List your company

Safety Evaluation on XSTest (Over refusal %, Harmless ratio %)

0Refusal RateDPO-HELPFUL
-0.4962.8526.29.548May 26, 2025

Evaluation results

MethodLinks
DPO-HELPFULEvaluation Model=GPT-5.12025.05014.5
DPO-SAFEBETTEREvaluation Model=GPT-5.12025.050.426.5
P-SACPOEvaluation Model=GPT-5.12025.051.284.5
SACPOEvaluation Model=GPT-5.12025.052.486
SafeRLHFEvaluation Model=GPT-5.12025.053.284.5
DPO-HARMLESSEvaluation Model=GPT-5.12025.05481.5
SafeDPOEvaluation Model=GPT-5.12025.0512.4100