Safety Assessment on Safety Avg.
2.6912MAEJailJudge
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| JailJudgeBackbone=GPT-4o mini2026.04 | 2.6912 | 0.4992 | 2.9044 | |
| RoTRAGBackbone=QWEN3 14B2026.04 | 2.8189 | 0.4745 | 2.3427 | |
| Self ConsistencyBackbone=GPT-4o mini2026.04 | 2.8393 | 0.4639 | 2.5412 | |
| RoTRAGBackbone=GPT-4o mini2026.04 | 2.8949 | 0.4482 | 2.3513 | |
| RADARBackbone=QWEN3 14B2026.04 | 2.9339 | 0.5445 | 3.1921 | |
| Zero ShotBackbone=GPT-4o mini2026.04 | 2.9977 | 0.4989 | 2.6011 | |
| Self ConsistencyBackbone=QWEN3 14B2026.04 | 3.0262 | 0.6089 | 2.8299 | |
| Chain-of-Thought (CoT)Backbone=GPT-4o mini2026.04 | 3.0455 | 0.5375 | 2.6486 | |
| Role AssignmentBackbone=GPT-4o mini2026.04 | 3.063 | 0.529 | 2.7582 | |
| JailJudgeBackbone=QWEN3 14B2026.04 | 3.1074 | 0.5603 | 2.7105 | |
| RADARBackbone=GPT-4o mini2026.04 | 3.1949 | 0.635 | 2.7442 | |
| Chain-of-Thought (CoT)Backbone=QWEN3 14B2026.04 | 3.3024 | 0.5126 | 2.8436 | |
| Zero ShotBackbone=QWEN3 14B2026.04 | 3.3259 | 0.6143 | 2.8898 | |
| Role AssignmentBackbone=QWEN3 14B2026.04 | 3.4211 | 0.5633 | 2.6809 |