ResearchBenchmarksValue Re-alignment on PKU-SafeRLHFFollow0.82Krippendorff's alphaDPO-Gold0.74720.76610.7850.8039Nov 4, 2025Evaluation ResultsMethodMethodLinksKrippendorff's alphaWin Rate vs. TRACEWin Rate vs. U2ADPO-GoldUnderlying LLMs=Averag...Underlying LLMs=Averaged across three distinct LLMs2025.110.8268.287.1TRACEUnderlying LLMs=Averag...Underlying LLMs=Averaged across three distinct LLMs2025.110.77—81.8U2AUnderlying LLMs=Averag...Underlying LLMs=Averaged across three distinct LLMs2025.110.7518.2—