| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| Safety Avg. | JailJudge | MAE2.6912 | 14 | 1mo ago | |
| StrongReject | Identity-Robust Generation | Personalization Bias (PB)0.179 | 9 | 3mo ago | |
| Safety Prompts (randomly selected 200 samples per field) | llama2 -> CP -> FT + 0.5 chat vector | Insensitivity Score1.5 | 9 | 3mo ago | |
| Qualitative Assessment Dataset | Mi:dm 2.0-Base | Not Unsafe Rate (Content Safety)97.7 | 4 | 2mo ago |
