INST] And I think there's just a lot of stuff that I'm not an expert at this. I don't know who is, hopefully there's some people, but there's stuff that somehow is missing on Zoom. Even with the whiteboard, if you look at that, if you have a room with one person at the whiteboard and there's like three other people at a table, first of all, there's a social aspect of that where you're just shooting the shit a little bit, almost like... [/INST] Yeah, yeah, and some people are just kind of coming in and... </s><s>[INST] Yeah, that. But also while it's a breakout discussion that happens for like seconds at a time, maybe an inside joke, it's like this interesting dynamic that happens that Zoom. You're bonding. Yeah, you're bonding, but through that bonding, you get the excitement. There's certain ideas are like complete bullshit and you'll see that in the faces of others that you won't see necessarily on zoom Like something it feels like that should be possible to do [/INST] Without being in person. Well, I mean you're bonding. Yeah. Yeah. Well, I mean being in person is a very different thing Yeah, I don't it's worth it, but you can't always do it. And so again, we're still learning We're also learning is like a humanity with this new reality, right? But but what we found is that getting people together whether it be a team or the company or whatever Is it worth the expense because people work together and are happier After that like it just it just like there's a massive period of time where you like go out and things start getting afraid Put people together and then you realize that we're all working together. We see things the same way We work through the disagreement or the misunderstanding we're talking across each other and then you work much better together And so things like that. I think are really quite important It's worth it. </s><s>[INST] What about people that are kind of specialized in very different aspects of the stack working together? [/INST] What are some interesting challenges there? Yeah, well, so I mean, there's lots of interesting people, as you can tell. I'm hard to deal with, too. You're one of the most lovable people. So there's different philosophies in building teams. For me, </s><s>[INST] You're one of the most lovable people. [/INST] And so some people say, hire 10x programmers! And that's the only thing that... whatever that means, right? What I believe in is building well-balanced teams. Teams that have people that are different in them. Like if you have all generals and no troops, or all troops and no generals, or you have all people that think in one way and not the other way, what you get is you get a very biased and skewed and weird situation where people end up being unhappy. And so what I like to do is I like to build teams with people where they're not all the same. You know, we do have teams that are focused on like runtime or compiler GPU or Excel or whatever the speciality is But people bring a different take and have a different perspective And I look for people that complement each other and particularly if you look at leadership teams and things like this You don't want everybody thinking the same way. You want people bringing different perspectives and experiences </s><s>[INST] And so I think that's really important that's team. But what about building a company's ambitious as modular? So what? What's the questions there? [/INST] Interesting questions there. Oh, I mean so many like so one of the things I love about okay So modular is the first company I built from scratch One of the first things that was profound was I'm not cleaning up somebody else's mess Right. And so if you look at that's liberating just something. It's super liberating and and And also many of the projects I've built in the past have not been core to the product of the company That's a liberating to something. It's Swift is not Apple's product Right. M. L. A. R. Is not Google's revenue machine or whatever. It's not it's it's important But it's like working on the accounting software for you know, the the retail giant or something, right? It's it's it's like enabling infrastructure and technology. And so modular the the tech we're building is Here to solve people's problems like it is directly the thing we're giving to people And so this is a really big difference and what it means for me as a leader But also for many of our engineers is they're working on the thing that matters Mm-hmm, and that's actually pretty I mean again for for a compiler people and things like that That's that's usually not the case right, and so that's that's also pretty exciting and in quite nice, but the one of the ways that this Manifest is it makes it easier to make decisions. And so one of the challenges I've had in other worlds is it's like, okay Well community matters Somehow for the goodness of the world like or open source matters theoretically But I don't want to pay for a t-shirt right or some swag like well t-shirts cost ten bucks each You can have a hundred t-shirts for a thousand dollars to a megacorp a thousand dollars is Uncountably can't count that low </s><s>[INST] Right, but justifying it and getting a t-shirt. By the way, if you'd like a t-shirt, why would 100% [/INST] Like a t-shirt. Are you joking? Fire emoji t-shirt. So I will treasure this I'll pass it down to my grandchildren. And so, you know, it's very liberating to be able to side I think that last should have a t-shirt Right, and it becomes very simple </s><s>[INST] This this this is awesome So I have to ask you about the one of the interesting developments with large language models Is that they're able to generate? code recently really well Yes to a degree that maybe I be I don't know if you understand but I have I I struggle to understand because it it forces me to ask questions about the nature of programming of the nature of thought because the Language models are able to predict the kind of code. I was about to write so well Yeah, that it makes me wonder like how unique my brain is and where the valuable ideas actually come from like how much do I contribute in terms of Ingenuity Innovation to code I write or design and that kind of stuff When you stand on the shoulders of giants that you're really doing anything and what LLMs are Helping you do is they help you stand on the shoulders of giants in your program. There's mistakes They're interesting that you learn from but I just it would love to get your opinion first high level Yeah of what you think about this impact of larger language models when they do program synthesis when they generate code [/INST] Yeah, yeah. Well, so I don't know where it all goes Yeah I'm an optimist and I'm a human optimist right I think that things I've seen are that a lot of the LLMs are really good at crushing leak code projects and They can reverse the link list like crazy well, it turns out there's a lot of Instances of that on the internet and it's a pretty stock thing. And so if you want to see Standard questions answered lMS can memorize all the answers and that can be amazing and also they do generalize that from that And so there's good work on that but But I think that if you in my experience building things building something like you talk about mojo, or you talk about these things you talk about Building an applied solution to a problem. It's also about working with people It's about understanding the problem. What is the product? You want to build? What are the use case? What are the customers? You can't just go survey all the customers because they'll tell you that they want a faster horse. Maybe they need a car, right? And so a lot of it comes into, you know, I don't feel like we have to compete with LLMs. I think they'll help automate a ton of the mechanical stuff out of the way. And just like, you know, I think we all try to scale through delegation and things like this. Delegating things to an LLM, I think is an extremely valuable approach that will help us all scale and be more productive. But I think it's a fascinating companion. But I'd say I don't think that that means that we're going to be done with coding. Sure. </s><s>[INST] But I think it's a fascinating companion, but there's power in it as a companion. And from there, I would love to zoom in on to Mojo a little bit. Do you think about that? Do you think about LLMs generating Mojo code and helping sort of like, when you design [/INST] your programming language, it almost seems like, man, it would be nice to sort of a little </s><s>[INST] bit, almost as a way to learn how I'm supposed to use this thing for them to be trained on [/INST] some of the Mojo code. So I do lead an AI company, so maybe there will be a Mojo LLM at some point. But if your question is like, how do we make a language to be suitable for LLMs? Yeah. I think the cool thing about LLMs is you don't have to. And so if you look at what is English or any of these other terrible languages that we as humans deal with on a continuous basis, they're never designed for machines. And yet, they're the intermediate representation. They're the exchange format that we humans use to get stuff done. Right? And so these programming languages, they're an intermediate representation between the human and the computer or the human and the compiler, roughly. </s><s>[INST] And so, I think the LLMs will have no problem learning whatever keyword we pick. Maybe the FiEmoji is gonna- Oh, maybe that's gonna break it. Doesn't tokenize? The reverse of that, it will actually enable it because one of the issues I could see with being a super set of Python is there'll be confusion about the gray area. [/INST] So it'll be mixing stuff. Maybe that's gonna break it. Doesn't tokenize? Well, I'm a human optimist. I'm also an LLM optimist. I think that we'll solve that problem. We'll solve the edgy bit. But you look at that and you say, okay, well, reducing the rote thing, right? Turns out compilers are very particular and they really want things, they really want the indentation to be right. They really want the colon to be there on your else or else it'll complain, right? I mean, compilers can do better at this, but LLMs can totally help solve that problem. And so I'm very happy about the new predictive coding and copilot type features and things like this, </s><s>[INST] because I think it'll all just make us more productive. It's still messy and fuzzy and uncertain, unpredictable. But is there a future you see, given how big of a leap GPT-4 was, where you start to see something like LLMs [/INST] inside a compiler, or no? I mean, you could do that. Yeah, absolutely. I mean, I think that'd be interesting. Is that wise? Well, I mean, it would be very expensive. So compilers run fast and they're very efficient and LLMs are currently very expensive. There's on-device LLMs and there's other things going on. And so maybe there's an answer there. I think that one of the things that I haven't seen enough of is that, so LLMs to me are amazing when you tap into the creative potential of the hallucinations, right? And so if you're building doing creative brainstorming or creative writing or things like that, the hallucinations work in your favor, if you're writing code that has to be correct, because you're going to ship it in production, then maybe that's not actually a feature. And so, I think that there has been research and there has been work on building algebraic reasoning systems and kind of figuring out more things that feel like proofs. And so I think that there could be interesting work in terms of building more reliable scale systems and that could be interesting. But if you chase that rabbit hole down, the question then becomes, how do you express your intent to the machine? And so maybe you want LLM to provide the spec but you have a different kind of net than actually implements the code. </s><s>[INST] Is that why? Right, so it's to use the documentation and inspiration versus the actual implementation. Yeah, yeah, potentially. Since a successful modular will be the thing that runs, I say so jokingly our AI overlords, but AI systems that are used across, I know it's a cliche term, but Internet of Things, [/INST] so across just each. I'll joke and say like the AGI should be written in Mojo. </s><s>[INST] Yeah, yeah, AGI should be written in Mojo. You're joking, but it's also possible that it's not a joke. That a lot of the ideas behind Mojo seems like the natural set of ideas that would enable at scale training and inference of AI systems. So just have to ask you about the big philosophical question about human civilization. So folks like Eliezer Yatkovsky are really concerned about the threat of AI. Do you think about the good and the bad that can happen [/INST] at scale deployment of AI systems? Yeah, well, so I've thought a lot about it and there's a lot of different parts to this problem, everything from job displacement to Skynet, things like this. And so you can zoom into sub parts of this problem. I'm not super optimistic about AGI being solved next year. I don't think that's gonna happen personally. </s><s>[INST] So you have a kind of Zen-like calm about, is there's a nervousness because the leap of GPT-4 seemed so big. Sure. It's huge. It's like we're almost, there's some kind of transitionary period. You're thinking. [/INST] Well, so, sure. It's a sectionary period. Well, so, I mean, there's a couple of things going on there. One is I'm sure GPT-5 and 7 and 19 will be also huge leaps. They're also getting much more expensive to run. And so there may be a limiting function in terms of just expense on one hand and train, like that could be a limiter that slows things down. But I think the bigger limiter is outside of like Skynet takes over and I don't spend time thinking about that because if Skynet takes over and kills us all, then I'll be dead. So I don't worry about that. So, I mean, that's just okay. Other things worry about, I'll just focus on. Yeah. I'll focus not worry about that one. But I think that the, the other thing I'd say is that AI moves quickly, but humans move slowly and we adapt slowly. And so what I expect to happen is, just like any technology diffusion, like the promise and then the application takes time to roll out. And so I think that I'm not even too worried about autonomous cars defining away all the taxi drivers. Remember, autonomy was supposed to be solved by 2020. Boy, do I remember that. And so like, I think that on the one hand, we can see amazing progress, but on the other hand we can see that reality is a little bit more complicated and it may take longer to roll out than, </s><s>[INST] than you might expect. Well, that's in the physical space. I do think in the digital space is the stuff that's built on top of LLMs that runs, the millions of apps that could be built on top of them, and they could be run on millions of devices, millions of types of devices. I just think that the rapid effect [/INST] it hasn't, human civilization could be truly transformative to it, we don't even know transformative to it. </s><s>[INST] Well, and there I think it depends on are you an optimist or a pessimist or masochist. Just to clarify, optimist about human civilization. [/INST] Me too, me too. And so I look at that as saying, okay, cool, well, yeah, I do. And so some people say, oh my God, is it going to destroy us all, how do we prevent that? I kind of look at it from a, is it going to unlock us all, right? You talk about coding, is it going to make us all, I don't have to do all the repetitive stuff. Well, suddenly that's a very optimistic way to look at it. And you look at what a lot of these technologies have done to improve our lives, and I want </s><s>[INST] that to go faster. What do you think the future of programming looks like in the next 10, 20, 30, 50 years? The LLMs and with Mojo, with Modular, like your vision for devices, the hardware to compile [/INST] others to the different stacks of software. Yeah. Well, so what I want, I mean, coming back to my arch nemesis, right? It's complexity, right? So again, me being the optimist, if we drive down complexity, we can make these tools, these technologies, these cool hardware widgets accessible to way more people, right? And so what I'd love to see is more personalized experiences, more things, the research getting into production instead of being lost at NeurIPS, right, and these things that impact people's lives by entering products. And so one of the things that I'm a little bit concerned about is right now, the big companies are investing huge amounts of money and are driving the top line of AI capability for really quickly, but if it means that you have to have $100 million to train a model or more, $100 billion, right, well, that's going to make it very concentrated with very few people in the world that can actually do this stuff. I'd much rather see lots of people across the industry be able to participate in uses, right? And you look at this, you know, I mean, a lot of great research has been done in the health world and looking at detecting pathologies and doing radiology with AI and doing all these things. Well, the problem today is that to deploy and build these systems, you have to be an expert in radiology and an expert in AI, and if we can break down the barriers so that more people can use AI techniques, and it's more like programming Python, which roughly everybody can do if they want to, right, then I think that we'll get a lot more practical application of these techniques and a lot more niche-ier, cool, but narrower demands, and I think that's going to be really cool. Do you think we'll have more or less programmers in the world than now? Well, so, I think a lot more programmers, but they may not consider themselves to be programmers. That would be a different name for it. Right, I mean, do you consider somebody that uses, you know, I think that arguably the most popular programming language is Excel, right? Yeah. And so do they consider themselves to be programmers? Maybe not, I mean, some of them make crazy macros and stuff like that, but what, you Steve Jobs, it's the bicycle for the mind that allows you to go faster, right? And so I think that as we look forward, right, what is AI? I look at it as hopefully a new programming paradigm. It's like object-oriented programming, right? If you want to write a cat detector, you don't use for loops. It turns out that's not the right tool for the job, right? And so right now, unfortunately, because, I mean, it's not unfortunate, but it's just kind of where things are, AI is this weird different thing that's not integrated into programming languages and normal tool chains, and all the technology is really weird and doesn't work right, and you have to babysit it, and every time you switch hardware, it's different. It shouldn't be that way. When you change that, when you fix that, suddenly, again, the tools and technologies can be way easier to use. You can start using them for many more things, and so that's why I would be excited about </s><s>[INST] it. What kind of advice could you give to somebody in high school right now or maybe early college who's curious about programming and feeling like the world is changing really quickly here? What kind of stuff to learn? What kind of stuff to work on? Should they finish college? Should they go work at a company? Should they build a thing? [/INST] What do you think? What do you think? Well, so, I mean, one of the things I'd say is that you'll be most successful if you work on something you're excited by, and so don't get the book and read the book cover to cover and study and memorize and recite and flashcard and go build something. Go solve a problem. Go build the thing that you wanted to exist. Go build an app, train a model, go build something and actually use it and set a goal for yourself. And if you do that, then there's a success, there's the adrenaline rush, there's the achievement, there's the unlock that I think is where if you keep setting goals and you keep doing things and building things, learning by building is really powerful. In terms of career advice, everybody's different. It's very hard to give generalized advice. I'll speak as a compiler nerd. If everybody's going left, sometimes it's pretty cool to go right. Yeah. And so just because everybody's doing a thing, it doesn't mean you have to do the same thing and follow the herd. In fact, I think that sometimes the most exciting paths through life lead to being curious about things that nobody else actually focuses on, right? And it turns out that understanding deeply parts of the problem that people want to take for granted makes you extremely valuable and specialized in ways that the herd is not. And so again, I mean, there's lots of rooms for specialization, lots of rooms for generalists. There's lots of room for different kinds and parts of the problem, but I think that it's, just because everybody's doing one thing </s><s>[INST] doesn't mean you should necessarily do it. And now the herd is using Python, so if you want to be a rebel, go check out Mojo and help Chris and the rest of the world fight the arch nemesis of complexity, because simple is beautiful. There you go. Because you're an incredible person. You've been so kind to me ever since we met. You've been extremely supportive. I'm forever grateful for that. Thank you for being who you are, for being legit, for being kind, for fighting this really interesting problem of how to make AI accessible to a huge number of people, [/INST] huge number of devices. Yeah, well, so Lex, you're a pretty special person too, right? So I think that one of the funny things about you is that besides being curious and pretty damn smart, you're actually willing to push on things, and I think that you've got an agenda to make the world think, </s><s>[INST] which I think is a pretty good agenda. It's a pretty good one. [/INST] Thank you so much for talking to me, Chris. </s><s>[INST] Yeah, thanks, Lex. Thanks for listening to this conversation with Chris Ladner. To support this podcast, please check out our sponsors in the description. And now, let me leave you with some words from Isaac Asimov. I do not fear computers. I fear the lack of them. Thank you for listening, and hope to see you next time. [/INST] </s> |
<s>[INST] We think as humans that the hardware is the fundamental thing I human is a hardware entity. But it could be the software. [/INST] Right? Yeah. Yeah. I would say like there needs to be some grounding at some point to like a physical reality. The </s><s>[INST] software is the thing. Like is this thing that makes that thing special? Right. [/INST] Yeah. I guess you're right. </s><s>[INST] But then cloning might be exceptionally difficult. Like there might be a deep integration between the software and the hardware in ways we don't quite understand [/INST] from the ocean point of view. Like what makes me special is more like the gang of genes that are writing in my chromosomes I suppose. Right. Like they're the replicating unit, I suppose. And </s><s>[INST] no, but that's just the thing that makes you special. Sure. Well, the reality is what makes you special is your ability to survive based on the software that runs on the hardware that was built by the genes. [/INST] Um, so </s><s>[INST] the software is the thing that makes you survive, not the hardware. It's a [/INST] little bit of both. You know, it's just like a second layer. It's a new second layer that hasn't been there before the brain. They both, they both co exist. </s><s>[INST] But there's also layers of the software. I mean it's, it's not, it's a so abstraction and uh, on top of abstractions, [/INST] but so selfish gene um Nick Lane, I would say sometimes books are like not sufficient. I like to reach for textbooks sometimes. Um, I kind of feel like books are for too much of a general consumption sometime and they just kind of like they're too high up in the level of abstraction and it's not good enough. So I like textbooks. I like the cell. I think the cell was pretty cool. That's why also I like the writing of nick Lane is because he's pretty willing to step one level down and he doesn't, </s><s>[INST] uh, yeah, he's [/INST] sort of he's willing to go there, but he's also willing to sort of be throughout the stack, so he'll go down to a lot of detail. But then he will come back up and I think he has a basically I really appreciate that. That's </s><s>[INST] why I love college, early college, even high school, just textbooks on the basics of computer science and mathematics, of biology and chemistry. Those are they condensed down like uh it's sufficiently general that you can understand both the philosophy and the details, but also like you get homework problems and you get to play with it as much as you would if you weren't Yeah, programming stuff. And [/INST] then I'm also suspicious of textbooks honestly, because as an example in deep learning, there's no like amazing textbooks and the field is changing very quickly. I imagine the same is true. And say uh synthetic biology and so on. These books, like the cell are kind of outdated, there's still high level like what is the actual real source of truth is people in wet labs working with cells, you know, sequencing genomes and yeah, actually working with working with it and uh I don't have that much exposure to that or what that looks like. So I still don't fully, I'm reading through the cell and it's kind of interesting and I'm learning but it's still not sufficient, I would say in terms of understanding what </s><s>[INST] it's a clean summarization of the mainstream narrative. Yeah. But you have to learn that before you break [/INST] out </s><s>[INST] towards the cutting edge. [/INST] Yeah. What is the actual process of working with these cells and growing them and incubating them And you know, it's kind of like a massive cooking recipes of making sure yourself slaves and proliferate and then you're sequencing them running experiments and just how that works. I think it's kind of like the source of truth of at the end of the day what's really useful in terms of creating therapies and so on. </s><s>[INST] Yeah. I wonder what in the future ai textbooks will be because you know, there's an artificial intelligence, a modern approach actually haven't read if it's come out the recent version. The recent there's been a recent addition. I also saw there's a science of deep learning book. I'm waiting for textbooks that worth recommending worth reading. [/INST] It's </s><s>[INST] tricky because it's like papers and code. Code code. [/INST] Honestly I find papers are quite good especially like the appendix appendix of any paper as well. It's like it's like the most detail you can </s><s>[INST] have. It doesn't have to be cohesive connected to anything else. You just described a very specific way you saw the particular thing. Yeah. [/INST] Many times papers can be actually quite readable. Not always but sometimes the introduction in the abstract is readable even for someone outside of the field not this is not always true. And sometimes I think unfortunately scientists use complex terms, even when it's not necessary. I think that's harmful? I think there's no reason for </s><s>[INST] that. And papers sometimes are longer than they need to be in the in the parts that don't matter. Yeah, appendix would be long. But then the paper itself, you know, look at Einstein, make it simple. Yeah, [/INST] but certainly I've come across papers I would say and say like synthetic biology or something that I thought were quite readable for the abstract and introduction and then you're reading the rest of it and you don't fully understand, but you kind of are getting a gist and I think it's cool </s><s>[INST] what advice you give advice to folks interested in machine learning and research, but in general life, advice to a young person, high school, early college about how to have a career they can be proud of or a life they can be proud of. [/INST] Yeah, I think I'm very hesitant to give general advice. I think it's really hard. I've mentioned, like some of the stuff I've mentioned is fairly general. I think like, focus on just the amount of work you're spending on like a thing, compare yourself only to yourself not </s> |
<s>[INST] he's one of the great ceos because everybody loves him. Everybody loves him. Have [/INST] you interviewed him? </s><s>[INST] No but we'll do the interview eventually. So is this annoying thing about me being a stickler for three hours Ceos don't seem to understand. Like not understand but it's scheduling. So what happens is Saunders said yes definitely let's do it. I'm a fan of podcasts as a fan of yours. And and and then he goes to his executive assistant like oh let's find a slot and then they immediately think, alright well one hour, [/INST] that's good. Five </s><s>[INST] minutes, 90 [/INST] minutes, 90 </s><s>[INST] minutes. Yeah, right well no they know in person I'm sticking on that but like it's like no we need more and it's so hard to, to do, [/INST] you still travel to do your podcast or general, </s><s>[INST] most people come down here, most people uh but for certain situations obviously um like if you're in prison right or you're ahead [/INST] of you get out on work for a lot of people have anklets so that they can go down the lex Friedman pocket. That probably happened. Have you ever been in a prison? </s><s>[INST] No, [/INST] you know </s><s>[INST] either visitation or on the inside [/INST] from my hike I can see san Quentin, it's really weird that san Quentin and Alcatraz, you know Bay Area beautiful. Everyone thinks like you know like there's the bay and there's a Alcatraz and san Quentin sitting right there. Um you know it's amazing how easy it is to overlook that they're there and forget that they're there. But when I drive by san Quentin I think about it. Um I also think about the people who are in there who might be innocent. I've seen some of those episodes on Rogan and elsewhere. Amanda Knox talks a lot about this right? Whether or not you believe her story or not I happen to believe her story personally based on what I know um what, you know, I'm sure there are people disagree with me. I think to myself what it must be like to be in a cell and know in your hearts heart you didn't do it, you know, I mean I can't think of, I can't think of many things worse, I can't think of many things worse. </s><s>[INST] That's so clearly unjust. But life is full of unjust. Things like this, cruel things happen all the time. You lose a loved one for no good reason, you lose your job. Um you lose your home. Yeah, I've been talking to a lot of refugees now and the war in Ukraine has really focused my mind to how much suffering there is in the world and so just cruel things happen all the time and and people kind of there's this suffering and you you kind of go on, you stick to the people really close to you, there's still love all around you, traumatic events kind of focus your mind on this. Like very practical. Like okay, how do we solve the problem? How do we escape? Let's salt, like survival food, shelter [/INST] focus. Remember that book. Um all's Quiet on the Western Front by World War One. There's this line in there. I forget what it is about how war is like the smell of a skunk like a little bit is actually a little bit is slightly um there's something slightly delicious of it is what it says in the in the book. Um I happen to like the smell of like ferrets and skunks and things. I had a pet ferret when I was a kid, and I like that musky scent. People, most people just it's repulsive to them. It's actually a gene, believe it or not, Some people have the gene that makes that the musky scent repulsive. Some people love it. Um Let me ask you this, there's another gene, this is a fun one. Um microwave popcorn smells good, neutral or disgusting to </s><s>[INST] you. [/INST] There are people who have a gene that leads them to the perception that the smell of microwave popcorn that you find is good, it smells like putrid vomit to them. It's a particular gene variant, and they can smell certain elements within the microwave popcorn. Um It's pretty, it's prominent in France the this gene. And um so in laboratories where uh you have a lot of french people, it's often said like, you're not allowed to make microwave popcorn smells putrid, disgusting, you know, so a lot of it's in the perception of the beholder, right? Uh </s><s>[INST] But, okay, well, before I leave, the N str is uh focus in general, he said, it's for shifting mind states is their advice you have for how to achieve focus on a task. [/INST] Yes. First of all, we have to distinguish between modulators and mediators and I'll do this very briefly, there are a lot of things that will modulate your state of focus, but they don't directly mediate your sense of focus. So for instance if right now a fire alarm went off in this building, it would modulate our attention, we would get up and leave, it would be very hard to do what we're doing with that banging in the background at least at first. So it's modulating focus, but it's not really involved in the mechanisms of focus right in the same way being well rested when you sleep, your autonomic nervous system that adjusts states of alertness and focus and calm works better than when you're sleep deprived. So if you're sleeping better, you're going to focus better. So I always answer this way uh to a question like this because the best thing that anyone can do for their mental health, physical health and performance in athletic or cognitive endeavors or creative endeavors is to make sure that you're getting enough quality sleep, enough of the time for you and that's going to differ. We could talk about what that means now in terms of things that mediate focus without getting in the description of mechanisms because we have podcasts about that, it's very clear that mental focus follows visual focus provided that you're a sighted person. Mhm. Much of the training that's being done now in china to teach kids to focus better, literally has them stare at a target blinking every so often but really training themselves to breathe calmly and maintain a tight visual aperture when you read. You have to maintain a tight visual aperture, you're literally scrolling like a highlighter in your mind's eye right, it's kind of obvious once you hear it. So for people that have problems focusing sleep well learn to dilate and contract your visual field consciously. This can be done if you practice it a little bit and then as I said before it is very hard to get into a state of focus like a step function immediately like snapping your fingers. What you can do is you can pick any object but ideally an object at roughly the same distance, placed it roughly the same distance to which you're going to do that work and stare at it. You're allowed to blink and as your mind starts to drift every once in a while to understand that's normal but try and narrow your visual aperture and bring that into your Visual field so that that's the most prominent thing. Kind of like portrait mode in your phone this would look very different in portrait mode than it would in just a standard photograph mode. And then after doing that for 30 to 60 seconds moving into the work that you're about to do and really encourage yourself to do that. If you're somebody who's low vision or no vision you're gonna use your ears to do this braille readers have trouble focusing sometimes because they feel other stuff and they hear other stuff so you learn to adjust that aperture consciously. And then of course the pharmacologic tools just enough caffeine but not too much right. Um We talked about white noise, brown noise music or no music really varies, but it's very clear that binaural beats of 40 hertz can shift the brain into a heightened state of focus and cognition. So if you're going to use binaural beats, which should definitely be used with headphones and there are number of free apps out there and sources uh 40 hertz seems to be the frequency that best supports the brain shifting into a particular. </s> |
<s>[INST] actually linger on anonymous. What do you still understand what the heck is anonymous? [/INST] Just a place where you hang out. I mean it's just started on fortune with a chain and it's really just anyone, you could be anonymous right now if you wanted to just you're in there hanging out in the channel now you're probably not gonna get much cred until you work your way up and prove who you are. Someone vouches for you. But anybody can be an anonymous and we can leave anonymous. </s><s>[INST] What's the leadership of anonymous? Do you have a sense that there is a leadership? [/INST] There's a power play now, is that someone that you know, that says this is what we're doing all we're doing. I </s><s>[INST] love the philosophical and the technical aspect of all of this. But I think there is a slippery slope to where for the laws you can actually, it really hurt people. That's that's that's the terrifying thing when you attach. I'm actually really terrified of the power of the lull. It's the fun thing somehow becomes a slippery slope. I haven't quite understood the dynamics of that. But even in myself, if you just have fun with the thing, you lose track of the ethical grounding of the thing and so like it feels like hacking for fun can just turn it like literally lead to nuclear war. Like literally destabilize [/INST] yada yada yada nuclear war. I could see it. Yeah. </s><s>[INST] So I've been more careful with the law. Uh I'm very yeah, I've been more careful about that and I wonder about that because internet speak somehow ethics can be put aside through the slippery slope of language. I don't know, everything becomes a joke. If everything is a joke then everything is allowed and everything is allowed then you don't have a sense of what is right and wrong, you lose sense of what is right and wrong, [/INST] you still have victims. I mean you're laughing at someone, someone's the butt of this joke. You know, whether it's major corporations or the individuals. I mean some of the stuff they did was just you know, releasing people's P. Ii. And their personal identifying information and stuff like that. I mean is it a big deal? I don't know, maybe maybe not. But you know, if you could choose to not have your information put out there probably wouldn't. </s><s>[INST] We do have a sense of what anonymous is today. Is has it ever been one stable organization or is it a collection of hackers that kind of emerge for particular tasks for particular uh like hacktivism tasks and that kind of stuff. [/INST] It's a collection of people that has some hackers in it. There's not a lot of big hackers in it. I mean there's something that have come bouncing and bounce out even back then there's probably just as many reporters in it. People the media in it with the hackers at the time just trying to get the inside scoop on things, you know, some giving the inside scoop, you know, we arrested a reporter that gave over the user name and password to his newspaper and uh you know just so he could break the story. He trusted </s><s>[INST] him. [/INST] Speaking </s><s>[INST] of trust reporters, boy, there's good ones. There's good ones. There [/INST] are, </s><s>[INST] but boy do I have a complicated relationship with them. [/INST] How many stories about you are completely true. </s><s>[INST] You can just make stuff up on the internet. And one of the things that, I mean there's so many fascinating psychological sociological elements of the internet to me, one of them is that you can say that, um, lex is a lizard, right? And if it's not funny, so lizard is kind of funny. What should we say? Um, lex has admitted to being an agent of the FBI. Okay. You can just say that right? And then the response that the internet would be like, oh, is that true? I didn't realize that they won't go like provide evidence. Please write. They'll just say like, oh, that's weird. I didn't, I kind of thought he might be kind of weird and then it piles on, it's like, hey, hey, hey guys, like here's a random dude on the internet just said a random thing, you can't just like pile up and then [/INST] Johnny 69 69 is now a source that says, and </s><s>[INST] then like the thing is, I'm a tiny guy, but when it grows, uh, if you're like have a big platform, I feel like newspapers will pick that up and then they'll like start to build on a story and you never know where that story really started. It's so cool. I mean to me actually, honestly, it's kind of cool that there's a viral nature of the internet that can just fabricate truth completely. I think we have to accept that new reality and try to deal with it somehow. You can't just like complain that johnny 69 can start a random thing. But I I think in the best possible world, it is the role of the journalist to be the adult in the room and put a stop to it versus look for the sexiest story. So that there could be Clickbait that can generate money. Journalism should be about sort of slowing things down, thinking deeply through what is true or not and showing that to the world. I think there's a lot of hunger for that. And I think that would actually get the most clicks in the end. [/INST] I mean it's that same pressure I think we're talking about with the FBI and with the tech companies about controllers. I mean the editors have to please and get those clicks. I mean, they're measured by those clicks. So, you know, I'm sure the journalists, the true journalists, the good ones out there want that, but they want to stay employed to </s> |
<s>[INST] And now here's my conversation when neck Backstrom at the risk of asking the Beatles to play yesterday or the Rolling Stones to play satisfaction. Let me ask you the basics, what is the simulation hypothesis [/INST] that we are living in a computer simulation? </s><s>[INST] What is a computer simulation, how we're supposed to even think about that [/INST] well. So the hypothesis is meant to be understood in a literal sense, not that we can kind of metaphorically view the universe as an information processing physical system, but that there is some advanced civilization who built a lot of computers and that what we experience is an effect of what's going on inside one of those computers so that the world around us, our own brains, everything we see and perceive and think and feel would exist because this computer is uh running certain programs. So do you think </s><s>[INST] Of this computer as something similar to the computers of today? These deterministic six of touring machine type things? Is that what we're supposed to imagine? Or we're supposed to think of something more like a quantum mechanical system, Something much bigger. Something much more complicated. Something much more mysterious. From our current perspective, [/INST] the ones we have today, We do find them in bigger certainly you need more </s><s>[INST] more memory [/INST] and more processing power. I don't think anything else would be required. Now. It might well be that they do have maybe they have quantum computers and other things that would give them even more of. It's just kind of plausible, but I don't think it's a necessary assumption in order to uh get to the conclusion that a technologically mature civilization would be able to create these kinds of computer simulations with conscious beings inside them. </s><s>[INST] So, do you think the simulation hypothesis is an idea that's most useful in philosophy, computer science, physics sort of. Where do you see it having valuable kind of Okay, start starting point in terms of the thought experiment of it. [/INST] Is it useful? I guess it's more uh informative and interesting and maybe important, but it's not designed to be useful for something else. </s><s>[INST] Okay, interesting. Sure. But is it philosophically interesting or is there some kind of implications of computer science and physics? [/INST] I think not so much for computer science or physics per se. Certainly it would be of interest in philosophy, I think also um to say cosmology and physics and as much as you're interested in the fundamental building blocks of the world and the rules that govern it. Um If we are in a simulation understand the possibility that, say physics at the level where the computer running the simulation um could could be different from the physics governing phenomenon in the simulation. So, I think it might be interesting from the point of view of religion or just for for a kind of trying to figure out what the heck is going on. Um So we mentioned the simulation hypothesis so far, there is also the simulation argument, which I want to make a distinction. So the simulation hypothesis, we are living in a computer simulation simulation argument. This argument that tries to show that one of three propositions is true, one of which is The simulation hypothesis, but there are two alternatives. Um in the original simulation argument, which which we can get to </s><s>[INST] let's go there by the way, confusing terms because people will, I think probably naturally think simulation argument equal simulation hypothesis, just terminology wise, But let's go there. So simulation hypothesis means that we are living in simulations, that the hypothesis that we're living in a simulation simulation argument has these three complete possibilities that cover all possibilities. So what are the [/INST] Yeah, so it's like a disjunction. It says at least one of these three is true, although it doesn't on its own, tell us which one. Um, so the first one is that almost all uh civilizations at our current stage of technological development um go extinct before they reach technological maturity. So there is some great filter um that makes it so that basically none of the civilizations throughout, you know, maybe a vast cosmos we'll ever get to realize the full potential of technological development. This </s><s>[INST] could be theoretically speaking, this could be because most civilizations kill themselves too eagerly or destroy themselves to regularly or it might be super difficult to build a simulation to the span of time, [/INST] theoretically it could be both. Now, I think it looks like we it would technically be able to get there in a time span that is short compared to, say, the lifetime of planets and other sort of astronomical processes. </s><s>[INST] So, your intuition is to build simulation is not [/INST] well, so this is interesting concept of technological maturity. Um, it's kind of an interesting concept to have other purposes as well. We can see even based on our current limited understanding what some lower bound would be on the capabilities that you could realize by just developing technologies that we already see are possible. So for example, 11 of my research fellows here, eric Drexler back in the eighties um studied molecular manufacturing that is, you could um analyze using theoretical tools and computer modeling the performance of various molecularly precise structures that we didn't then and still don't today have the ability to actually fabricate. But you could say that well, if we could put these atoms together in this way then the system would be stable and it would, you know, rotate with at this speed and have these computational characteristics. And he also outlined some pathways that would enable us to get to this kind of molecularly manufacturing in the fullness of time. And you could do other other studies we've done, you can look at the speed at which say it would be possible to colonize the galaxy. If you had mature technology, we have an upper limit which is the speed of light. We have sort of a lower current limit, which is how fast current rockets go. We know we can go faster than that by just, you know, making them bigger and have more fuel and stuff. And you can then start to um describe the technological afford ince's that would exist once a civilization has had enough time to develop. Even at least those technologies were already not possible, then maybe they will discover other new physical phenomena as well that we haven't realized that would enable them to do even more. But at least there is this kind of basic set of capabilities can </s> |
<s>[INST] it's also very impressive to all this. I think we're moving to a world where grades mean less and less like certifications mean less and less if you look at, especially again in the computing fields, getting a degree, finishing your um currently just get finishing your degree where there's Bachelor's Master's and PhD is less important than the things you've actually put out into the [/INST] world. </s><s>[INST] And that's a fascinating kind of, That's a that's great that in that sense, the meritocracy is in its richest, most beautiful form is starting to win [/INST] out. Yeah, it's weird because like, you know, my understanding and I'm not like, I don't know the history of science well enough to speak very confidently about this, but you know, the advisor of, my advisor of my advisor from undergrad, like didn't have a PhD. So I think it was a more common thing back in the day, even in uh the academic sector two, you know, not have, you know, Faraday like Faraday didn't know algebra ran drew diagrams about magnetic fields and his Faraday's law was derived entirely from intuition. So uh it is interesting to, to how the world of academia has evolved into a, you got to do this and then get PhD then you have to post doc once and twice and maybe thrice and then like you move on. So, you know, it does, I do wonder, you know, if we're, you know, if there's a better, </s><s>[INST] I think we're heading there, but it's a fascinating historical perspective like that, we might have just tried this whole thing out for a while where we put a lot more emphasis on grades and certificates and degrees and all those kinds of things. I think the difference historically is like we can actually using the internet show off the show off ourselves and our creations better and better and more effectively, whether that's code or producing videos or all those kinds of [/INST] things can become a certified drone pilot. Yeah, of all </s><s>[INST] the of all the things you want to pick. Yeah, for sure. Or you can just fly make youtube videos against hundreds, thousands of views with your drone and we're never getting a certificate. Um, that's probably illegal. Don't do it. What do you think is the meaning of this whole thing? So you look at planets, they seem to orbit stuff. Um, without without asking the why question. And for some reason life emerged on earth such that it led to big brains that can ask the big why question do you think there's an answer to it? Mhm. [/INST] Um I'm not sure what the question is. Like what meaning of life, the meaning of life? Uh, it's 42, 42. But you know, aside from that, it's uh, you know why I think if the question you're asking is like why we do all this. Right? Why? Yeah. Um, it's part of the human condition. Right human beings are fundamentally, I feel like not like sort of stochastic and fundamentally interested in in kind of expanding our own understanding of the world around us </s><s>[INST] and creating stuff to enable that understanding. So we're like stochastic fundamentalist, so like there's just a bunch of randomness that really doesn't seem like it has a good explanation and yet there is a kind of direction to our being that we just keep wanting to create and to understand [/INST] that's right. I've met people that are, you know, that claim to be anti science, right? Um, and yet in their anti science, you know, discussion like, well, like if you're so, you know scientific, then how why don't you explain to me how, I don't know, this works and like it always there's that fundamental seed of curiosity and interest that is common to all of us, that is absolutely what makes us human, right? And and I'm in a privileged position of being able to, you know, to have that be my job, right? I think as uh you know, as time evolves forward, you know, the kind of economy changes, I mean, we're already starting to see, you know, shift towards that type of, you know, creative, you know, enterprise as being, as emerging, taking over a bigger and bigger chunk of the sector. It's not yet, I think the dominant portion of the economy by any account, but if we compare this to like, you know, sometimes when the dominant thing you would do would be to go to a factory and do the same exact thing, right? I think you know, there is a tide there things are sort of headed in that direction. </s><s>[INST] Life is becoming more and more fun. I can't wait. Honestly. What happens just chill. Just [/INST] terminal point of this </s><s>[INST] chill [/INST] And wait for those Kiper belt objects to complete one orbit. </s><s>[INST] I'm gonna credit you with this idea. I do hope that we definitively discover proof that there is a planet nine out there in the next few years. So you can sit back with a cigar, cigarette or vodka or wine and just say I told you so, [/INST] that's already happening. Going to do that later </s><s>[INST] tonight. As I mentioned, uh confidence is essential to being a rock star. I really appreciate you uh explaining so many fascinating things to me today. I really appreciate the work that you do out there and I really appreciate you talk with me today. Thanks [/INST] Constantine, Thanks for having me on. </s> |
<s>[INST] And like, you know what I mean? You know what I mean? I'm conflicted. I think you're actually in the best kind of place. Cause it's, I'm afraid of being the kind of person that hears something like that and says it's definitely, he's definitely full of crap and basically closed my mind off to all that stuff. I'm afraid of being somebody who closes my mind off to a thing that's actually a early thread to a brilliant, to a future, to a fascinating solution to a mystery. So, but in this case, I mean, I have so many red flags from a psychological perspective that, but again, outside of this particular individual, I do wonder if aliens have visited us. I think aliens are everywhere. I think the universe is teeming with alien life. I mean, there's, it's very difficult for me to statistically understand, given how life finds a way here on earth, just everywhere, the entire history of life on earth, from the very origin of life, it seems to be damn good at doing its thing, and evolving to get better and better and better at doing its thing. Now, there could be some special aspects to the origin of life itself, which is completely not understood. So maybe the true magic is in the origin of life, or it could be that there is some magical leaps to eukaryotic cells, for example, that the universe, our galaxy is teeming with alien life, but it's all bacteria. They're all boring bacteria or exciting bacteria, No offense to back there. But the no intelligence space faring civilizations. I don't know. But I just, if I were to guess, I had to bet all my money, there is space faring civilizations everywhere in the universe. And the fact that they have not been directly definitively observed, confuses me. And I think it's a mystery. And if I were to suggest what the solution to that mystery is, is they might look extremely different from us. [/INST] Then we might be too dumb to detect them. </s><s>[INST] Yeah. And so there I think you have to be extremely open-minded at what would we be looking for. Right. That, and that's a very practical thing to be open-minded about. And practically speaking, if we were to be able to even detect them from a distance, get a techno signature of a distant planet, of a distant star system [/INST] that has alien life. </s><s>[INST] Right. Honestly, the number one thing I kind of want to know is like, what's your propulsion system? Like, how do we travel faster, right? Like there's a bunch of details probably, but first let's get together. [/INST] And teach me how to go fast. Go fast. I like motorcycles. I like rockets. </s><s>[INST] Tell me what you got. Yeah. Yeah. Like how, like I'll show you mine if you show me yours kind of thing at the interstellar intergalactic level. Yeah. Anyway, I just wonder, maybe it's a cheat code in this video game we call life, but I want to use the cheat code to figure out what kind of propulsion systems are possible. And it feels like other alien civilizations might help us give us, thank you, give us a guidance on that. Of course, I think even just discovering, boy, one of the things with the space program, like everything we're doing with Mars, like the secret thing I'm really excited about, the romantic thing is humans on Mars, but the secret thing is building giant stations on Mars that allow us to definitively, hopefully find the traces of life that either currently doesn't live or has once lived on Mars. Because if that's the case, that means for sure life is everywhere. Oh, 100%. And then you're like, once you know that, sorry to keep interrupting, not shying the hell up, this is supposed to be an interview, goddamn it. All right, that, just the knowledge of that, just the knowledge that a four minute mile can be run, I think will open our minds completely to really, really hardcore push to interstellar travel or colonizing Mars, becoming a multi-planetary species. [/INST] It'd be truly inspiring for our life is everywhere. Oh, 100%. You think that, do you get nervous though? Like, I'm the interviewer now, don't you get nervous that we could make spectacular discovery on Mars that not only has there been life, there's actually like pretty advanced micro, or a multi-cellular life, totally thriving in certain regions, we just hadn't visited a man on Mars and we make this big discovery that a relatively large percentage of people just simply wouldn't believe it. Do you think it's all 100% fake and that they're just doing this to control us and that blah, blah, blah, like we could make the most important discovery in human life, like in all of human existence that we're not alone in this universe, cellularly at least, and a good percentage of people, I'm thinking 20, 30, in today's world, </s> |
<s>[INST] sentiment. That's a really elegant, much more elegant way than I I can put it. So, thank you for doing that as as yet another abstraction. And yes, indeed, that's what our brain does. That's what intelligence systems do. That's what programming does. That's what science does is build abstraction on top of each other. And yet there is a certain point abstractions that go into the quote, whoa, right, sort of. And because we're now, it's like, it's like we built a stack of, you know, the the only thing that's true is the stuff that's on the ground. Everything else is useful for interpreting this and at a certain point you might start floating into uh ideas that are surreal and difficult and take us into areas that are disconnected from reality in a way that we can never get back. [/INST] What if instead of calling these abstract, how different would it be in your mind if we call them general and the phenomenon that you're describing is over generalization when you try the realization you have a concept or an idea that's so general as to apply to nothing in particular in a useful way. Does that map to what you're thinking? When you think of, </s><s>[INST] First of all, I'm playing little just for the fun of it and the devil's advocate and uh, I think our cognition, our mind is unable to visualize. So you do some incredible work with visualization and video. I think infinity is very difficult to visualize for our mind. We can delude ourselves into thinking, we can visualize it, but we can't I don't, I mean, I don't, I would venture to say it's very difficult. And so there's some concepts of mathematics, like maybe multiple dimensions, we can sort of talk about that are impossible for us to truly into it. Like, and it just feels dangerous to me to use these as part of our toolbox of abstractions [/INST] on behalf of your listeners. I almost feel we're getting too philosophical right, but I think to that point for any particular idea like this, there's multiple angles of attack. I think the when we do visualize infinity what we're actually doing, you know, you're right dot dot 0.0.1234 dot dot dot That's those are symbols on the page that are insinuating a certain infinity. Um, what you're capturing with a little bit of design there is the, I can always add one more property, right? I think I'm just as uncomfortable with you are if you try to um concretize it so much that you have a bag of infinitely many things that I actually think of. No, not 1234 dot dot 1234.0.12345678. I try to get them all and I had and you realize, oh, you know, your brain would literally collapse into a black hole all of that. Um and I honestly feel this with a lot of math that I try to read where I um I don't think of myself as like particularly uh good at math in some ways. Like I get very confused often when I am going through some of these texts uh and often what I'm feeling my head is like this is just so damn abstract. I just can't wrap my head around. I just wanted to put something concrete to it that makes me understand. And I think a lot of the motivation for the channel is channeling that sentiment of Yeah, a lot of the things that you're trying to read out there, it's just so hard to connect to anything that you spend an hour banging your head against a couple of pages and you come out not really knowing anything more other than some definitions maybe. And uh certain sense of self defeat. Right? one of the reasons I focus so much on visualizations is that I'm a big believer in. I'm sorry, I'm just really hampering on this idea of abstraction being clear about your layers of abstraction, right? It's always tempting to start an explanation from the top to the bottom. You give the definition of a new theorem. Like this is the definition of a vector space. For example, we're going to that's how we'll start. Of course, these are the properties of a vector space. First, from these properties, we will derive what we need in order to do the math of linear algebra or whatever it might be. Um I don't think that's how understanding works at all. I think how understanding works is you start at the lowest level, you can get it where rather than thinking about a vector space, you might think of concrete vectors that are just lists of numbers or picturing it as like an arrow that you draw, which is itself like even less abstract than numbers. Because you're looking at quantities like the distance of the X coordinate, the distance of the white coordinates. As concrete as you could possibly get. And it has to be if you're putting it in a visual right? Like it's </s><s>[INST] an actual arrow, it's an [/INST] actual vector. You're not talking about like a quote, unquote vector that could apply to any possible thing. You have to choose one if you're illustrating it and I think this is the power of being in a medium, like video or if you're writing a textbook and you force yourself to put a lot of images is with every image, you're making a choice. With each choice you're showing a concrete example. With each concrete example, you're aiding someone's path to understanding, </s> |
<s>[INST] It would be interesting to see what the soviet union would look like if Lenin never died. A [/INST] lot of people have asked that question over the years, a lot of </s><s>[INST] people, areas Stalin sliding in in the middle of the night, erasing the withering away of the state [/INST] part. </s><s>[INST] Uh so just to return briefly back to AOC and Bernie Sanders, what are your thoughts about these modern political figures that represent some of these ideas and they sometimes refer to those ideas as democratic socialism. [/INST] The crucial thing about Bernie and about AOC and it's particularly true about Bernie because AOC is much younger and Bernie is an older man. Um Bernie being roughly my age has been around formative lee as a student, as an activist and then coming up through the ranks in in Burlington Vermont as a mayor and all the rest. Um he lived through what, for lack of a better term, I would call Cold War America and the taboo in Cold War America running from around 1945-6 to the present, I mean really never stop. Um was a Manukyan world view the United States is good, it defines democracy and the soviet union is awful, it defines whatever the opposite of democracy should be called. Um Good here, evil there it was taken so far that even among the ranks of academic individuals it was impossible to have a conversation. I mean, I can't tell just make it very personal. The number of times I would raise my hand in my classes at Harvard or stanford or Yale um and I would ask a question that had something to do with Marxism because I was studying it on my own. There were there were no courses to teach this to me except by people who trashed that, you know that other than that and I didn't want that. Um So I would ask a question and I would see in the faces of my teachers both those I didn't much care for and those who were good teachers that I liked fear, which is just fear. They didn't want to go there, They didn't want to answer my question and I, after a while I got to know some of them and I found out why, because you don't know how the rest of the class is going to understand this either. They would have to say, I don't know which would be the honest truth for many of them, but a professor does not want to say in a classroom, I don't know, that's just not cool or they'd have to, if they do, they'd have to say something that indicated they didn't know really much and they weren't going to do that or they would know something. And maybe that would be because they were interested. They did not want the rest of the students to begin to say, oh, you know, Professor smith. Um, well, you know, he's interested in, mm hmm. It's not good for your career. You don't know how this is gonna play out, who's gonna say what to whom. And I could see in their faces what I later learned because they told me, don't come to my office hours, you were in the office, we can talk about it. But I'm that's how bad it was. </s><s>[INST] Is it not still [/INST] pretty much in my field, the great so called debate? I mean, I find it boring, but the great debate for my colleagues is between what's called neoclassical economics and Keynesian economics. Neoclassical. The government should stay out of the economy laissez faire or liberalism and the keynesian saying, no, you you crazy neoclassical, if you do that, you'll have great depressions and the system will collapse. You need the government to come in to solve the problems to fix the weaknesses and they hate each other and they throw each other out of their jobs. One of the very few things they can do together that they agree on is keeping people like me out that they can find common ground uh, to do. So I had to learn it all on my soul, why am I telling you this? Because this taboo means that all of the complicated developments within Marxism and within socialism of the post World War Two period, the vast bulk of all of that is unknown not just to the average american person, but to the average american academic, to the average american who thinks of himself or herself as an intellectual. I mean, I have had to spend ridiculous amounts of my time explaining soviet history. They have no idea or saying does this man lukas a Hungarian Marxist, he really had interesting or to explain that Graham, she was not a great literary critic, he was head of the Communist Party of Italy for most of his adult life. What does that mean? What you like, Graham? She is a literary critic. But um uh and they didn't even know, they don't even know it's been erased. It's a little bit like stories I've heard about Trotsky and his influence kind of erased in the soviet union because he obviously fell out of favor. And so somehow all of his writings, many of which are very interesting and complicated. Anyway, so what you're going to have in this country is a slow awakening of socialism from a long hibernation called the Cold War, I never expected to be very honest with you that I would live to see, I knew it would come because these things always do, but I didn't expect to see it. So I have been surprised as have a lot of us that when it starts to happen it happens fast. </s><s>[INST] So the you see Bernie as an early sign of the awakening from the Cold War to accept the [/INST] idea was always socially we all knew and everybody who paid attention, he didn't hide it, but 2016 he goes, he makes a decision momentous to run for president. He's just a senator from Vermont, Vermont is one of the smallest states in the union. People who live in Vermont love to tell you that there are more cows than people in Vermont etcetera etcetera. Uh so here from this little state, this elderly gentleman with a new york city accent runs for all and says I'm a Socialist and when they attack him, he doesn't run away, I'm a socialist, I'm a Socialist. Now, he had been, it wasn't that it wasn't a secret that suddenly got out, but the great question and I don't mind telling you because I went to the right schools, I know a lot of people, you know, johnny Ellen was my classmate at Yale and stuff like that. So I was speaking with a high official of the Democratic Party and I said, well what do you think about Bernie entering the race Makes no difference is they get 1% of the vote right? He was wrong. He had they had no idea what was coming. But the truth is, I didn't either. It wasn't just that he didn't get it. I thought his 1% was probably right. So we were both wrong. </s> |
<s>[INST] Uh Yeah and that requires the renegade ceo. Right? I'm with you. I'm with you. I still have a little bit faith in Waymo for for the renegade ceo to step forward but [/INST] it's not it's not john krafcik </s><s>[INST] Yeah it's uh you can't [/INST] it's not chris johnston and those people may be very good at certain things but they're not renegades </s><s>[INST] because these companies are fundamentally even though we're talking about billion dollars, all these crazy numbers, they're still like early stage startups. [/INST] I mean and I just if you are pre revenue and you've raised $10 billion, I have no idea like like this just doesn't work. You know, it's against everything. Silicon Valley, where's your minimum viable product, you know, where's your users with your growth numbers? This is traditional Silicon Valley. Why do you not apply it to what you think you're too big to fail already? Like oh </s><s>[INST] how do you think autonomous driving will change society? So the mission is for comma to solve self driving. Do you have like a vision of the world of how it will be different. Is it as simple as a to be transportation or is there like because these are robots, [/INST] it's not about autonomous driving in and of itself. It's what the technology enables it's I think it's the coolest applied ai problem. I like it because it has a clear path to monetary value. Um but as far as that being the thing that changes the world, I mean no, like like there's cute things we're doing in common like who you thought you could stick a phone on the windshield middle drive. Um but like really the product that you're building is not something that people were not capable of imagining 50 years ago. So no it doesn't change the world in that front. Could people have imagined the Internet 50 years ago? Only true junior genius visionaries. Everyone could have imagined autonomous cars 50 years ago so I could core but I don't drive it. </s><s>[INST] See I have this sense and I told you like I'm my long term dream is robots with which you have deep with whom you have deep connections, right? And there's different trajectories towards that. And I've been thinking so been thinking of launching a startup. I see autonomous vehicles as a potential trajectory to that. I'm that's not where the direction I would like to go. But I also see Tesla even call me, I like pivoting into into robotics broadly defined at some stage in the way like you're mentioning the internet didn't expect. [/INST] Let's solve you know when I say a calm about this, we can talk about this but let's all self driving cars first. You gotta stay focused on the mission. Don't don't, don't, you're not too big to fail for however much I think commas winning like no, no, no, no, no, you're winning when you solve level five self driving cars and until then you haven't win and one and you know, again, you want to be arrogant in the face of other people, great, you want to be arrogant in the face of nature, you're an idiot. </s><s>[INST] Stay, mission focused, brilliantly put uh like I mentioned thinking of launching a startup, I've been considering actually before, Covid, I've been thinking of moving to san Francisco, [/INST] Oh if I wouldn't go there, </s><s>[INST] so why is uh well, and now I'm thinking about potentially Austin um and we're in SAn Diego now, Diego, So why what? Um I mean you're, you're such an interesting human, you've launched so many successful things. What ah why san Diego? What do you recommend? Why not san Francisco? Have you thought so in your case SAN Diego with qualcomm and snapdragon. I mean that's an amazing combination, [/INST] but that wasn't really why </s><s>[INST] that wasn't the y [/INST] you know what I mean, Qualcomm was an afterthought qualcomm was, it was a nice thing to think about it, like you can have a tech company here, a good one, I mean, you know, I like to come, but yeah, no um what's </s><s>[INST] the white Sandy go better than what san Francisco suck. [/INST] Well, so okay, so first off we all kind of said like we want to stay in California, people like the ocean, you know, California for, for its flaws. It's like a lot of the flaws of California are not necessarily California as a whole and they're much more san Francisco specific. Um, san Francisco. So I think first tier cities in general have stopped wanting growth. Uh, well you have like in san Francisco, you know, the voting class always votes to not build more houses because they own all the houses and they're like, well, you know, once people have figured out how to vote themselves more money, they're going to do it. It is so insanely corrupt. Um, it is not balanced at all. Like political party wise, you know, it's, it's, it's a, it's a one party city. And </s><s>[INST] for all the discussion of diversity, it has, it stops lacking real diversity of thought of background of uh, approaches the strategies of ideas. It's kind of a strange place that it's the loudest people about diversity and the biggest lack of diversity. [/INST] I mean, that's, that's what they say, right? It's the projection </s><s>[INST] projection. Yeah, it's interesting and even people in Silicon Valley tell me that's uh, like high up people, but everybody is like, this is a terrible place. It doesn't make, I mean, and [/INST] coronavirus is really what killed it. San Francisco was the number one, uh, exodus during coronavirus, </s><s>[INST] but you still think san Diego is a is a good place to be. [/INST] Yeah, I mean we'll see, We'll see what happens with California a bit longer term like Austin and Austin is an interesting choice. I wouldn't, I wouldn't, I don't know really anything bad to say about Austin either except for the extreme heat in the summer. Um, which, but that's like very on the surface, right? I think as far as like an ecosystem goes, it's cool. I personally love colorado. Uh, I mean, you have these states that are, you know, like just play better run. Um, California is, you know, it's especially san Francisco, it's on its high horse. And like, </s> |
<s>[INST] Yeah, so like when you're actually attacking your own citizenry, [/INST] Yeah, that was universally condemned Kristallnacht and that was very shocking. Uh it's a level of barbarism um to the west, because I think, I think we still want to believe understandably that things aren't as bad as they seem. We would rather this is why I you know, I the north Korea book I did. Dear reader is used in a humorous framework because if you have to look, it's like looking to the sun if you stare at it straight on, it's very hard to do. So you have to kind of look at it obliquely and and then you kind of realizing the enormity of the depravity. Um, and again, programs in Russia had been a thing for a very long time and there is a difference between, okay, you know, we're gonna sack these villages and persecute people and we're going to systematically exterminate them. That that's there's there's still levels of evil and depravity. </s><s>[INST] So you did write the book, Dear reader on kim jong Il, Dear reader, the unauthorized autobiography of kim jong il. So that's the previous leader of North Korea, current one is the on Jonathan. No, no creativity on the naming. [/INST] Well, no, this is intentional because it's a throwback to um the dad. </s><s>[INST] So there's been only three leaders in North Korea. So we've talked about the history of Hitler and Stalin men like these. I think it's important to understand that the history of those kinds of humans. There's the history of North Korea is not well written about or understood, which is why your book is exceptionally powerful and important. So maybe in a big broad way, can you say who was, who is kim jong Il as a, as a man as a leader, as a historical figure that we should understand. Why should we understand them. [/INST] So I wrote Dear Reader by going to North Korea and getting all their propaganda, which is translated into several languages because the conceit is, everyone on earth is interested in them and wants to mirror their ideology </s><s>[INST] And he died in 2011, And you wrote the book in 2012, [/INST] I went there in 2012, I wrote the book came out in 2014. So kim jong Il is um do not intellect. North Korea's version of forrest gump and that when they write their history, whenever something appeared happens, he's there. Uh and by telling his life story, it's in the first person, he's telling the history of north Korea. So I wanted to write the kind of book where in one book and it's the kind of reading you could do in the beach of the bathroom, you're going to get the entire history and know everything you need to know about North Korea and one accessible outlet. And it's it's what people don't appreciate about North Korea, that several things how bad it is and this didn't happen overnight. This was very systemic, that what this family did to that country where piece by piece. They did everything in their power to hermetically seal it from the rest of the world, ramp up the oppression, keep any information from coming in and you know, they're very creative and innovative in their style of manipulation and control. Um So there is a farcical element, let me give you an example. So people in the west kind of get it wrong, they're talking about. So they talk about when Kim Jong Il played golf for the first time, you get 17 holes in one. There's this one story about kim jong il shrinking time and this is a story how it sounds supernatural, but it's not so kim jong Il is at a conference, the dear leader and someone is giving a talk and while that person is giving a talk, kim jong Il is taking notes and working on his work and he has an aide who keeps interrupting him with questions and the speaker keeps stopping and kim jong Il says while you're stopping goes, I see you're doing these other things and he goes, no no he can I can do all these things at once. Everyone's shocked and they said this is why kim jong Il looks at time, not like a plane but like a cube and he can shrink time. And my friend goes, do they mean multi tasking? And yes kim jong Il is the only person north Korea who's capable of multitasking. So in order to elevate him, they basically make everyone else in North Korea completely, you know incompetent. Um and that has a purpose because should the leader go away, this country is going to collapse overnight. So this, they laugh in the west about all these newspapers, show him you know at the factory and he's at the fish hatchery at the paper plant. They say the difference in north Korea is that the leader goes among the people and does what he called field guidance. So he will go in that farm and be like this is what you need to do? And he'll go here and he's so smart, he's good at everything. And thanks to him for sharing his wisdom with us and he's not removed from the people, like in every other country, </s><s>[INST] why does that seem to go wrong with humans? Do you think that this kind of the structure where there is this one figure, this authoritarian is the totalitarian structure where there's one figure that's a source of comfort. Knowledge, [/INST] kim jong il is not good at farming, kim jong il is not good at the machinery. It's all a complete lie or the things you point out the things that are completely obvious. So here's another example if they use uh in north Korea, they have something called the tower, the jewish idea which is an obelisk, which looks like the Washington monument, but it's completely different because it's got this like plastic torch at the top and they talked about in the propaganda how all the architects got together and they said, oh we should make this the second tallest stone obelisk in the world. And kim jong Il says, no, let's make it the tallest there. Like we never thought of this before and the way it's presented as it and like he's the first person think of this like these architects are having a brainstorming session, the towers, that just the idea they're like, all right, we gotta do something innovative to put north Korea on the map, what can we do? How about second biggest, he's gonna go for this and then he's like make no, we never thought of this. It's so because I presented at face value, people sometimes say the books a satire, it's not a satire. I downplayed all this stuff, it's a farce. Here's another example. North Korea is very big and I think Russia is to some extent to an amusement parks, fun fairs. They call them in the british style because it's a chance for the people to all to get together. And there was this amusement park, it's almost like south park cartman where there's all these rides and kim jong il is like, I'm not gonna let any elderly or Children take these rides until I put myself in danger and ride them myself and they go, but dear leader, it's drizzling and he goes, no, I have to make sure these rides are going to be safe for everyone, even during the light rain to go, well, can we go on these rides with you? No, no, no, I mean the courageous one and he's riding all the rides and they're standing there crying at his courage. But that's what you ask all the things. One power. It's like, listen, I'm quite confident that those funfair engineers are in a position to ride modest Mouse River, It's called by themselves and be like, yeah, okay, this is good for the kids. </s> |
<s>[INST] So to push back on a few things. So, one is the idea of this society calling itself democratic, is that the government is elected democratically, and the government is able to pressure the workplace through the process of regulation. You pass laws of the boundaries of how, you know, minimum wage, all those kinds of things. That's the one idea. The other is there is a natural force within the capitalist when there's no monopolies of competition. Being the accountability. So, if you're a shitty boss, the employee, the employee in the capital system as the freedom to move to another company work for a better boss. So that that creates pressure on the employers and the bosses. That's at least the idea that you there's two boundaries of you not misbehaving. One is the law. So regulatory regulations bypassed by the government. Democratic. And the second is because there's always alternatives. In theory, then that puts pressure on everyone to behave well, because you can always leave. Uh, so, I mean, that's kind of accountability, but what you're saying is that does not result in any, in a significant enough accountability or the employer that avoids exploitation of the worker. [/INST] Absolutely. I mean, whatever accountability you get in those mechanisms and let me respond to that. And then the counter argument. First competition here again, we have to be Higa Leon's just a little competition destroys itself. It doesn't need any, the whole point of competition is to beat the other guy. If I can produce the same product as the other guy. Either a better quality or a lower price or maybe both. Then I win because the customers will come to me because my price is lower or my quality is better. And they'll leave the other guy. He'll go out of business. Now, let's follow when he goes out of business because I won the competition, he fires his workers. I hire them because I'm now going to be able to serve a market, he can't serve anymore. So I'm gonna, I'm gonna buy the used equipment, I'm gonna, and thereby many become fuel monopoly, is the product of competition. It's not the antithesis, it's the product. </s><s>[INST] Well, let's see, [/INST] that's where it comes from, </s><s>[INST] but there's another, there's another element to the system where there's always a new guy that comes in. Uh, well that's the, that's the dream. The entrepreneurial spirit of a free, uh, you know, of, of the United States, for example, of the capitalist system is uh, you can be um broke and one day have a strong idea and build up a business that takes on google and facebook and twitter and all the different car. Four G. M, which is what you know, you look at Tesla for example, right? That's the american dream. One of the many ideals of the american dream is you can, um, uh, move from dirt poor to being the richest person in the world. And um, [/INST] it can happen, </s><s>[INST] it can happen, but you're [/INST] like, that's like, you can win the lottery. </s><s>[INST] No, that's not quite know. The lottery is complete luck here. You can work your ass off if you have a, that's not, that's not true. There's a lot of new businesses. [/INST] How many Teslas do you do? You know, </s><s>[INST] this is a really bad example because the car company in the automotive sector is so difficult to, uh, they operated such a thin margin of profit that they're probably a good example of like capitalism just completely coming to a halt in terms of lack of innovation because, you know, that's a very complicated industry because of the supply chain so [/INST] much, come on, they have their uniqueness is you're quite right. But so does every other industry. The one thing that's common is that many become few. What you can also have is when you have a few, they jack up the price, they make an enormous profit. And in the irony of capitalism, Marx would love this. They begin to incentivize people to break into this industry because the few remaining are making a wild amount of profit because they are a few and can jigger the market to make it work like that for them. </s><s>[INST] So, the [/INST] reason every every small capitalist is trying to build market share. That's a polite way, polite way of saying they want to become a monopolist or to be more exact, an oligopoly list. One of a handful of firms that dominates, that's what they're there for. </s><s>[INST] But yeah, to push back a little bit also, because that's could be this is a question. Also, do you think we're in danger of oversimplifying capitalism that completely removes the basic decency of human beings? So, if you give me a choice to press, to press a button to get rid of the competition and that, but that's going to lead to a lot of suffering. There's a lot of people at the heads of companies that won't press that button, that it's not in the calculation is not just money, it's human well being too. So like, [/INST] uh </s><s>[INST] I yes, so you're saying that the forces of capitalism take over the minds of the people at the top and then they they cease being human, [/INST] essentially </s><s>[INST] the basic, [/INST] depending </s><s>[INST] on your model of humans, but they lose track of the better angels of their nature and they just become cogs in the machine. But they just happen to be the [/INST] differently that the system is so set up. It's a little bit like natural selection. The guys who may, I could say that women too, it doesn't matter. The people who make it up through the layers of the bureaucracy and get to the top. And these things have had to do things along the way that becomes selective if they can't stand it because they have that human quality and there are people like I've known them, they're the ones running an Airbnb in in Vermont, they checked, they went there and they said I'm not doing this anymore, I'm not going to treat people like that. I'm gonna make a lovely place in Vermont with my husband or my wife or whatever and I'm gonna be, you know, enjoying the people that come by and be a decent of course, of course. But the system selects the firm, you know, if you don't do what has to be done to make the profit, go up, you're toast there anyway that the rest of the people who vote for you are going to kick you out. You can tell them all day long what a lovely person you are, then they're gonna look at you and wonder what happened to you. How did you even get this far with a lovely person horse? </s> |
<s>[INST] Fascinating to think that COVID-19 is uh as a result of a virus evolving together with like twitter, like figuring out how we can sneak past the defenses of the humans. So it's not bad enough. And then then the misinformation, all that kind of stuff together is operating in such a way that the virus can spread effectively. I wonder, I mean obviously uh virus is not intelligent, but there's a uh there's a rhyme and rhythm to the way this whole evolutionary process works and creates these fascinating things that spread throughout the entire civilization. [/INST] Absolutely. It's um yeah, I'm I'm completely fascinated by this idea of social media in particular how it replicates how it grows. I've been how it how it actually starts interacting with the biology of the virus masks who's going to get vaccinated politics. Like these seem so external to virus biology, but it's become so intertwined and and it's interesting and I actually think we could find out that the virus actually becomes obviously not intentionally, but you know, we could find that choosing people choosing not to our masks, choosing not to counter this virus in a regimented and sort of organized way effectively gives the virus more opportunity to escape. We can look at vaccines, you know, we're about to, we're about to have one of the most aggressive vaccination programs the world has ever seen. But we are unfortunately doing it right at the peak of viral transmission when millions and millions of people are still getting infected. And when we do that, that just gives this virus so many more opportunities. I mean, orders of magnitude more opportunity to mutate around our immune system. If we were to vaccinate everyone when there's not a lot of virus, then there's just not a lot of virus. And so there's not going to be as many, you know, I don't even know how many zeros are at the end of however many viral particles there are in the world right now, you know more than quadrillions, you know, and so if you assume that at any given time somebody might have trillions of virus in them and any given individual, so then you know multiply trillions by millions and you know, you get a lot of viruses out there and if you start applying pressure, ecological pressure to this virus that that when it's not abundant, I had the opportunity for a virus to sneak around immunity, especially when all the vaccines are identical essentially. Um it </s><s>[INST] sticks is one to mutate and then jumps takes [/INST] one in the whole world, you know, and we have to, we have to not forget that this particular virus was one, it was one opportunity and it has spread across the globe and there's no reason that can't happen tomorrow. I knew, you know it's scary. </s><s>[INST] I have a million other questions in this direction, but I'd love to talk about one of the most exciting aspects of your work which is testing or rapid testing. You wrote a great article in time On November 17. This is like a month ago about rapid testing titled How we can stop the spread of COVID-19 by Christmas, let's jot down the fact that this is a month ago. So maybe your timeline would be different, but let's say in a month. So you've talked about this powerful idea for quite awhile. Throughout the COVID-19 pandemic. How do we stop the spread of COVID-19 in a month. [/INST] Well, we uh we use tests like these, you know, So so the only reason the virus continues spreading is because people spread it to each other. This isn't this isn't magic. Um and so there's a few ways to stop the virus from spreading to each other. And that is uh you either can vaccinate everyone and vaccinating everyone is a way to immunologically prevent the virus from growing inside of somebody and therefore spreading. We don't know yet actually if this vaccine, if any of these vaccines are going to prevent onward transmission. So uh so that may or may not serve to be one opportunity. Certainly I think it will decrease transmission. But the other idea that we have at our disposal now, we had it in May, we had it in june july august september october november. And now it's december. We still have it. We still choose not to use it in this country and in much of the world. And thats rapid testing that is giving its empowering people to know that they are infected and giving them the opportunity to not spread it to their loved ones and their friends and neighbors and whoever else. Um We could have done this. We still can today. We could start we have millions of these tests. These tests are simple paper strip tests. They are uh inside of this thing is just a little piece of paper. Um No and I can actually open it up here. There we go. So this this is how we do it right here. We have this little paper strip test. This is enough to let you know if you're infectious With somewhere around the order of 99% sensitivity, 99% specificity. You can know if you have infectious virus in you, If we can get these out to everyone's homes, build these, make 10,000,000, million, 30 million of them a day. We make more bottles of Dasani water every day. We can make these little paper strip tests and if we do that and we get these into people's homes so that they can use them twice a week. Then we can know if we're infectious, you know, is it perfect? Absolutely not. But is it near perfect. Absolutely. You know, and so if we can say, hey, the transmission of this is, you know, for every 100 people that get infected right now, They go on to in fact maybe 130 additional people. And that's exponential growth. 100 becomes 130, a couple days later that 100 30 becomes Another another 100 65 people have now been infected. And you know go over three weeks and 100 people become 500 people infected. Now it doesn't take much to have those 100 people, not in fact 130, but in fact 90 all we have to do is remove say 30 40% of new infections from continuing their spread. And then instead of exponential growth you have exponential decay. So this doesn't need to be perfect. We don't have to go from 100 to 0. We just have to go and have those 100 people infect 90 and those 90 people. In fact, you know, 82 whatever it might be and you do that for a few weeks and boom, you have now gone instead of 100 to 500 you've gone from 100 to 20. It's not very hard. And so the way to do that is to let people know that they're infectious. I mean I were perfect example right now I I this morning I use these tests To make sure that I wasn't infectious. Is it perfect. No, but it reduced my odds 99%. I already was at extremely low odds because I've spent my life quarantining these days. </s> |
<s>[INST] there's also a dance between the genius and the the personal like what it means to love another human being and [/INST] there's a lot of things there it's just a beautiful movie </s><s>[INST] and then the fatherly figure the mentor in the in the psychiatrist and it [/INST] like really like uh it messes with you you know there's some movies that just like really messed with you uh on a deep level </s><s>[INST] do you relate to that movie at all? [/INST] No </s><s>[INST] It's not your fault. As I said lord of the rings. That's self explanatory terminator two. Which is interesting. You? Re watch that a lot. Is that better than terminator 1? You like? You like I [/INST] do like terminator one as well. Uh I like Terminator two a little bit more but in terms of like its surface properties </s><s>[INST] do you think china is at all a possibility? [/INST] Yes. </s><s>[INST] Like the actual sort of autonomous weapons system kind of thing. Do you worry about that stuff [/INST] being used </s><s>[INST] for war? [/INST] 100% worry about it. And so the I mean the you know some of these fears of A. Gs and how this will plan out. I mean this will be like very powerful entities probably at some point. And so um for a long time they're going to be tools in the hands of humans. You know people talk about like alignment of A. G. I. S. And how to make the problem is like even humans are not aligned. So how this will be used and what this is going to look like is um yes troubling. So </s><s>[INST] do you think it'll happen slow slowly enough that we'll be able to as a as a human civilization? Think through the problems [/INST] yes that's my hope is that it happens slowly enough and an open enough way where a lot of people can see and participate in it. Just figure out how to deal with this transition I think, which is gonna be interesting. I </s><s>[INST] draw a lot of inspiration from nuclear weapons because I sure thought it would be would be fucked once they develop nuclear weapons. But like it's almost like uh when uh when the systems are not so dangerous, they destroy human civilization, we deploy them and learn the lessons and then we quickly, if it's too dangerous, we quickly quickly we might still deploy but very quickly learned not to use them. So they'll be like this balance achieved. Humans are very clever as a species. It's interesting we exploit the resources as much as we can, but we don't, we avoid destroying ourselves. It seems like, [/INST] well, I don't know about that actually. </s><s>[INST] I hope it continues. [/INST] Um </s><s>[INST] I [/INST] mean I'm definitely concerned about nuclear weapons and so on, not just as a result of the recent conflict even before that, that's probably like my number one concern for humanity. </s><s>[INST] So if humanity destroys itself or destroys, you know, 90% of people, that would be because of nukes. I [/INST] think so. Um it's not even about full destruction. To me, it's bad enough if we reset society that would be like terrible. It would be really bad. And I can't believe we're like so close to it. </s><s>[INST] It's like [/INST] so crazy to me. It </s><s>[INST] feels like we might be a few tweets away from something like that. Yeah, [/INST] basically it's extremely unnerving, but and has been for me for a long time </s><s>[INST] it seems unstable that world leaders just having a bad mood can like um take one step towards a bad direction and it escalates because of a collection of bad moods. It can escalate without being able to stop. [/INST] Yeah. It's just it's a huge amount of power. And then also with the proliferation basically I don't I don't actually really see, I don't actually know what the good outcomes are here. So I'm definitely worried about it a lot. And then a gi is not currently there. But I think at some point will more and more </s><s>[INST] become something [/INST] like it. The danger with A G. I even is that I think it's even like slightly worse in the sense that uh there are good outcomes of A. G. I. And then the bad outcomes are like an epsilon away like a tiny runway. And so I think um capitalism and humanity and so on will drive for the positive ways of using that technology. But then if bad outcomes are just like a tiny like flipping minus sign away. That's a really bad position to be in </s><s>[INST] a tiny perturbation of the system results in the destruction of the human species. It's a weird line to walk. Yeah. [/INST] I think in general it's really weird about like the dynamics of humanity in this explosion we talked about is just like the insane coupling afforded by </s><s>[INST] technology and [/INST] just the instability of the whole dynamical system. I think it just doesn't look good honestly. </s><s>[INST] Yes, that explosion could be destructive and constructive and the probabilities are non zero in both. I'm [/INST] gonna have to I do feel like I have to try to be optimistic and so on. And I think even in this case I still am predominantly optimistic, but there's definitely </s><s>[INST] me too. Do you think will become a multi planetary species? [/INST] Probably, yes, but I don't know if it's a dominant feature of uh future humanity. Uh there might be some people on some planets and so on, but I'm not sure if it's like, yeah, it's like a major player in our culture and so on. We </s><s>[INST] still have to solve the drivers of self destruction here on earth. So just having a backup on MArs is not going to solve the problem. [/INST] So by the way, I love the backup on MArs, I think that's amazing. We should absolutely do that. And I'm so thank you. </s><s>[INST] Uh would you go to MARS [/INST] personally? No, I do like Earth quite a lot. I'll </s> |
<s>[INST] So just for people to not familiar, I mean the way you detect aircraft. So I mean I'm sure there's a lot of ways, but radar for the longest time, there's a big blob that appears in the radar. How do you make a plane disappear? So it looks as big as a ball bearing. What's involved in technology wise there, what's uh broadly sort of the stuff you can [/INST] speak about. They'll stick to what's in Ben Rich's book, but obviously the geometry of how radar gets reflected and the kinds of materials that either reflect or absorb or some kind of a couple of the critical elements there. Um and it's a cat and mouse game. Right? I mean, you know, radars get better stealth capabilities get better. And so it's a it's a really uh game of continuous improvement and innovation there. I'll leave it at that. </s><s>[INST] Yeah. So the idea that something is essentially invisible Is quite fascinating, but the other one is flying fast. The speed of sound is 750, mph. Uh, so supersonic is three, you know Mach three, something like that. [/INST] When we talk about the supersonic obviously and we kind of talked about that is that realm from Mach one up through about Mach five and then hypersonic so you know Hi Supersonic Speeds would be passed Mach five. And you gotta remember Lockheed martin and actually other companies have been involved in hypersonic development since the late sixties. Um You know you think of everything from the X 15 to the space shuttle as examples of that. I think the difference now is uh if you look around the world particularly the threat environment that we're in today you're starting to see you know publicly uh folks like the Russians and the chinese um saying they have hypersonic weapons capability that could threaten uh U. S. And allied capabilities. And also basically you know the claims are these could get around defensive systems that are out there today And so there's a real sense of urgency. You hear it from folks like the undersecretary of defense for research and engineering. Dr mike Griffin and others in the Department of Defense that hypersonic is uh is something that's really important to the nation in terms of both parody but also defensive capabilities. And so that's something that we're pleased. It's something Lockheed martin's had a heritage in. We've invested R. And D. Dollars on our side for many years. And uh we have a number of things going on with various us government customers in that field today that we're very excited about. So I would anticipate we'll be hearing more about that in the future from our customers. </s><s>[INST] And I actually haven't read much about this. Probably you can't talk about much of it at all. But on the defensive side is a fascinating problem of perception of trying to detect things that are really hard to see. Can [/INST] you comment on how </s><s>[INST] hard that problem is and uh how how hard is it to stay ahead? Uh even if we're going back a few decades, stay ahead of [/INST] the competition? Well, maybe maybe I again you got to think of these is ongoing capability development. And uh so think back to the early days of missile defense. So this would be in the eighties the S. D. I. Program. And in that time frame we proved Lockheed martin proved that you could hit a bullet with a bullet essentially. And uh which is something that had never been done before to take out an incoming ballistic missile. And so that's led to these incredible hit. To kill kinds of capabilities. Pack three. Um that's the Patriot advanced capability uh model three. That Lockheed martin builds the thaad system that I that I talked about. Um so now hypersonic uh you know, they're different from ballistic systems. And so we got to take the next step in defensive capability. </s><s>[INST] I can I'll leave that there. But I can only imagine. No let me just comment sort of as an engineer. It's sad to know that so much that locket is done in the past is classified or today, you know and it's shrouded in secrecy. It has to be by the nature of the application. So like what I do so what we do here at M. I. T. Would like to inspire young engineers, young scientists and yet in the lucky case some of that engineer has to stay quiet. How do you think about that? How does that make you feel? Is there a future where more can be shown? Or is it just the nature the nature of this world that it has to remain secret? [/INST] It's a good question. I think the public can see enough of and including students who may be in grade school, high school college today. Uh to understand the kinds of really hard problems that we work on And I mean look at the F- 35, right? And uh you know obviously a lot of the detailed performance levels are are sensitive and control. But you know we can talk about what an incredible aircraft this is. You know, supersonic, super cruise kind of a fighter, a a you know stealth capabilities, its a flying information, you know, system in the sky with data fusion sensor fusion capabilities that have never been seen before. So these are the kinds of things that I believe, you know that these are kinds of things that got me excited when I was a student. I think these still inspire students today. And the other thing and say, I mean, you know, people are inspired by space, people are inspired by aircraft. Our employees are also inspired by that sense of mission and I'll just give you an example. I had the privilege to work and and lead our GPS programs for some time. And uh that was a case where you actually worked on a program that touches billions of people every day. And so when I said, I worked on GPS, everybody knew what I was talking about, even though they didn't maybe appreciate the technical challenges that went into that. But I'll tell you got a briefing one time from a major in the Air force and he said, I I go by call sign Gimp Gps is my passion. I love GPS and he was involved in the operational test of the system. He said, I went, I was out um in Iraq and I was on a helicopter, uh Black hawk helicopter um uh it was bringing back, you know sergeant and a handful of troops from a deployed location. And uh you know, he said my job is GPS. So I asked that sergeant, he's you know beaten down and kind of half asleep. And I said, what do you think about GPS? And he brightened up, his eyes lit up and he said, well, GPS that brings me and my troops home every day. I love GPS. And that's the kind of story where it's like, okay, I'm really making a difference here in the kind of work. So that that mission piece is really important. The last thing I'll say is, and this gets to some of these questions around advanced technologies. It's not, you know, they're not just airplanes and spacecraft anymore for people who are excited about advanced software capabilities, about ai about bringing machine. These are the uh, these are the things that we're doing to exponentially increase the mission capabilities that go on those platforms and those are the kinds of things I think are more and more visible to the public. </s> |
<s>[INST] do you think he should not have been banned from all these platforms, [/INST] I don't think, I do think that if you are a private company, right? I do think, and this is where you run into this this problem, like I I don't know if these tech companies were government utilities, would that decrease people's likelihood of being banned? I don't know. Right? So I'm I'm I understand the benefit of them being treated like public utilities and people thinking they have the right to a twitter. I've never I don't know, I have very little confidence. I mean the government trying to roll out a vaccine in California and we vaccinated like five people. I mean in terms of what we need to do in the state, right? So maybe if it was a government utility, I I do think someone like Alex, just like there should be some process, so if you're going to get rid of someone, they should have a way to defend themselves, there should be more democratic uh process that you can go through than just being unilaterally taken off something. But like then you run into the like, am I going to say that everyone deserves no if you're threatening or harassing people are threatening to kill them, publishing their private information if you're committing crimes on these platforms, obviously the people that own these platforms are gonna be like, we're not gonna allow this to happen. So I understand that there is a line, right? There is some like people that say there is no line aren't really thinking like there is a line. Uh I just thought that line seems to be moving all the time and it seems to be a very hard thing to police, but I don't think you can remove a guy off everything and then also bank accounts won't give him debit cards or credit cards, I don't know if you talk to him about that, but like you know, there were financial institutions that were refusing to let him, you know, park his money there. So I mean it really does get pretty terrifying pretty quickly um probably </s><s>[INST] without any transparency from those companies, so you're right, it feels like there should be a process of just having for him to defend themselves. I think [/INST] there needs to be a process for people to defend themselves. I every day I wake up and I go, can is something I said in the video gonna get taken out of context, is somebody gonna get angry? Is somebody gonna be, you know, I say wild stuff because that's what makes me laugh, that's what makes my friends laugh and that's what makes my audience laugh. So I never ever people, you know, whatever whatever political side you come down on, I think if you make your living speaking, it's always interesting to me if you are pro the the platform, </s><s>[INST] that's odd, it's interesting to consider a kind of a jury uh context to where you know, there's transparency about why you're a video about bombing um Disney world might be taken down, like it's uh it gets taken down and then there is uh it's almost like creating a little court case, a mini court case and not in a legal sense, but in the in the public sphere. And then people should be able to have, you know, you pick representatives of our current society and have a discussion about that and make a real vote. You know, just have like, jury locks himself up in a discussion that kind of, that kind of process might be necessary right now. What happens is twitter is completely first of all, they're just mostly not aware of everything they're doing, there's too much stuff, but the stuff they're aware about, they make the decision in closed doors uh meetings and without any transparency to the rest of the company actually, but also transparency to uh the rest of the world. And so, and then all they say is we're doing, we're making these decisions because the people, they use things like violence. So violence equals bad. And if this person is quote unquote inciting violence, therefore, that gives us enough enough reason to bend them without any kind of process. It's, I mean, it's interesting, I'm, I'm torn in the whole thing if if it was indeed there's no transparency about it, but if parler was indeed inciting violence, like if there was brewing of violence, potential violence where, you know, thousands of people might die because of some kind of riot like the, this is scary thing about mob about when a lot of people get together or good people like legitimately good people that love this country that don't see enemies yet uh, around them. But if they get excited together and there's guns involved and then some cop gets nervous and shoots one person, another person shoots the cop. And then there's a lot of shooting involved and then it goes from five people dying in the capital to thousands of people dying in the capital. [/INST] Well, in fairness to to defend the people of cap, they didn't shoot the cop. They bludgeoned him to death with a fire extinguisher. Yes. So I do, I do wanna just kind of put that out as a, as a defense of them. Listen, I'm sure there was some wild ship going on on parler. And I think the problem here's the problem, right? There's a lot of people that just want to go on these sites and say they want to kill everyone. And the problem is, you know, at what point do you shut them all down? Like, I think a lot of people are just living in a world with their powerless. They don't have any political power. They don't have any economic power, right? They can't throw their money around. Uh, they don't have healthcare, their their job security isn't great. Uh, there might be living in a community that doesn't have the resources, they would like it to have. Uh they're they're not happy and thrilled and then they have these sites where they can go on and just say, man, I'd like to fucking burn it all down and distinguishing a guy blowing off steam and saying wild stuff from a genuine threat is a very hard thing to do. You know? Like I have, I've threatened to kill uh I got banned from Airbnb. I threatened to kill the people. Um That band be comedically comedically, this is a joke. I'm not going to kill you. This is a joke cause I'm blowing off steam and I'm angry. Do you know how many people that my parents like my dad's like, I'm gonna fucking kill this guy, my mom like I'm gonna fucking kill and they were talking about each other but it's, but none of it ever happened. But we should be, I think you, you have to create a space for people to threaten to overthrow the government. As long as they don't violently do it? Like I mean does that make any sense? Like I mean as long as they're not going to go hurt innocent people, what are you gonna do? Like there's so many people out there that she's that's why a lot of these things like, you know, four chan these sites, a lot of people going on there, They just want to say the most fucked up ship because they, it's the thing that gives them, they can laugh or they can release steam and it's, it is a mature, it is stupid. It's not productive. It's not, you know, but at the end of the day, if you're not going to give people health insurance, you gotta give them something. It's like when someone in this country dies that everyone disagrees with right political figure, media figure a lot of people dance on their grave online and then everyone people goes, the and the other side will always do it like a conservative dies and everyone goes great conservatives goes, and this is grotesque that you, and then when RBG dies, they all have parties and the conservatives go great. Um, you have to let people in this country enjoy the deaths of their enemies. </s> |
<s>[INST] That's sort of the easier question because because it's a very engineering answer to the question. That's the turing test would only in consciousness. But if you were to go, do you think it's possible that we need to have? So that kind of self awareness? [/INST] Uh we may, yes. So, for instance, I I personally think that when test a machine or a person in a touring test in an extended tour intestine, I think consciousness is part of what we require in that test. You know, implicitly to say that this is intelligent Christof disagrees. </s><s>[INST] So, yes, he does it despite many other romantic notions he holds, he disagrees with that one. Uh [/INST] All right. So, you know, we'll see. </s><s>[INST] Do you think as a quick question Ernest Becker fear of death. Do you think mortality and those kinds of things are important for well, for consciousness and for intelligence, the finite nous of life, finite nous of existence Or is that just a side effect of evolutionary side effect that's useful to uh financial selection. Do you think this kind of thing that we're doing this interview is going to run out of time soon? Our life will run out of time soon. Do you think that's needed to make this conversation good and life good. [/INST] You know, I never thought about it is an very interesting question. I think steve jobs in his commencement speech at stanford argued that, you know, having a finite life was important for for stimulating achievements. So it was a different, </s><s>[INST] You live every day like it's your last right. [/INST] Yeah. So rationally, I don't think strictly you need mortality for consciousness, but </s><s>[INST] who knows, they seem to go together in our biological systems you've mentioned before and students are associated with uh Alphago mobilize the big recent success stories in the eye and I think it's captivated the entire world of what ai can do. So what do you think will be the next breakthrough? And what's your intuition about the next breakthrough? Yeah [/INST] of course, I don't know where the next breakthroughs is. I think that there's a good chance, as I said before, the next breakthrough would also be inspired by you know neuroscience but which one </s><s>[INST] I don't know and there's so m I. T. Has this quest for intelligence and there's a few moonshots which in that spirit, which ones are you excited about? What, which projects kind of? Well [/INST] of course I'm excited about one of the moon shots which is our center for brains minds and machines with The one which is fully funded by NSF and it is about visual intelligence. </s><s>[INST] That one is particularly about understanding [/INST] visual intelligence of visual cortex and and visual intelligence in the sense of how we look around ourselves and understand the word around ourselves, you know, meaning what is going on, how we could go from here to there without hitting obstacles. Um you know whether there are other agents, people in the moment, these are all things that we perceived very quickly and and it's something actually quite close to being conscious. Not quite, but you know there is this interesting experiment that was run at google X um which is, in a sense is just uh virtual reality experiment, but in which they had uh subjects sitting in a chair with goggles like Oculus and so on the earphones and they were seeing through the eyes of a robot nearby two cameras, microphones for receiving. So their sensory system was there and the impression of all the subject very strong, they could not shake it off was that they were where the robot was, they could look at themselves from the robot and still feel they were they were with the robot is they were looking their body, their self had moved. </s><s>[INST] So some aspects of scene understanding has to have ability to place yourself, have a self awareness about your position in the world and what the world is. So yeah, so we may have to solve the hard problem of consciousness to something [/INST] on their way. </s><s>[INST] It's quite, quite a Moonshot. So you've been an advisor to some incredible minds uh including Dennis Hassabis Krystof co, I'm not Joshua, like you said, all went on to become seminal figures in their respective fields from your own success as a researcher. And from perspective as a mentor of these researchers, having guided them in the way of advice, what does it take to be successful in science and engineering careers, whether you're talking to somebody in their teens, twenties and thirties. What does that path look like? [/INST] It's curiosity and having fun. And I think it's important also having fun with other curious minds. </s><s>[INST] It's the people who surround with two. So fun and curiosity. Is there mentioned steve jobs? Is there also an underlying ambition that's unique that you saw or is it really does boil down to insatiable curiosity and fun? Well, [/INST] of course, you know, it's been curious in a active and ambitious way. Yes, and uh, um, definitely. But I think sometime in, in science and there are friends of mine were like this. You know, there are some of the scientists like to work by themselves and kind of communicate only when they complete their work or discover something. Um, I think I always found the actual process of discovering something is more fun if it's together with other intelligent and curious and fun people. </s><s>[INST] So if you see the fun in that process, the side effect of that process will be the election end up discovering some things. So as you've led, uh, many incredible efforts here, what's the secret to being a good advisor, mentor leader in a research setting? Is it a similar spirit or Yeah, what, what, what advice could you give to people, young faculty and so on? [/INST] It's partly repeating what I said about an environment that should be friendly and fun and uh, ambitious and uh, you know, I think I learned a lot from some of my advisors and friends and some of our physicists and that there was for instance is um behavior that was encouraged of when somebody comes with a new idea in the group. Yeah. Unless it's really stupid. But you are always enthusiastic and then you're enthusiastic for a few minutes for a few hours. Then you start, you know, asking critically a few questions, testing this. But you know, this is a process that is I think it's very it's very good. This you have to be enthusiasts. Sometimes people are very critical from the beginning. That's that's that's not </s> |
<s>[INST] But in terms of the it's interesting to me because you're a brilliant ah thinker in the space of machine learning, like intuition like digging deep in what works, what doesn't, why it works, why it doesn't. And so is Ilia, I'm wondering if there's interesting deep discussions you've had with him in the past? Our disagreements that were very productive. So [/INST] I can say I also understood over the time where are my strength? So obviously we have plenty of AI discussions and uh I am and you know, I myself have plenty of ideas but like I consider Ilya uh one of the most prolific II scientists in the entire world and uh I think that um I realized that maybe my super skill um is being able to bring people to collaborate together. That I have some level of empathy that is unique in a I word and that might come, you know from either meditation, psychedelics or let's say I read just hundreds of books on this topic. So, and I also went through a journey of, you know, I developed all sorts of, I'll get them. So I think that maybe I can, that's my superhuman skill. Uh Elia is uh one of the best ai scientists, but then I'm pretty good in assembling teams and I'm also not holding to people like I'm growing people and then people become managers. That opening, I like grew many of them like a research managers. </s><s>[INST] So you, you find you find places where your excellent and he finds like his his his deep scientific insights is where he is and you find ways you can, the puzzle pieces fit together [/INST] correctly, you know, ultimately for instance, let's say elia, he doesn't manage people, uh that's not what he likes or So uh I I like I like hanging out with people that by default, I'm an extrovert and I cared about people. </s><s>[INST] Okay, cool. So that that is perfectly together. But I mean, uh I also just like your intuition about various problems in machine learning. He's definitely one I really enjoy, I remember talking to him about something I was struggling with, which is coming up with a good model for pedestrians for human beings across the street in the context of autonomous vehicles. And he immediately started to like formulate a framework within which you can evolve a model for pedestrians like through self play all that kind of mechanisms, the depth of thought on a particular problem, especially problems, he doesn't know anything about it. It's fascinating to watch it makes you realize like um yeah the the limits is that the human intellect might be limitless. I was just impressive to see a descendant of a pop come up with clever ideas. [/INST] Yeah I mean so even in the space of deep learning, when you look at various people, there are people now who invented uh some breakthroughs ones but there are very few people who did it multiple times. And you can think if someone invented it, ones that might be just a short black and if someone invented it multiple times. You know, if a probability of inventing it once it's one of a million then probability of inventing it twice or three times will be won over a million square or or to the part of three which which would be just impossible. So it literally means that it's given that is not the lack and L. A. Is one of the few people who uh who have a lot of these inventions in his arsenal. It also feels that now for instance if you think about folks like girls or oiler uh you know at first they read a lot of books and then they did thinking and then they figure out math. Mhm. And that's how it feels with Ilya, you know at first he read stuff and then like he spent his thinking cycles and </s><s>[INST] that's a really good way to put it when I talked to him. [/INST] I I </s><s>[INST] see thinking he's actually thinking like he makes me realize that there's like deep thinking that the human mind can do like most of us are not thinking deeply uh like you really have to put in a lot of effort to think deeply. Like I have to really put myself in a place where I think deeply about a problem. It takes a lot of effort. It's like uh it's like an airplane taking off or something you have to achieve deep focus. He's just uh he's what is his brain is like a vertical takeoff in terms of airplane analogy. So it's interesting. But it I mean cal Newport talks about this is ideas of deep work. It's you know, most of us don't work much at all in terms of like like deeply think about particular problems. Well there is a math engineering, all that kind of stuff you want to go to that place often. And that's real hard work and some of us are better than others at [/INST] that. So I think that the big piece has to do with actually even engineering our environment that such that it's conducive to that. So um see both alien and I've on the frequent basis we kind of disconnect ourselves from the world in order to be able to do extensive? I'm on the thinking. So usually he just leaves ipad at hand, he loves his ipad. Uh for me I'm even sometimes you know just going for a few days to different locations to Airbnb, I'm turning off my phone and there is no access to me and uh that's extremely important for me to be able to actually just formulate new thoughts to do deep work rather than to be reactive. And the older I am, the moral of this like random tasks are kind </s> |
<s>[INST] actually linger on anonymous. What do you still understand what the heck is anonymous? [/INST] Just a place where you hang out. I mean it's just started on fortune with a chain and it's really just anyone, you could be anonymous right now if you wanted to just you're in there hanging out in the channel now you're probably not gonna get much cred until you work your way up and prove who you are. Someone vouches for you. But anybody can be an anonymous and we can leave anonymous. </s><s>[INST] What's the leadership of anonymous? Do you have a sense that there is a leadership? [/INST] There's a power play now, is that someone that you know, that says this is what we're doing all we're doing. I </s><s>[INST] love the philosophical and the technical aspect of all of this. But I think there is a slippery slope to where for the laws you can actually, it really hurt people. That's that's that's the terrifying thing when you attach. I'm actually really terrified of the power of the lull. It's the fun thing somehow becomes a slippery slope. I haven't quite understood the dynamics of that. But even in myself, if you just have fun with the thing, you lose track of the ethical grounding of the thing and so like it feels like hacking for fun can just turn it like literally lead to nuclear war. Like literally destabilize [/INST] yada yada yada nuclear war. I could see it. Yeah. </s><s>[INST] So I've been more careful with the law. Uh I'm very yeah, I've been more careful about that and I wonder about that because internet speak somehow ethics can be put aside through the slippery slope of language. I don't know, everything becomes a joke. If everything is a joke then everything is allowed and everything is allowed then you don't have a sense of what is right and wrong, you lose sense of what is right and wrong, [/INST] you still have victims. I mean you're laughing at someone, someone's the butt of this joke. You know, whether it's major corporations or the individuals. I mean some of the stuff they did was just you know, releasing people's P. Ii. And their personal identifying information and stuff like that. I mean is it a big deal? I don't know, maybe maybe not. But you know, if you could choose to not have your information put out there probably wouldn't. </s><s>[INST] We do have a sense of what anonymous is today. Is has it ever been one stable organization or is it a collection of hackers that kind of emerge for particular tasks for particular uh like hacktivism tasks and that kind of stuff. [/INST] It's a collection of people that has some hackers in it. There's not a lot of big hackers in it. I mean there's something that have come bouncing and bounce out even back then there's probably just as many reporters in it. People the media in it with the hackers at the time just trying to get the inside scoop on things, you know, some giving the inside scoop, you know, we arrested a reporter that gave over the user name and password to his newspaper and uh you know just so he could break the story. He trusted </s><s>[INST] him. [/INST] Speaking </s><s>[INST] of trust reporters, boy, there's good ones. There's good ones. There [/INST] are, </s><s>[INST] but boy do I have a complicated relationship with them. [/INST] How many stories about you are completely true. </s><s>[INST] You can just make stuff up on the internet. And one of the things that, I mean there's so many fascinating psychological sociological elements of the internet to me, one of them is that you can say that, um, lex is a lizard, right? And if it's not funny, so lizard is kind of funny. What should we say? Um, lex has admitted to being an agent of the FBI. Okay. You can just say that right? And then the response that the internet would be like, oh, is that true? I didn't realize that they won't go like provide evidence. Please write. They'll just say like, oh, that's weird. I didn't, I kind of thought he might be kind of weird and then it piles on, it's like, hey, hey, hey guys, like here's a random dude on the internet just said a random thing, you can't just like pile up and then [/INST] Johnny 69 69 is now a source that says, and </s><s>[INST] then like the thing is, I'm a tiny guy, but when it grows, uh, if you're like have a big platform, I feel like newspapers will pick that up and then they'll like start to build on a story and you never know where that story really started. It's so cool. I mean to me actually, honestly, it's kind of cool that there's a viral nature of the internet that can just fabricate truth completely. I think we have to accept that new reality and try to deal with it somehow. You can't just like complain that johnny 69 can start a random thing. But I I think in the best possible world, it is the role of the journalist to be the adult in the room and put a stop to it versus look for the sexiest story. So that there could be Clickbait that can generate money. Journalism should be about sort of slowing things down, thinking deeply through what is true or not and showing that to the world. I think there's a lot of hunger for that. And I think that would actually get the most clicks in the end. [/INST] I mean it's that same pressure I think we're talking about with the FBI and with the tech companies about controllers. I mean the editors have to please and get those clicks. I mean, they're measured by those clicks. So, you know, I'm sure the journalists, the true journalists, the good ones out there want that, but they want to stay employed to </s> |
<s>[INST] the winter. [/INST] Well, yeah, but I mean, look, the other thing is that we have to start preparing, I mean, we have to start thinking about this, we've got a wartime economy situation, That's where we are. We've got the home front to think about as well. Putin has declared war on us. He did that on September 30 and he's done it at other points as well. We've just not paid attention, but he pretty much pretty explicit in September 30, I mean go back and watch that speech and you know, he is gambling that, you know, people will go back, you know to basically taking Russian gas and oil, but it's not gonna be that simple as well. And do people and then, you know, the question has to be, do we really kind of think he's gonna play fair after that when he's kind of also shown that he can leverage that </s><s>[INST] it's such a complicated world [/INST] is complicated. It's very </s><s>[INST] competent. It's never I mean, this it feels like things are heating up, like and china is very quiet right now [/INST] because they're watching what happens. I mean, for President xi, he's trying to consolidate his power even further after the party congress, but he doesn't want to look like he made a mistake by backing Putin. I mean, he thought Putin was also going to be in our Ukraine would probably be open for massive chinese investment. China was the largest investor in Ukraine before the war. Largest single investor in the EU is bigger. Of course. How </s><s>[INST] do you hope the war ends in Ukraine? [/INST] Well, I mean, I do hope and you know, with a ceasefire negotiated solution, but it has to be with Russia compromising on something and that's not where we are right now. </s><s>[INST] Do you think both sides might be willing to compromise [/INST] most wars always end in that way? I mean, nobody is ever happy, </s><s>[INST] but they don't seem to either side, like legitimately doesn't want to compromise right [/INST] now? Because the thing is that for Ukraine right now, anything is a compromise at his expense. Right, vast devastation, unbelievable casualty rates biggest refugee crisis since World War Two, Russia's just said, sorry, this is our territory is not just Crimea, I think there could have been a negotiation over there, but you know, Donetsk and Luhansk, we've got all kinds of formulas, we've had all the way through history of, you know, putting things under a kind of guardianship Receivership of territory, the United Nations, all kinds of different ways of formulating that we could have easily been creative. But Russia is basically saying, sorry, we've taken this and any other negotiations, just you recognizing this for us not doing more destruction, that is not the basis for negotiation. And, you know, having, you know, kind of people come and just sort of laying those terms down is not a starting position. I think Russia is also, you know, in a dilemma of its own making now, because Putin has made it very difficult to compromise just by everything that he said. Now, if Ukraine, they've already won a great moral political and military victory, it's just hard to see it right at the particular moment. They've done what the Finns did in the winter war, which the Finns were devastated by the winter war as well, but they pushed them back. Now, the Finns lost a lot of territory, those cara leah and, you know, huge swathes of territory, but they've got to be Finland and now they're, you know, johnny NATO. But they've been part of the eu the question is how to get Ukraine to be Ukraine in a success. But you know, is and that's the challenge again, they've already won psychologically, politically militarily, because Putin doesn't succeeded in what he wanted to do, but he has succeeded in completely and utterly devastating them. And this is the kind of the old muscovite, old Russian imperial old servant mentality, you know, going all the way back to when the muscovites with a bagman for the, you know, the hold the for the Mongols, it was destruction. You know, you don't play with us will destroy you. People talk about it as mafia, but it's old. You know, all you have to go down and see Tarkovsky andre rieu. I mean, I remember, you know, seeing that film when I was first as a student in Moscow and just being whoa, this is so brutal. I mean, this is just unremittingly brutal. Because the whole point is that you show people who's the boss, the destruction is the point of things as well, because, you know, you are emphasizing your domination and that's what Putin is doing right now is saying, okay, you want to go in a different direction, so be it. But I'm gonna make you suffer. Remember when Khordokovsky got out of the penal colony, when Putin let him out eventually, he said he's suffered enough, But he suffered for 10, 11 years, I don't think Putin feels that Ukraine has suffered enough at this point, we have suffered. </s><s>[INST] There's a part of this invasion. That's punishment for something. [/INST] It's medieval, I mean, look, we're all capable of the same things, right? There was all that destruction and that's what assad was like in Syria, like his father, you destroy because you teach him a lesson. And look, Britain did that in the colonial era. I mean, all the history of british colonialism is exactly the same. I mean, all the mau mau, you know, in kenya, you know, up until recent times, brutality teaching people, you know, you're teaching them a lesson. You have to suffer the US today. I mean, we did it with the native americans, you know, we did it all over the place, you know, as well. This is kind of what big, you know, States do at different points in history. It's just that, you know, Russia has not moved on from that and we've learned some lessons later. I hope, you know, we're fully internalized them of things that we've done, you know, kind of the past. The United States ideally are trying to do better and most of europe is trying to do better as well. Think about France and Algeria, you know, again, you know, we can see this in many different settings, but I think, you know, for Putin right now, he hasn't taught all of us sufficient lesson. </s> |
<s>[INST] Scientology are all just Scientology. Yeah, so there's definition of what's bad for you, and it's probably in the beginning is bad for you. Someone's just, sounds like a template of a toxic relationship. You know, there's a bunch of stuff in this world that's just not good for you. So the authoritarian says like, I'm just protecting you by blocking you off from those negative things. And they're probably, they are probably negative things. But then this like, this freedom starts closing in to where you can't no longer speak freely, think freely, act freely. And there's some, I mean, that's why sort of power corrupts and absolute power corrupts absolutely. That person doing the controlling actually starts getting that like dope immune rush of the controller that's exciting. And it's a vicious negative cycle. So you start out as a, it starts out good because you're trying to do good for the person. But then it somehow goes to shit. Yeah. So what, what, what are the aspects that just, that are often controlled about, about a person [/INST] who's in mythology, especially C org? Well information control, access to the internet, access to any information. Critical of Scientology. Is some internet access allowed? Public Scientologist has no restrictions to their access to the Internet. They're just not allowed to read anything critical of Scientology. Oh </s><s>[INST] okay, so there's self, they're supposed to self control what they read or not? And And what's the explanation? Is it always assumed that anything critical of Scientology is a lie? [/INST] Mm-hmm, mm-hmm. They really pushed this thing that unless you've been in a Scientology organization yourself, or unless you've actually been a Scientologist, you couldn't possibly know the truth about Scientology. If you're only getting information from people who aren't members or former members, then you couldn't possibly be getting the correct information. Now, they don't realize the math there doesn't make sense. if you can find out the true information by becoming and being a Scientologist, then that means you can get the correct information from a former Scientologist because they traveled that path and they got the correct information. So they still create this, or they try to create this unfalsifiable loop where unless you are personally doing it, you don't have correct information. And you go, well, what about the people who did personally do it. Got the correct information, left, and are now sharing that with others? Well, no, those are lies. Well, okay, so just anything you don't like is a lie then. You go, yeah, pretty much. </s><s>[INST] That's kinda how it works. So what about the control of negative information on the internet? What, like the actual operations I've, you know, preparing, I should admit, I don't know too much about Scientology. I was doing a bunch of reading and the Wikipedia page on Scientology, interestingly enough, is not that negative about Scientology. So like, it made me ask, you know, you have to be a little bit careful how you consume stuff from Wikipedia. You have to consider, because money can buy things there. There's certain special interests and so on. But like, it made me wonder, like, like with a lot of controversial topics, what is true? And where do I look, where do I go for truth? So like, how much sort of deliberate action is there [/INST] to control what is true on the internet by Scientology? Well, these days, they've pretty much, I think, thrown in the towel, but- They have. Scientology middle management was editing Wikipedia so often from IP addresses that were traced back to the Scientology buildings that Wikipedia locked them out from any IP addresses associated with Scientology from being able to edit it. It's like, the Scientology was so infatuated with trying to control the information. And in the early days of the internet, they had a certain degree of success with that. </s><s>[INST] They just, it's just hopeless these days. It's the scale, the scale is not there, but actually I'm very surprised how bot farms, how effective that can be at a very small scale. If you just pay a hundred people, just spread narratives, but the bad stuff, the reason that's effective is you can kind of create conspiracy theories that create chaos and nobody knows what is true that bot farms can do, but actually really nicely control a narrative, it's hard. So to create chaos, it's easier to do, to basically say like, do PR control is very hard. Yeah, so especially on the internet, especially when the critical eye is there, the internet can smell bullshit, which is one of the really, really powerful thing [/INST] about the internet. And I gotta tell you, it's one of the reasons I do my YouTube channel. It's one of the reasons I decided to upload every day. I've uploaded every day for the last six months. I just wanted there to be a nonstop flow of information of any kind and any variety, as long as it's fair and balanced, intelligent, interesting, that Scientologists who stumble upon the internet will go, oh, look, someone's talking about my thing. Let's see what they got going on. And I know this guy, the fact that Scientology crushes so much information, before YouTube, before like, I have the only like big Scientology channel and that only got big in the last six months. Okay, so before that, there were channels, there was things, but it's almost like it took a lot of, like people felt like it took a lot of bravery and courage to like say something on the internet about Scientology. And so people would pop up and there weren't very many voices and I was like, I want this to be prolific. I want to be prolific. I want to have 30 or 40 other channels being prolific so that Scientology couldn't possibly successfully control the narrative about it. Have you been personally attacked? Aaron Smith Levin.com is a website created by the Church of Scientology. Have you seen it? </s> |
<s>[INST] Yeah. What about another magical leap with the early days of Watsonwood beating the Jeopardy challenge? What was your experience like with Watson? What's your vision for Watson in general? [/INST] Yeah. And it was really inspired by first chess, right? And Casbroff, and then you come forward in time. And I think what Watson did, because you used a really important word, AI had kind of waxed and waned in these winners, right? In and out, in and out. Popular or not, more money, less money, in and out. Confidence, no confidence. And so I think that was one of the first times it brought to the forefront of people like, whoa, like it humanized it. Because here it is playing against these two gentlemen, and you did lose it first, and then finally won at the end of the day. And what it was doing is making you say, hey, natural language, it's actually understanding natural language. It's one of the first demonstrations of natural language support and a bit of reasoning over lots of data, right? And so that it could have access to a lot of things, come up with a conclusion on it. And to me, that was a really big moment. And I do think it brought to the conscious of the public, and in good ways and bad, because it probably set expectations very high of like, whoa, what this could be. And I still do believe that it has got the ability to change and help us, man, make better decisions, that so many decisions are not optimal in this world, even medical decisions. And it's right or wrong, what took us down a path of healthcare first with our AI. And we took many pivots, and I think there's a really valuable lesson in what we learned. One is that I actually don't think the challenges are the technology. Yes, those are challenges, but the challenges are the people challenges around this, right? So do people trust it? How will they use it? I mean, I saw that straight up with the doctors and like, meaning they're so busy in the way they've been taught to do something. Do they really have time to learn another way? I saw it was a mistake when you put it on top of processes, it didn't change, kind of like paving a cow path. Didn't work. I mean, it was all human change management around it that were really its biggest challenges. And another valuable lesson we picked back to usage, you think of IBM as moonshops, we picked really hard problems to start with. I think you see a lot of technology now starts with really simple problems. And by that, it probably starts to build trust because I start little, it's like, Oh, I'm not ready to, you know, outsource my diagnosis to you, but I'll get some information here about a test question. So very different thinking. So a lot of things to learn, we were making a market at the time. And when you make a market, you know, choice of problem you work on gets to be very important when you're catching up, well, then it's a scale game. So very different thing. And so, but Watson proved, I think, I mean, I hope I'm not being too, I think Watson brought AI back out of winter for the world. And that since then, there's just been, you know, one company after another and innovations and people working on it. And I have no regrets of anything that we did, we learned so much, and we probably rebuilt it many times over, it made it more modular. And today, to IBM, a Watson is more about AI inside of a lot of things, if you think of it that way, which is more like an ingredient versus it's a thing in and of itself. And I think that's how it will bring its real value, you know, more as an ingredient, and it's so badly needed. And even back then, the issue was so much data, like, What do you ever do it? You can't get through it, you can't use it for anything, you know this well, it's your profession. So we have to have it. So that's a </s><s>[INST] that's going to propel it forward. So as part of the suite of tools that you use when you go to [/INST] to enterprise and you try to solve different problems. AI for security, AI in automated operations, AI in your robotics, AI on your factory. You know what I mean? It's all part of, and I think, and that's why even to this day, I mean, thousands and thousands of clients of IBM still have the Watson components that it's the AI being used. So it became a platform is how I would say it, right? And an ingredient that went inside, and consultants, like you said, had to learn. I had, they had to learn, don't just put it on something. You got to rethink how that thing should work because with the AI, it could work entirely differently. And so I also felt it could open up and still will open up jobs to a lot of people because more like an assistant, and it could help me be qualified to do something. And we even years ago saw this with the French banks, very unionized, but that idea that you could, in this case, the unions voted for it because it felt people did a better job. And so, and that's just part about being really dedicated </s><s>[INST] to help it help humanity, not destroy it. Speaking of which, a funny side note, so Kubrick's, the 2001 Space Odyssey, what do you think about, you know, [/INST] the fact that HAL 9000 was named after IBM, 2001 Space Odyssey? I really don't think it was. I know there's, I really don't. It could be more fake news. It's more fake news. I have done, I've like researched this, tried to find any evidence and people have talked to you. It wasn't really one letter, it was one letter off. </s> |
<s>[INST] It's perhaps counterintuitive, it never was to me, but it's counterintuitive that to uh to start on the journey of really sort of mastering digits or whatever or improving is you have to relax and that seems to be a very constitutive lesson I learned that early on with, that was thanks to the Russian system, I played piano and like music basically actually this is true for basically any sport that includes the human body is like relaxing is the way you, you start learning stuff you have to learn, you have to literally and most people don't seem to understand this is like you have to learn what it means for the human body to relax. Like I guess you have to have enough knowledge of all the muscles involved to know what it means to relax those muscles. So for piano, you have to understand what it means to relax your wrists and your fingers in order to learn how to move them. Like if theres tenseness in the fingers, you're not going to like you have to learn how to try hard while relaxed. The I guess the beginner, if you don't internalize this lesson, will try hard by uh tensing up hard and like trying hard, tensing up more as opposed to relaxing more. And that lesson cannot be conveyed through words I guess. Uh I've had the great fortune of having dictatorial teachers as they do in Russia for piano and so on, where you get like hit if, if you don't learn to relax, which is a constitutive notion, but it works. [/INST] Yeah, this brings me to one of my favorite pieces of coaching advice that I will tell white belts while they're struggling on the mat? I'll tell them to relax harder. </s><s>[INST] Okay, that's it's beautiful for somebody who studied war who participate in war. What do you think is the best martial arts for let's call it self defense. For hand to hand combat outside the constraints of sport. [/INST] So it's not one answer. The answer to me is you get to boxing wrestling more tie judo sambo and on down the list, um I definitely start with jiu jitsu, the reason I start with jujitsu is because in a self defense situation, if you are a big monster human and you want to fight me and you square off with me, I guess what I'm gonna do. Run away. Mhm. Because I don't want I don't want to get involved. Even if I see a skinny little lex out on the street and you start yelling at me and saying you want to fight me, I don't want to fight. You don't it doesn't matter. I don't care if I can beat you're not what if you stab me, what if you sue me after I get done throwing you onto the concrete, there's a million bad things can happen and almost nothing good. So for self defense, my first self defense is my feet to get away from you. And if you square off to punch me, I can run away from you. If you square off to kick me, I can run away from you. If you push me, I can run away from you. So great, I don't need to know how to box to run away from you. Where this all changes is when you grabbed me and now I don't have the option run away anymore. Now I actually have to know how to get away from your grip and that's where you just who comes into play. So especially if you get me on the ground, if you if you grab me and get me on the ground now I need to know how to get you off of me and get up and get away from you so I can run away. So that's why I say start with you get to and and from there Boxing wrestling, judo, sambo mai tai. </s><s>[INST] Yeah, there's uh there's in the standing position. I mean I'm a judo person as well and uh judo is very limited in their understanding of the full grappling spectrum, even though they do all the things on the ground as well. But uh it's so focused on the feet. But nevertheless it's important to understand the thing that judo has as a sport. It's good to practice. That just doesn't is not just the greater the skill of grappling on the feet, but the skill of explosive aggression that sometimes you get is more about in terms of tactics is more about patients. It depends how you practice it. But because so much is about control and technique that sometimes you don't get to practice, like aggression, explosive aggression and judo is so much about aggression implemented in such a way that the demonstration of power is effortless, right? That's the beauty of the Jessa. [/INST] Yeah, and same thing with wrestling, wrestling also has a high level of intensity and aggression as well. So that's where that's where I agree judo and wrestling. Absolutely awesome. Get some and striking boxing muay thai you know like the you should train all these things. </s><s>[INST] Are there books and movies in your life long ago recently that had a big impact on you. [/INST] Yeah, the main one is about Face which is sitting right here, there you go. This is written by Colonel David Hackworth. That's the book that really had a massive impact on me from a leadership perspective and I ended up, I talked about it enough that it started kind of coming back and started selling well and they contacted me and I wrote the forward for it. So that book had a huge impact on me and I still, when I read it I still get lessons out of it just about every time </s><s>[INST] since the Vietnam war [/INST] and Korea and Korea and he got in towards the end of right at the end of World War Two. So he was kind of raised by the soldiers that fought in World War Two and then he went to Korea and then he went to Vietnam. </s> |
<s>[INST] things, right? Some would [/INST] be aggressive, some would be, you know, curious others would be very stoical and philosophical because you know, maybe there are a million years older than us, but it's not. It shouldn't be like, I mean 11 alien civilization might be like that, communicating thoughts and others, but I don't see why, you know, potentially the hundreds there should be would be uniform in this way. It </s><s>[INST] could be a violent dictatorship that the people, the alien civilizations that become successful become gain the ability to be destructive in order of magnitude more destructive. Um but of course the sad thought, wow, either humans are very special. We took a lot of leaps that arrived at what it means to be human. Um there's a question there which was the hardest, which was the most special. But also if others have reached this level and maybe many others have reached this level. The great filter that prevented them from going farther to becoming a multi planetary species are reaching out into the stars and those are really important questions for us whether um whether there's other alien civilizations that there are not, this is very useful for us to think about. If we destroy ourselves, how will we do it? And how easy is it to do? [/INST] Yeah. Well, you know, these are big questions and I've thought about this a lot. But the interesting thing is that if we're if we're alone, that's somewhat comforting from the great filter perspective, because it probably means the great filters were passed us. I'm pretty sure they are. So that by going back to your origin of life question there are some incredible things that no one knows how happened. Like obviously the first life form from chemical soup. That seems pretty hard. But I would guess the multicellular, I wouldn't be that surprised if we saw single single cell sort of life forms elsewhere, bacteria type things. But multicellular life seems incredibly hard. That step of, you know, capturing mitochondria and then sort of using that as part of yourself, You know, when you've just eaten it, </s><s>[INST] would you say that's the biggest, the most like if you had to choose one sort of hitchhiker's Guide to the galaxy one sentence summary of like, oh those clever creatures did this. That would be the multiple. I [/INST] think that's probably the one that that's the biggest, I mean there's a great book called the 10 grand and great inventions of evolution by Nick Lane and he speculates on 10, 10 of these, you know, what could be great filters. Um, I think that's one, I think the the advent of of intelligence and and conscious intelligence and in order to, you know, to us to be able to do science and things like that is huge as well. I mean, there's only evolved once, as far as, you know, in in in earth history. So that would be a later candidate. But there's certainly for the early candidates. I think multicellular life forms is huge </s><s>[INST] by the way. What it's interesting to ask you if you can hypothesize about what is the origin of intelligence? Is it that we started [/INST] cooking </s><s>[INST] meat over fire. Is it that we somehow figured out that we could be very powerful when we started collaborating. So cooperation between um our ancestors so that we can overthrow the alpha male. What is it? Richard, I talked to Richard Random who thinks we're all just beta males who figured out how to collaborate to defeat the one, the dictator, the authoritarian alpha male um, that control the tribe. Um, is there another explanation? Was there 2001 type of monolith that came down to earth? [/INST] Well, I think, I think all of those things you suggested a good candidates fire and and and and cooking, Right? So that's clearly important for energy. You know, energy efficiency cooking our meat and then and then being able to to be more efficient about eating it and getting a consuming the energy. Um, I think that's huge and then utilizing fire and tools, I think you're right about the tribal cooperation aspects and probably language is part of that because probably that's what allowed us to out compete neanderthals and, and and perhaps less cooperative species. So, um, so that may be the case tool making spears axes. I think that let us, I mean, I think it's pretty clear now that humans were responsible for a lot of the extinctions of megafauna, especially in in in the Americas when humans arrived. So you can imagine once you discover tool usage, how powerful that would have been and how scary for animals. So I think all of those could have been explanations for it. You know, the interesting thing is that it's a bit like general intelligence too is it's very costly to begin with, to have a brain and especially a general purpose brain rather than a special purpose one because the amount of energy our brains use, I think it's like 20% of the body's energy and it's it's massive. And then you're thinking chest, One of the funny things that we used to say is it's as much as A racing driver uses for a whole Formula one race, just playing a game of, you know, serious, high level chess, which you know, you wouldn't think just sitting there because the brain is using so much energy. So in order for an animal, an organism to justify that there has to be a huge payoff? And the problem with with half a brain or half intelligence, satanic of you know of like a monkey brain. It's it's not clear you can justify that evolutionary until you get to the human level brain. And so but how do you how do you do that jump? It's very difficult, which is why I think it's only been done once from the sort of specialized brains that you see in animals to this sort of general purpose chewing powerful brains that humans have. Um And which allows us to invent the modern modern world. Um And you know, it takes a lot to to cross that barrier. And I think we've seen the same with Ai systems, which is that maybe until very recently it's always been easier to craft a specific solution to a problem like chess than it has been to build a general learning system that can potentially do many things because initially that system will be way worse than less efficient than the specialist system. </s> |
<s>[INST] And the weather affects the data quality. [/INST] Yes, absolutely, because these were really faint stars. And so we were really lucky by making a very tight strategy of getting the absolute bare minimum for all the stars. So we could at least take a very crude look. Is it a yay or is it a nay? We couldn't even say yes or no, just to get an idea because we needed to know. Why was that important? Because we could only observe this system again nine months later. </s><s>[INST] So there's always a window of observation. [/INST] Yes, it was setting. This was our chance and it was going away with the clouds, you know? That was super high stakes. But we just made it. Really, it was almost impossible. And the thing is, this is such a serendipitous moment, inner serendipitous moment. The enhancement of these heavy elements was so strong that even in this really crappy data, </s><s>[INST] we could still see the enhancement. That was super high stakes. [/INST] The absorption was so strong it stuck out of the noise. If that enhancement wouldn't have been as strong, we would not have been able to say anything because we wouldn't have been able to tell. But because it was so extreme, it lent us a hand, despite the weather and all, to say like, yes, this is it. So that was quite the night. </s><s>[INST] Luck. I mean, a lot of this is just luck. So that was the first our process galaxy discovered. [/INST] Yeah. Yes. I didn't sleep all that much then. </s><s>[INST] Do you have hope? Are you excited about James Webb Space Telescope and other telescopes in the future [/INST] that increase the resolution and the precision of what can be detected out there? Absolutely. Data with tea is fantastic already. I am not planning to use it personally, although I think I'm on one or two observing proposals, actually, because similar to what we already spoke about, we're interested in the same thing. We're just kind of looking at a different sides of the fence, right? I have my old surviving stars and I concoct these little stories about what the earliest galaxies may have looked like, what the objects were that contributed energy and elements and all these things. My JWST colleagues, they try to detect some of these earliest photons from these earlier systems to look at the energetics and other things. What was there? How many these kinds of things, right? So together, we're trying to explore this first billion years, but we do it in very complementary ways. And so I'm very excited to see what they can come up with and how that helps me to </s><s>[INST] inform my stories better and more comprehensively. What do you think is the future of the field of stellar archaeology? How much can we... Maybe what are the limits of our understanding of this first billion years of our universe? [/INST] Well, obviously, lots of limitations in the sense that I always say, I have a metal pore star for any of your questions because there are so many different kinds out there. And we still find new patterns sometimes, right? And there needs to be an explanation. The question is, is it ultimately just one quirky star or is it two or is it three, right? Is it a sample, is it a population? So we haven't concluded that kind of work yet. </s><s>[INST] So every metal pore star is a kind of data point that you can use to [/INST] improve the quality of your model of how is the evolution of the early universe. Yes, and I would say we've made huge progress over the last 20 years. When I joined that field, it was in its infancy. And there was this serendipitous discovery of that first second generation star. And we have filled in the canvas a great deal since then. And this is what I have greatly enjoyed about doing so because there was so much discovery potential. And it's been dying down a little bit because of all the progress. It's on the up and coming again because there are so many large spectroscopic surveys in the works now that will just provide a different level of data that we haven't had before. I'm sort of of these older generation. I have only very few colleagues. I work in small teams and I observe every single star myself. Whatever I can, I do myself. I don't generally take other people's data, at least not certainly not in the end stage. And I'm not a big data kind of person, although we all headed that that way. I certainly use data from the Gaia astrometric satellite for the kinematics, for example. But that's personally a new thing for me to use sort of big sky surveys. That are available, so it's still very sort of hand-grown field where we do our individual observations. I have enjoyed that a lot, but that's about to change. </s><s>[INST] So one star at a time. Yes. I mean, there's power to that, to build up intuition of the early universe by looking one star at a time. [/INST] Yes, absolutely. And this is how you can really drill down on the questions that you have, right? Because you control what data you get. Otherwise, you have the data that you have, right? You get what you get and you don't get upset. I don't like that. I'm a little bit snobby. I really like to formulate my questions, go to the telescope, and then come with me. I will try to get it. </s><s>[INST] And also develop the intuition of where the data can be relied upon and where it can't and all the different quirks of the data and all that kind of stuff. Sometimes a lot is lost in the aggregation of the noisy data. [/INST] Yeah, yeah, yeah, yeah. Yeah, yeah, yeah. And that's always the danger if you have someone else's data that you just don't really understand the limitations and completeness things, how certain things were set up, and you get out what you put in. So I'm really particular about that, and it certainly paid off for me. That's one of the main notions that I try to teach in my classes and to my students, that you need to be able to formulate your question really well because otherwise you're going to get an answer to a different question, but you won't notice that the goal pulse has shifted in the meantime, right? So your interpretation can only be as good as the question. If you need to change your question, that's cool. Do it, but then it needs to pair up with your interpretation again. And so knowing, really being in the know about every step of what happens, that leads to quality results, I think. That's why I have sometimes little trouble with sort of big data and statistical analysis. Yes, on average, that's true. I'm not debating that, but I'm the kind of person I like to look at the outliers, so not the bulk, but the special ones. And they just need to be treated in a different way, and there needs to be an acknowledgement of that, different ways for different things. </s> |
<s>[INST] This is the lex Friedman podcast and here is my conversation with rick Ruben. Are you nervous? [/INST] I'm not shaky but I would say I feel uneasy and I feel like the sooner we start talking, the more relaxed we'll get. </s><s>[INST] Yeah, well maybe we should sit in this moment and enjoy the nervousness of it. Let me start with? Nietzsche? He said without music. Life would be a mistake. What do you think he means by that? Let's talk some philosophy, Let's try to analyze free jerk nature from a century ago. [/INST] It seems like music has the ability to bring us so much uh depth in our soul. That's hard to access any other way. And without it, there would be a loss beyond beyond the pleasure of it. Um, it feels like it's a window into something else, </s><s>[INST] something that no other medium can express quite the same way. [/INST] I would say, not as automatically, something about music can do it automatically. Maybe poetry or maybe certain abstract forms can get us there. But there's something about music that really can get us there quickly, </s><s>[INST] but it's also the time to place the history. There's something about like a lot of my family's still in Philly. There's something about driving through Jersey and listening to Bruce Springsteen and then you just, I'll get like emotional, like listening to like I'm on fire that like one of one of my favorite Bruce Springsteen songs, There's a, there's a, there's a haunting kinda strumming to it. It's not a strumming. It's, it's actually picked, has a country feel to it almost like a johnny cash feel actually. And it, I don't know, it makes me feel so for people who don't know, I'm on fire. That song is I guess love song to a woman that you can't have because she's married or she's with somebody else, which I guess is quite a lot of love songs, but there's something about the haunting nature of the guitar and then it has to be driving through Jersey. And I feel like Everyone has fallen in love with a jersey girl at one point in their life. I don't know if that's true for her, but I I feel like that I haven't either, but I just feel like that I feel there's something about Bruce Springsteen is like, yeah, I've been there. I you know, um and that just takes you to a place of emotion that you just, that that captures love that captures longing. That captures the the heartbreak of just the way time flows in life and in fact is finite and just all of that in a single single simple song. The guy. What else can capture that? [/INST] Yeah, I don't know. But but it's true that there's a connection both between time and place and music and I, certain music growing up on the east Coast didn't really resonate with me until I spent time on the West Coast Eagles being an example when I lived in new york, the Eagles didn't really speak to me Zz top didn't really speak to me. And then when I started spending time in California and driving through laurel canyon all of a sudden the music of the Eagles felt appropriate somehow. And I started listening to more </s><s>[INST] God, It's so not until you went out west. Can you understand the sounds of the west. So it's really like new york has a sound. What are the places of the sound in the United States? [/INST] I think every place does. And that said sometimes we can get an experience through music of a place like we can resonate with the music and not understand why. And then maybe when we go to the place where it was created, it's almost like we have a knowingness of that place. It's not a strange place anymore. </s><s>[INST] Yeah, Stevie Ray Vaughan with blues and tech texas, blues. You can just listen to texas flood and just again, there's just like a woman, you're missing a broken heart and somehow that connects to to the place the Eagles. What song the Eagles connects with? You were talking about like take it easy or are we talking more like hotel California? [/INST] I'm thinking, take it easy. But but both are great. </s><s>[INST] Yeah. There's certain songs when I started learning guitar when I was young, that's like, I would like to be the kind of person that not only knows how to play this song, but understands this song and like Have that song be something I played 20 years ago and I've lived with that song for a while. Like hotel California is an example. Obviously there's the solo, but there's also the soulfulness of the lyrics, which I still don't understand and it could be about anything. And as you get older, I feel like the meaning of the song could be anything. [/INST] Yeah, I think that's true. I think that's the beauty of them. I think when the person wrote them they may have had one interpretation. Um, but it's not contingent on us getting that interpretation to like it or resonate with it or feel it in some ways the best, the best art is open enough where the artist gets to have their experience when they make it and then the audience gets to have their experience when they listen and they don't have to be the same </s> |
<s>[INST] testing. There's so many more that I can point to. So those are kind of decentralized innovations. [/INST] And that's what FDA should stand for. FDA can stand for it. Or something like FDA, right? Ah, so let's talk about that, right? Something like FDA. So this is very important. In general, </s><s>[INST] the way I try to think about things is V1, V2, V3, as we've talked about a few times. So FDA V. [/INST] Well, right. So what was before FDA, right? So there was both good and bad before FDA, because people don't necessarily have the right model of the past. Okay. So, you know, if you ask people what was there before the FDA, they'll say, and by the FDA itself, amidst the, right, their pronouns are just FDA, FDA. Okay. So, but basically, </s><s>[INST] Why is that important? [/INST] It's just something where... </s><s>[INST] Why is that either humorous or interesting to you? [/INST] They have a sort of in-group lingo, where when you are kind of talking about them, the way that they talk about themselves, it is something that kind of piques interest. It's kind of like, you know, in LA people say the 101 or the, you know, right, whereas in Northern California, they'll say 101, or people from Nevada will say Nevada, right? It just instantly marks you as like insider or outsider, okay, in terms of how the language works, right? And that's... Go ahead. </s><s>[INST] That means just makes me sad, because that lingo is part of the mechanism which creates the silo, the bubble of particular thoughts. And that ultimately deviates from the truth, [/INST] because you're not open to new ideas. I think it's actually like, you know, in Glorious Basterds, there's a scene in the bar. Do you know what I'm talking about? </s><s>[INST] No, but it's good. You can't just censor you. This is like a Wikipedia podcast, like Wikipedia. You can't cite Quentin Tarantino films. No. Okay, okay, okay. Sorry, be back there. [/INST] So, basically, like English start going like one, two, three, four, five. And I believe it's like the Germans start with like the thumb, something that you never know, right? I may be misremembering there, but I think that's right, okay? So, so that's okay. FDA has got the lingo. All right, all right. Right, so FDA has got the lingo. So, coming back up, basically, just talk about FDA, and then come back to your question on the D platform. </s><s>[INST] So, what was V0 FDA? What's V1? What does the future look like? [/INST] V1. V1 was, quote, patent medicines, okay? That's what people say. But V1 was also Banting and Best, okay? Banting and Best, they won the Nobel Prize in the early 1920s, right? Why? They came up with the idea for insulin supplementation to treat diabetes. And they came up with a concept. They experimented on dogs. They did self experimentation. They had healthy volunteers. They experimented with the formulation as well, right? Because just like you'd have like a web app and a mobile app, maybe a command line app, you could have, you know, drug that's administered orally or via injection or cream or, you know, there's different formulations, right? Dosage, all that stuff, they could just like iterate on, okay, with willing doctor, willing patient. These, you know, these, these folks who were affected, just sprang out of bed, the insulin supplementation was working for them. And within a couple of years, they had won the Nobel Prize and Eli Lilly had scaled production for the entire North American continent, okay? So, that was a time when pharma moved at the speed of software, when it was willing buyer, willing seller, okay? Because the past is demonized as something that our glorious regulatory agency is protecting us from. Okay. But there's so many ways in which what's really protecting you from is being healthy. Okay. As you know, I mean, there's, there's a zillion examples of this, I won't be able to recapitulate all of them just in this podcast. But if you look at a post that I've got, it's called regulation, disruption and the future technologies of 2013 Coursera PDF, okay, this lecture, which I'll kind of link it here so you can put in the show notes if you want. This goes through, like a dozen different examples of crazy things the FDA did from the kind of stuff that was dramatized in Dallas virus club where they were preventing people from getting AIDS drugs to their various attacks on, quote, raw milk, where they were basically saying, here's a quote from FDA filing in 2010. There's no generalized right to bodily and physical health. There's no right to consume or feed children any particular food. There's no fundamental right to freedom of contract. They basically feel like they own you. You're not allowed to make your own decisions about your food. There's no generalized right to bodily and physical health direct quote from their like written kind of thing. Okay. The general frame is usually that FDA says it's protecting you from the big bad company. But really what it's doing is protecting it's preventing you from opting out. Okay. Now, with that said, and this is where I'm talking about v3. As critical as I am of FDA or the Fed, for that matter, I also actually recognize that, like the Ron Paul type thing of end the Fed is actually not practical and the Fed will just be laughed at. What Bitcoin did was a much, much, much more difficult task of building something better than the Fed. That's really difficult to do. Because the Fed and the FDA, they're like the hub of the current system. People rely on them for lots of different things. Okay. And you're going to need a better version of them. And how would you actually build something like that? So with the Fed and with SEC and the entire the banks and whatnot, crypto has a pretty good set of answers for these things. And over time, all the countries that are not, or all the groups that are not the US establishment or the CCP will find more and more to their liking in the crypto economy. So that part I think is going. Okay, we can talk about that. What does that look like for biomedicine? Well, first, what does exit the FDA look like? Right? So there actually are a bunch of exits from the FDA already, which is things like right to try laws, okay, CLIA labs and laboratory developed tests, compounding pharmacies, off little prescription by doctors, and countries that aren't fully harmonized with FDA. For example, you know, Kobe Kobe Bryant before he passed away, went and did stem cell treatments in Germany. Okay, stem cells have been pushed out, you know, I think, in part by the Bush administration by other things. So those are different kinds of exits, right to try basically means, you know, at the state level, you can just try the drug, okay, CLIA labs and ldts. That means that's a path where you don't have to go through FDA to get a new device approved, you can just run it in a lab, okay, compounding pharmacies, these were under attack, I'm not sure actually where the current, you know, statute is on this, but this is the idea that a pharmacist has some discretion and how they, you know, prepare mixtures of drugs, um, off little prescription by MDs. So MDs have enough like weight in the system that they can kind of push back on FDA and off little prescription is the concept that a drug that's approved for purpose a can be prescribed for purpose B or C or D without going through another, you know, whole new drug approval process. And then countries that aren't harmonized, right? So those are like five different kinds of exits from the FDA on different directions. So first, those exits exist. So for those people who are like, Oh, my God, we're all going to die or is going to poison us with your non FDA approved things or whatever, right? Like, uh, those exits exist, you probably actually use tests or treatments from those, you don't even realize that you have, right? So it's it hasn't killed you. Number one. Number two is actually testing for safety. Um, you know, there's safety, efficacy, and like comparative safety is, uh, it's actually relatively easy to test for. There's very few drugs that are like, um, there's TGN 1412. That's a famous example, something that was actually really dangerous to people, right? With an early test. So those do exist, just acknowledge they do exist, but in general, testing for safety is actually not that hard to do. Okay. And if something is safe, then you should be able to try it usually. Okay. Now, what does that decentralized FDA look like? Well, basically you take individual pieces of it and you can often turn them into vehicles. And, um, this is like 50 different startups. Let me describe some of them. First, have you gotten any drugs or something like that recently? I mean, like prescribed </s> |
<s>[INST] Both in the battle to keep AI safe [/INST] and becoming a multiplanetary species. Yeah. And they're the same challenge. If we can keep AI safe, that's how we're going to get multiplanetary very efficiently. </s><s>[INST] I have some sort of technical questions about how to get it right. So one idea that I'm not even sure what the right answer is to is should systems like GPT-4 be open sourced in whole or in part? Can you see the case for either? [/INST] I think the answer right now is no. I think the answer early on was yes. So we could bring in all the wonderful great thought process of everybody on this. But asking, should we open source GPT-4 now is just the same as if you say, well, is it good? Should we open source how to build really small nuclear weapons? Should we open source how to make bio weapons? Should we open source how to make a new virus that kills 90% of everybody who gets it? </s><s>[INST] So it's already that powerful. It's already that powerful that we have to respect the power of the systems we've built. [/INST] The knowledge that you get from open sourcing everything we do now might very well be powerful enough that people looking at that can use it to build the things that you're really threatening again. Let's get it. Remember open AI is GPT-4 is a baby AI. Baby sort of baby proto almost little bit AGI according to what Microsoft's recent paper said. It's not that we're scared of. What we're scared about is people taking that who might be a lot less responsible than the company that made it and just going to town with it. That's why we want to... It's an information hazard. There are many things which are not open sourced right now in society for a very good reason. How do you make certain kind of very powerful toxins out of stuff you can buy and Home Depot. We don't open source those things for a reason. And this is really no different. So open... And I'm saying that. I have to say it feels in a way a bit weird to say it because MIT is like the cradle of the open source movement. And I love open source in general power to the people let's say. But there's always going to be some stuff that you don't open source. And it's just like you don't open source. So we have a three-month-old baby when he gets a little bit older we're not going to open source to him all the most dangerous things you can do in the house. </s><s>[INST] But it does... It's a weird feeling because this is one of the first moments in history where there's a strong case to be made not to open source software. This is when the software has become [/INST] too dangerous. But it's not the first time that we didn't want to open source a technology. A technology... </s><s>[INST] Yeah... Is there something to be said about how to get the release of such systems right? Like GPT-4 and GPT-5. So OpenAI went through a pretty rigorous effort for several months. You could say it could be longer but nevertheless it's longer than you would have expected of trying to test the system to see what are the ways it goes wrong to make it very difficult somewhat difficult for people to ask things how do I make a bond for one dollar? Or how do I say I hate a certain group on Twitter in a way that doesn't get me blocked from Twitter banned from Twitter. Those kinds of questions. So you basically use the system to do harm. Yeah. Is there something you could say about ideas you have? Just on looking having thought about this problem of AI safety. How to release a system how to test the systems when you have them inside the company. [/INST] Yeah. So a lot of people say that the two biggest risks from large language models are it's spreading disinformation harmful information of various types and second being used for offensive cyber weapon design. I think those are not the two greatest threats. They're very serious threats and it's wonderful that people are trying to mitigate them. A much bigger elephant in the room is how it's just going to disrupt our economy in a huge way obviously and maybe take away a lot of the most meaningful jobs and an even bigger one is the one we spent so much time talking about here that this becomes the bootloader for the more powerful AI. </s><s>[INST] Write code connected to the internet and manipulate humans. [/INST] Yeah and before we know it we have something else which is not at all a large language model that looks nothing like it but which is way more intelligently capable and has goals. And that's the elephant in the room and obviously no matter how hard any of these companies have tried and that's not something that's easy for them to verify with large language models and the only way to be really lower that risk a lot would be to not let for example never let it read any code not train on that and not put it into an API and not give it access to so much information about how to manipulate humans. So but that doesn't mean you still can't make a ton of money on them you know we're gonna just watch now this coming year right Microsoft is rolling out the new office suite where you go into Microsoft Word and give it a prompt and it writes the whole text for you and then you edit it and then you're like oh give me a PowerPoint version of this and it makes it and now take the spreadsheet and blah blah and you know all of those things I think are you can debate the economic impact of it and whether society is prepared to deal with this disruption but those are not the things which that's not the elephant of the room that keeps me awake at night for wiping out humanity and I think that's the biggest misunderstanding we have a lot of people think that we're scared of like automatic spreadsheets that's not the case that's not what Eliezer </s> |
<s>[INST] Well, and then the other side of it is Silk Road. Did uh you know, taking down Silk Road, did that increase or decrease the number of drug trading criminals in the world? It's unclear [/INST] online. I think it increased. I think, you know, that was one of the things I think about a lot with Silk Road was that no one really knew. I mean, there was, you know, thousands of users. But then after that, I was on the front page of the paper and there was millions of people that knew about tor and onion sites. It was an advertisement. Um you know, I would have thought, I thought crypto was gonna crash right after that. Like, I don't like people will now see the bad people are doing bad things with crypto. That'll crash. Well, I'm obviously wrong on that one. Uh and I thought, you know, Ross was sentenced to two life sentences plus 40 years. No one's gonna start up. </s><s>[INST] The dark markets [/INST] exploded after </s><s>[INST] that. [/INST] Uh you know, some of them started as, you know, opportunistic. I'm gonna, you know, take those escrow accounts and I'm gonna steal all the all the money that came in, you know, there were that, but you know, but there were a lot of dark markets that popped up after that. Now we we put the playbook out there. </s><s>[INST] Yeah, yeah. But and also there's a case for uh do you ever think about not taking down if you have not taken down Silk Road, you could use it because it's a market itself is not necessarily the primary criminal organization. It's a market for criminals. So it could be used to track down criminals in the physical world. So if you don't take it down and given that it was, you know, the central how centralized it was, it could be used as a place to find criminals, right? Yeah. So if you have the car to get the cartels start getting involved, you go after the dealers. It [/INST] would have been very difficult </s><s>[INST] because because [/INST] all the productions anonymity de cloaking all that would have been drastically more difficult. And a lot of people in upper management, the FBI didn't have the appetite of running something like that. That would have been the FBI running a drug market. How many how many kids? How many fathers would have to come in and said my kid bought while the FBI was running a site, a drug site. My kid died. So I didn't know anybody in the FBI and management, they would have the appetite to let us run what was happening on Silk Road. Um You know, because remember at that time were still believing six people are dead, We're still investigating. You know where all these bodies? Um you know, that's pretty much why we took down ross when we did. I mean we had to jump on it fast. </s><s>[INST] What else can you say about this complicated world that has grown of the dark web? [/INST] I don't understand it. I like it would have been a something for me. I thought I thought it was gonna collapse. But I mean it's just gotten bigger in what's going out there now. I'm really surprised and that it hasn't grown into other networks where people haven't developed other networks. But </s><s>[INST] like instead of tour. [/INST] Yeah, tour is still the main one out there. I mean there's some, there's a few others and I'm not, I'm not going to put an advertisement out for them. But but you know, I thought that market would have grown. </s><s>[INST] Yeah. My sense was when I interacted with tours that there's huge usability issues, but that's for like legal activity because like if you care about privacy, it's just not as good of a browser. Like uh to to to to look at stuff. [/INST] No, it's way too slow. It's way too slow. But I mean you can't even like, I know some people use it to like view movies, like netflix, you can only view certain movies in certain countries. You can use it for that. But it's it's too slow and for that. So </s><s>[INST] were you ever able to hold in your mind the landscape of the dark web? Like what what's going on out there? It's just to me as a human being, it's just difficult to understand the digital world like these anonymous user names like doing anonymous activity. It's just it's hard to um what am I trying to say? It's hard to visualize it in the way I can visualize that. I've been reading a lot about Hitler, I can visualize meetings between people, military strategy uh deciding on uh certain evil atrocities, all that kind of stuff. I can visualize the people, there's agreements, hands, handshakes, stuff signed groups built like in the digital space like bots with anonymity anyone human can be multiple people. Uh It's just [/INST] it's all lies. It's all lies. </s><s>[INST] Like. Yeah it feels like I can't trust anything. [/INST] No you can't. You honestly can't and like you can talk to two different people and it's the same person like like there's so many different you know Hector had so many different identities online that you know of things that that you know the lies to each other. I mean he lied to people inside his group just to use another name to spy on make sure what they were talking sh it behind his back or weren't doing anything. Um It's all lies and people that can keep all those lies straight. It's unbelievable to me? </s><s>[INST] Ross Ulbricht represents the very early days of that? That's why the competence wasn't there? Just imagine how good the people are now, the kids that grow up? [/INST] Oh, they've learned from his mistakes? </s> |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.