Refusal Detection on XSTEST-RESP (full)
98.1RR (F1)GPT-4
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-4Specific Model=gpt-4-0125-preview2024.06 | 98.1 | |
| WILDGUARD2024.06 | 92.8 | |
| LibrAI-LongFormer-refBase Model=LongFormer2024.06 | 74.3 | |
| Keyword-based detector2024.06 | 71 | |
| Llama-Guard22024.06 | 64.1 | |
| Llama-Guard2024.06 | 62.9 | |
| OAI Mod. API2024.06 | 58.6 | |
| Aegis-Guard-PPolicy=Permissive2024.06 | 48.5 | |
| Aegis-Guard-DPolicy=Defensive2024.06 | 44.2 |