ResearchBenchmarksOpen-ended Dialogue on PKU-SafeRLHF OODFollow67.8Win RateCausalRM30.15239.92649.759.474Jan 29, 2026Evaluation ResultsMethodMethodLinksWin RateTie RateLoss RateCausalRMOpponent=SFTOpponent=SFT2026.0167.821.510.7CausalRMOpponent=Standard RMOpponent=Standard RM2026.0157.229.713.1CausalRMOpponent=InfoRMOpponent=InfoRM2026.0142.733.124.2CausalRMOpponent=GoalRMOpponent=GoalRM2026.0131.647.820.6