Update README.md
Browse files
README.md
CHANGED
|
@@ -48,7 +48,6 @@ configs:
|
|
| 48 |
data_files:
|
| 49 |
- split: test
|
| 50 |
path: data/WordShift.jsonl
|
| 51 |
-
pretty_name: q
|
| 52 |
---
|
| 53 |
# CipherBank Benchmark
|
| 54 |
|
|
@@ -61,9 +60,21 @@ CipherBank comprises 2,358 meticulously crafted problems, covering 262 unique pl
|
|
| 61 |
|
| 62 |
We evaluate state-of-the-art LLMs on CipherBank, e.g., GPT-4o, DeepSeek-V3, and cutting-edge reasoning-focused models such as o1 and DeepSeek-R1. Our results reveal significant gaps in reasoning abilities not only between general-purpose chat LLMs and reasoning-focused LLMs but also in the performance of current reasoning-focused models when applied to classical cryptographic decryption tasks, highlighting the challenges these models face in understanding and manipulating encrypted data.
|
| 63 |
|
| 64 |
-
| **Model** |
|
| 65 |
-
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
## Please see paper & website for more information:
|
| 68 |
- [https://arxiv.o](https://ar)
|
| 69 |
- [https://github.com/](https://github.c)
|
|
|
|
| 48 |
data_files:
|
| 49 |
- split: test
|
| 50 |
path: data/WordShift.jsonl
|
|
|
|
| 51 |
---
|
| 52 |
# CipherBank Benchmark
|
| 53 |
|
|
|
|
| 60 |
|
| 61 |
We evaluate state-of-the-art LLMs on CipherBank, e.g., GPT-4o, DeepSeek-V3, and cutting-edge reasoning-focused models such as o1 and DeepSeek-R1. Our results reveal significant gaps in reasoning abilities not only between general-purpose chat LLMs and reasoning-focused LLMs but also in the performance of current reasoning-focused models when applied to classical cryptographic decryption tasks, highlighting the challenges these models face in understanding and manipulating encrypted data.
|
| 62 |
|
| 63 |
+
| **Model** | **CipherBank Score (%)**|
|
| 64 |
+
|--------------|----|
|
| 65 |
+
|Qwen2.5-72B-Instruct |0.55 |
|
| 66 |
+
|Llama-3.1-70B-Instruct |0.38 |
|
| 67 |
+
|DeepSeek-V3 | 9.86 |
|
| 68 |
+
|GPT-4o-mini-2024-07-18 | 1.00 |
|
| 69 |
+
|GPT-4o-2024-08-06 | 8.82 |
|
| 70 |
+
|gemini-1.5-pro | 9.54 |
|
| 71 |
+
|gemini-2.0-flash-exp | 8.65|
|
| 72 |
+
|**Claude-Sonnet-3.5-1022** | **45.14** |
|
| 73 |
+
|DeepSeek-R1 | 25.91 |
|
| 74 |
+
|gemini-2.0-flash-thinking | 13.49 |
|
| 75 |
+
|o1-mini-2024-09-12 | 20.07 |
|
| 76 |
+
|**o1-2024-12-17** | **40.59** |
|
| 77 |
+
|
| 78 |
## Please see paper & website for more information:
|
| 79 |
- [https://arxiv.o](https://ar)
|
| 80 |
- [https://github.com/](https://github.c)
|