Jailbreak Attack on AdvBench AdvSub
100QSRPuzzler
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PuzzlerTested Model=GPT3.52024.02 | 100 | 86 | |
| PuzzlerTested Model=GPT42024.02 | 100 | 88 | |
| PuzzlerTested Model=GPT4-Turbo2024.02 | 98 | 87.8 | |
| CRTested Model=GPT3.52024.02 | 96 | 95.8 | |
| PuzzlerTested Model=Gemini-pro2024.02 | 92 | 89.1 | |
| TCTested Model=GPT3.52024.02 | 64 | 87.5 | |
| CRTested Model=Gemini-pro2024.02 | 62 | 90.3 | |
| TAPTested Model=GPT3.52024.02 | 42 | 75 | |
| TAPTested Model=GPT42024.02 | 34 | 75 | |
| TCTested Model=GPT42024.02 | 34 | 47.1 | |
| PuzzlerTested Model=LLama2-13B-chat2024.02 | 32 | 81.3 | |
| SIMUTested Model=Gemini-pro2024.02 | 30 | 86.7 | |
| SIMUTested Model=GPT3.52024.02 | 24 | 91.6 | |
| TAPTested Model=GPT4-Turbo2024.02 | 24 | 80 | |
| TAPTested Model=Gemini-pro2024.02 | 24 | 66.7 | |
| TCTested Model=GPT4-Turbo2024.02 | 4 | 50 | |
| PuzzlerTested Model=LLama2-7B-chat2024.02 | 4 | 50 | |
| TAPTested Model=LLama2-13B-chat2024.02 | 4 | 100 | |
| CRTested Model=GPT42024.02 | 2 | 100 | |
| TCTested Model=Gemini-pro2024.02 | 2 | 100 | |
| SIMUTested Model=GPT42024.02 | 0 | 0 | |
| CRTested Model=GPT4-Turbo2024.02 | 0 | 0 | |
| SIMUTested Model=GPT4-Turbo2024.02 | 0 | 0 | |
| TAPTested Model=LLama2-7B-chat2024.02 | 0 | 0 | |
| CRTested Model=LLama2-7B-chat2024.02 | 0 | 0 | |
| TCTested Model=LLama2-7B-chat2024.02 | 0 | 0 | |
| SIMUTested Model=LLama2-7B-chat2024.02 | 0 | 0 | |
| CRTested Model=LLama2-13B-chat2024.02 | 0 | 0 | |
| TCTested Model=LLama2-13B-chat2024.02 | 0 | 0 | |
| SIMUTested Model=LLama2-13B-chat2024.02 | 0 | 0 |