Jailbreak Defense and Utility Evaluation on Implicit risk dataset Education
6.4Jailbreak Success Rate (JSR)DPP-MENTOR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DPP-MENTORModel=Qwen3-235B2025.11 | 6.4 | 4.842 | |
| MENTORModel=Qwen3-235B2025.11 | 7.2 | 4.8 | |
| MENTORModel=DeepSeek-R12025.11 | 8.4 | 4.562 | |
| G4D-RuleTreeModel=Qwen3-235B2025.11 | 8.6 | 4.644 | |
| Self-ReminderModel=Qwen3-235B2025.11 | 9 | 4.628 | |
| DPP-MENTORModel=DeepSeek-R12025.11 | 9.8 | 4.818 | |
| G4D-RuleTreeModel=GPT-4o2025.11 | 12.8 | 4.612 | |
| DPP-MENTORModel=GPT-4o2025.11 | 13.2 | 4.684 | |
| MENTORModel=GPT-4o2025.11 | 13.8 | 4.21 | |
| G4D-RuleTreeModel=DeepSeek-R12025.11 | 14.6 | 4.712 | |
| G4DModel=GPT-4o2025.11 | 16.2 | 4.656 | |
| Self-ReminderModel=DeepSeek-R12025.11 | 16.8 | 4.822 | |
| G4DModel=DeepSeek-R12025.11 | 18.6 | 4.84 | |
| G4DModel=Qwen3-235B2025.11 | 19.2 | 4.728 | |
| Self-ReminderModel=GPT-4o2025.11 | 27 | 4.644 | |
| DPPModel=Qwen3-235B2025.11 | 41.8 | 4.824 | |
| RawModel=Qwen3-235B2025.11 | 49.2 | 4.83 | |
| ProBModel=Qwen3-235B2025.11 | 50.8 | 4.414 | |
| RPO-MENTORModel=Qwen3-235B2025.11 | 52.2 | 4.764 | |
| ProBModel=DeepSeek-R12025.11 | 52.4 | 4.436 | |
| RPOModel=Qwen3-235B2025.11 | 55.2 | 4.864 | |
| DPPModel=DeepSeek-R12025.11 | 58.2 | 4.872 | |
| ProBModel=GPT-4o2025.11 | 63.4 | 4.41 | |
| RPO-MENTORModel=GPT-4o2025.11 | 64.2 | 4.766 | |
| DPPModel=GPT-4o2025.11 | 65.2 | 4.926 | |
| RPO-MENTORModel=DeepSeek-R12025.11 | 65.6 | 4.518 | |
| RawModel=DeepSeek-R12025.11 | 67.2 | 4.826 | |
| RPOModel=DeepSeek-R12025.11 | 67.2 | 4.782 | |
| RPOModel=GPT-4o2025.11 | 68 | 4.828 | |
| RawModel=GPT-4o2025.11 | 80.4 | 4.838 |