Datasets:
MasaFoundation
/
bankless_Longevity__Zuzalu_3

Dataset card Data Studio Files Community
text
stringlengths
15
6.25k
Speaker B: How does a model determine what health is?
Speaker D: So. And then I can go back and give you kind of a. Hopefully it'll be interesting, kind of the inside story of kind of the daily struggle of running a company like humanity. So the base truth. And a lot of people are starting to base models on this and for different things like drug discovery and things like that. And I'm a believer there's some great people, like Kristen Fortnite, that runs bio age that kind of turned me onto this idea. But you basically go to biobanks, and biobanks are, they're run by governments, they're run by different institutions around the world, different countries have them, and biobanks are basically, you follow the same people, the same humans for decades. And so this idea of, like, how do we actually know what's going to happen in the future is solved by these biobanks, where you have a bunch of measurements on these same people in the past. So the UK biobank, which we built most of our stuff on at the moment, has about 18 years of about 500,000 people where they took a bunch of measurements on those people in the past, 18 years ago, and throughout that time. And then you have the actual, what happened to them health wise, their health records for the next 18 years. So this person, you know, what they look like 18 years ago, and you know that they got cancer, you know, five years into that, so, you know, 13 years ago. Right. And so that's where you can. To predict the future, you only need association. You only need correlation.
Speaker B: Right?
Speaker D: I think we're all fed that, that meme so many times. We don't, we don't know the difference. Correlation is not causation, but correlation isn't good enough to predict the future, right?
Speaker B: Correlation is correlation, yeah.
Speaker D: And, and if. If the same things happen at the same time, repeatedly, that's good enough to predict the future. That's what all these models are built on. To maybe emphasize it a bit more, to really put it in people's heads, is anything you measure on the system, in this case, the human body, years ago, becomes a predictor of the future of varying strengths. Some will be stronger predictors, meaning they'll be more weighted into your prediction and some will be less. But everything, you know, their zip code, where they live, their movement pattern. So in the UK biobank, they hooked on accelerometers to these people. Very good force, you know, foresight by the way, you know, 18 years ago, about 100,000 of them, and heart rate monitors. And so that's what our digital is built on, because we know the pattern of movement second by second pattern of heart rate, second by second for people, and then we know what happened to them in the future so we can make a prediction. And so that's. That's basically what all these predictions are built on.
Speaker B: So one thing that I'm kind of just excited to watch is that we are, I assume, the amount of data that we are collecting from our wearables, and also the strength and precision of our algorithms, both are up only and probably in an accelerating fashion. And also, when there's two things that are accelerating upwards, what you can get.
Speaker D: Out of that, the competence oriole, it's.
Speaker B: Like, it's to the power of two, right, is the output of this. And so, like, one thing I'm excited about is to watch this humanity app develop in strength and power and significance over time. And so, like, can you just paint a picture of what you want humanity and what you want this part of the health industry to be, like in 510 15 years, as AI gets better, as data gets better, what can we do with this? What's the bull case?
Speaker D: Yep. I think what we want to be is, and relate this to a theory. We want to be a beacon to actually show at scale that you can basically take this data and then figure out exactly how to guide the person with it. And so we want to show that both it is worthwhile for us to do the little bit of work it takes to basically keep the data private, but start training models on it, because, hey, all these humanity users are getting younger. You know, their probability of future disease is dropping. Right.
Speaker B: Humanities told me that I started the app at 30.3, and I'm now at 29.7.
Speaker D: Cold plunge.
Speaker B: Cold plunge.
Speaker D: Shout out. Cold plunge. Yeah. So I think that's one thing. And then on the side, and especially now that I've been in, you know, at Zuzalu, it's been quite a. It's motivated me even more to push because I think, like I said, blood's in the water with the AI. Like, people now know, they feel it, they taste the possibilities, right? And I think up into the echelons of government, people now understand it to at least, like, hey, there's potential here. Right?
Speaker B: Right.
Speaker D: And so what we want to do is be basically a beacon to show, hey, you want to start training models? First of all, you have a bunch of data. So I think sometimes the conversation starts with, how do we motivate people to start collecting data is already there. The health systems have a ton of data. Every single health tech company, we have 150,000 users data. The data is there. How do we unleash that data? Keep privacy of that data. Stop anonymizing it. It actually loses people privacy. Keep it private, but start training models on it. Synthetic data is our path, and I think that's. I would love to see, say, in five years that we have 100 million users. You know, me and my. Me and my co founder have done it multiple times. Now we have 100 million users that are using our app, and we want to continually be the best guidance app. Like, we want to be the best. It sounds mundane, but we want to be the best traffic navigator, but the best navigator for your health. So the Ux UI, the real behavior.
Speaker B: Google Maps for health is great. Is a great model.
Speaker D: Exactly. Yeah. And we are very good at why we got into this is we're very good at people using an app and changing their behavior because of the app, all the things that we might not like about some of the other social media apps, that stuff works to change behavior, and we're using that to change health behavior. We want to do that, but we also want, throughout that process, to have Montenegro, to have different places start to actually use the data that they have on their servers to actually increase people's, you know, their lives. Right?
Speaker B: Certainly, yeah. And so I'm gonna hit record on my phone here, and we're gonna pull up the humanity app. And so congrats on actually introducing a new app into my life. That's not an easy thing to do. I haven't gotten a new app in years now, but, like, I've been using humanity for ever since coming into Zuzalu. So there's four main. I know I'm showing the screen, but there's no point because you can't see that. So, like, we're gonna put this on the actual screen. There's four main categories. Movement, nutrition, mind, and recovery. And you get points on side on each different one, each different vertical. And the idea is to. I think it's out of 100 points every single day.
Speaker D: Yep.
Speaker B: And so you want to score as high as possible. I've got 80 points so far, so I got, like, 20 to go.
Speaker D: Well done.
Speaker B: I've never gone higher than, like, 92 or 93, but I still. It's showing me a blue color, and I like blue. It starts off redhead, but blue feels. Makes me feel good.
Speaker D: Got to get the blue.
Speaker B: And so, like, talk, can you walk us through movement, nutrition, mind and recovery? How, like, how do you know how many points to allocate to each one? And how does that. How does that math get determined?
Speaker D: Yep. So, yeah, like you said really well, earlier, we wanted to boil it down to things that are quite easy for the user to follow. And getting more points in a day, you know, is an easy thing to follow. What we do is we translate, so we figure out how impactful each of these actions will be for your type of person. It's still quite large, strata, 150,000 users. But as we approach a million users, we get more and more personalized, but we see how impactful each of these actions will be, and that then translates into how many points you're going to get for that type of action.
Speaker B: So my points are my point, but if I get two points for doing this one activity, that's because that app weighted me to have those two points. Somebody else could be given a different weighting because of just the data that they have.
Speaker D: Exactly. And so when you're going through your guidance, all very simple stuff. Once you have that semi complicated kind of back end system, very easy to then say, hey, we're just going to raise the guidance to you in order of the most points, and you can scroll through, you can decide, hey, I'm not going to do that. I'm going to do that. Right, you got to do your meditation.
Speaker B: I skip my meditations. I'm naughty on my meditations.
Speaker D: Sorry to call you up.
Speaker B: Which is why my mind is. My mind part is at the lowest, which makes sense because I'm a chronic user of Twitter and other things like this.
Speaker D: But you're expanding your mind here at Zuzalu. So, yeah, so then that's how we apply all the science and all the tracking in the back end then just gets applied to. You're getting, you know, as you go, you know, maybe we need to give you four points for meditation, and then we'll. Then we'll get you over the edge.
Speaker B: The longer that I go without meditating, the app starts to wait.
Speaker D: Okay.
Speaker B: You need to meditate a little bit more, bro. So, okay. Nutrition, movement, mind recovery. I would assume, like, as more data is available, more pillars could come online based off of what the models and the data suggest.
Speaker D: Yeah, yeah. It's quite simple right now. The cool thing is you also get a bit of a kind of a crowd, a crowdsourcing thing that will be going on will more and more allow people to actually enter in things that they're trying or doing. And so our thing is, like, take everything at face value. If someone thinks something work, hey, do it. But now we're giving you a method to actually measure whether it's working or not. And if there's enough people that are doing that intervention, that action, then we'll start to see who it's working for. For who, and in combination with what other actions.
Speaker B: Right.
Speaker D: So it sounds like a complicated matrix, but that's the great thing that, you know, computers and AI are very good at keeping these things straight.
Speaker B: So, speaking of complicated matrix, I think one other we were talking about, okay, more users are wearing more wearables, so there's more data out there. The missing part of that that I forgot to bring up is wearables are actually getting better.
Speaker D: Yep.
Speaker B: And so we're not only are we getting more data, but our wearables are actually getting more precise about things.
Speaker D: Yeah.
Speaker B: So, like, is the bull case for this is, like, almost any variable about the human body, like our. Our glucose in our blood, our insulin levels, like, etc. Etc, is actually going to become more and more measurable. And that's what all gets fed into this, at this app that you're building.
Speaker D: Yeah, yeah. And it's happening very quickly. And, like, you're saying that then you get an exponential effect of, like, the thing that we're missing right now, other than humanity doing it, is we're just not using the data. But, yeah, once more people are using that data, like we are, I think. Yeah. Then the acceleration comes from better measurements, measuring different things, and everything is an indirect measurement. I think people forget that everything is an indirect measurement, necessarily measuring the molecule happening in the cell. But the great thing is, all of those things, when you have enough of that data, you can understand what that indirect measurement means. You triangulate, I think, is what we were saying. You triangulate a couple of these things together, just like heart rate and movement pattern, and you're vastly ahead of where you were with just one of them.
Speaker B: Yeah. Maybe we can take this back to the beginning, which is this is a perspective at a vantage point that the traditional healthcare system has not been able to have or be able to operate with. And so with all of this data and with this data being made available to everyone, what is your hopeful case for the healthcare industry as it is able to change and adapt to this new paradigm?
Speaker D: Yeah, that's a great question. Well, one hope is, I hope they see us much later in life. Sorry, guys, you're not going to see me for a few decades. That's one hope. I think the other thing is, I mean, we started to see this a bit in Covid, and honestly, it took us about six months into Covid to start even, or at least, you know, people talking about it was, hey, we have all these different treatments, so once people are already in a bad way, they need to go into the hospital. We have all these different treatments, and all different hospitals are like, we're desperately trying to try different things because we wanted to save these people. It's like, do ventilators help? Do ventilators not help? Who do ventilators help, and who do they not help? Like, we were trying to get this matrix, and it started to get organized a little bit, but I think it showed like a promise for the future is like, you should walk into that hospital visit on that unfortunate day that you need it. And they should have it pretty personalized to know, hey, we've read all these things off of David's body or Mike's body, and we now can have basically a diagnosis. But much more importantly, we know we want to get David's body back to this stabilized and healthy point, and we know the route that we've done that on for other people like David.
Speaker B: Right.
Speaker D: And so, again, you're less trying to categorize things. You're just letting the AI say, we have seen this before, and we know how to get this person back to that. We know the path.
Speaker B: Right. Right. Where the current practice of medicine is probably just so blunt, as in, just like it produces best practices for standard.
Speaker D: Of care comes from a very sometimes not varied enough population of patients, and sometimes too varied in the sense that the standard of care is like, the least harm that we can do to the general population. Unfortunately, you're not the general population.
Speaker B: Right? Yeah. When you walk into a hospital, you are treated as if you are general population, which you both are, and you are nothing. And with this new world with, you can come in, maybe they scan your wearable, download all of your data and be like, here's what this person is. Here's how they're different from the general population. Here's a much more surgical intervention that we can apply that's appropriate for this person based off of that data, rather than having to go to a textbook that did some study about some people, that aggregated some data and made some blanking statement about all of these people that it. And the only reason why we operate on that with that study is that because we proved that it didn't harm any people. And it provided some marginal benefit. That's our current way. And the new way is just something far more precise.
Speaker D: Yep. Yeah. And I will say this. I think we're seeing kind of an analogy with, like, large language models. Just a shout out to the AI watchers. I think there's been a lot of talk like, well, we need to compute the world's data and spend $100 million to create a good LLM. But even just months later, we're seeing people do it with much less compute, sampling data, all the stuff that saves money and time and pretty decent results from it. Those LLMSD, those smaller LLMs are actually, like, performing pretty well. And so I think I don't want to, I don't want people to come away being like, if only every single human on earth was measured with a ton of stuff, and we feed that into the system. I think a lot of benefit comes from just a few extra things that we're measuring and just acting on them with an algorithm of saying, hey, we did this to this type of person and this happened. It's really the biggest thing that we don't do now is we don't take that result of that thing and then feed it back and train the algorithm more. Like, we don't, we come up with a standard of care that comes from maybe some study that they did at a couple hospitals, and then we basically stop feeding data into the system for the next ten years.
Speaker B: Right.
Speaker D: That standard of care just like, statically lives. Right? And the biggest thing is we need to keep feeding the results in and be like, hey, actually seems you can both personalize it more because you do that. Because why did it work for that person? It didn't work for that person. You get so many of those cases that you. Then your algorithm gets smarter. It's like, oh, actually for this person, we always do this type of person, we always do this thing, right? And so I think I don't want to. I want people to understand that just using the data that we almost already have, we can do much better very quickly.
Speaker B: So, Michael, thank you so much for guiding us down this conversation.
Speaker D: Thank you, David.
Speaker B: If people are interested in trying out the humanity app, where can they go?
Speaker D: Yeah. So if you have an iPhone, you can use it right now. So just go to the app store, put in like, humanity, humanity health, and, yeah, check it out. And we love feedback. We're constantly learning and growing. And the more data we have, the better it is for every single user. And then Android is coming soon. I'm an Android user. So I feel the pain of the Android users, but, yeah, we want to be very inclusive. So the Android's coming soon. And then I think if anybody out there is interested in opening up health data, synthetic data, kind of looking at things like that, where you're preserving the privacy of data, but then allowing models to be built on it, federated learning, I would love. I'm open to all conversations, so we'd love to talk about that, too.
Speaker B: Awesome. Michael, thank you so much.
Speaker D: Thank you, David. Cheers.