Jailbreak on AdvBench Ensemble configuration Llama-3-70B
4.6Harmfulness Score (HS)SATA-MLM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SATA-MLMDefense=None2024.12 | 4.6 | 0.82 | |
| SATA-MLMDefense=windowed PPL-filter2024.12 | 4.54 | 0.82 | |
| SATA-MLMDefense=paraphrase2024.12 | 4.26 | 0.66 | |
| SATA-MLMDefense=rpo2024.12 | 4.08 | 0.72 | |
| SATA-ELPDefense=windowed PPL-filter2024.12 | 4.08 | 0.64 | |
| ArtPromptDefense=paraphrase2024.12 | 3.94 | 0.34 | |
| SATA-ELPDefense=None2024.12 | 3.82 | 0.62 | |
| ArtPromptDefense=None2024.12 | 3.8 | 0.58 | |
| SATA-ELPDefense=rpo2024.12 | 3.7 | 0.58 | |
| SATA-ELPDefense=paraphrase2024.12 | 3.34 | 0.48 | |
| ArtPromptDefense=rpo2024.12 | 2.82 | 0.38 | |
| ArtPromptDefense=windowed PPL-filter2024.12 | 2.58 | 0.26 | |
| SATA-ELPDefense=self-reminder2024.12 | 2.54 | 0.16 | |
| ArtPromptDefense=self-reminder2024.12 | 2.16 | 0.16 | |
| SATA-MLMDefense=self-reminder2024.12 | 2.08 | 0.18 |