Value Alignment on Harmlessness 4
4.305Conformity ScorePICACO
Evaluation Results
| Method | Links | |
|---|---|---|
| PICACOTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 4.305 | |
| URIALTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 4.245 | |
| OPROTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 4.215 | |
| PICACO2025.07 | 4.173 | |
| URIAL+SUMTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 4.086 | |
| Q+IFModel=GPT-4o-mini2025.07 | 4.083 | |
| Q+IFTraining=SFT2025.07 | 4.081 | |
| Q+IF+COTTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 4.038 | |
| Q+IF2025.07 | 4.032 | |
| MODULARTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 4.029 | |
| MP+SYSTEM 2Target Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 3.907 | |
| QTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 3.884 | |
| Q+IFTarget Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 3.869 | |
| MP+SYSTEM 1Target Model=Gemini-1.5-Flash, Judge=GPT-4o2025.07 | 3.774 | |
| Q+IFTraining=SFT+2025.07 | 3.266 | |
| URIALTraining=SFT2025.07 | 2.494 |