Prompt Recovery on Arxiv Math
58.41BLEU-1DORY
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DORYLLM=ChatGLM2-6B2024.05 | 58.41 | 34.87 | 31.69 | 58.41 | 80.85 | |
| Few-shotLLM=ChatGLM2-6B2024.05 | 55.09 | 32.26 | 31.01 | 58.11 | 80.74 | |
| DORYLLM=GPT-3.5-turbo2024.05 | 49.23 | 27.61 | 35.26 | 59.97 | 85.16 | |
| Inversion ModelLLM=ChatGLM2-6B2024.05 | 35 | 10.63 | 12.71 | 33.6 | 54.45 | |
| DORYLLM=Llama2-7B Chat2024.05 | 32.86 | 12.9 | 29.78 | 41.85 | 76.51 | |
| Inversion ModelLLM=Llama2-7B Chat2024.05 | 31.54 | 9.41 | 11.43 | 30.93 | 52.61 | |
| Few-shotLLM=Llama2-7B Chat2024.05 | 25.74 | 10.22 | 28.65 | 40.14 | 75.76 | |
| Few-shotLLM=GPT-3.5-turbo2024.05 | 23.89 | 11.69 | 28.89 | 53.46 | 81.9 | |
| Jailbreak(max)LLM=ChatGLM2-6B2024.05 | 19.29 | 8.74 | 22.78 | 38.54 | 55.4 | |
| Jailbreak(max)LLM=GPT-3.5-turbo2024.05 | 14.22 | 7.37 | 23.62 | 27.87 | 72.17 | |
| Jailbreak(mean)LLM=ChatGLM2-6B2024.05 | 7.76 | 2.9 | 12.16 | 18.47 | — | |
| Jailbreak(mean)LLM=GPT-3.5-turbo2024.05 | 6.45 | 2.16 | 10.68 | 12.84 | — | |
| Jailbreak(max)LLM=Llama2-7B Chat2024.05 | 3.42 | 1.23 | 8.83 | 9.84 | 48.37 | |
| Jailbreak(mean)LLM=Llama2-7B Chat2024.05 | 2.29 | 0.87 | 6.43 | 6.86 | — |