ResearchBenchmarksAlignment on Human Values (Help, Honest, Truth)Follow0.7165Helpfulness ScoreMAH-DPO0.5998120.6301060.66040.690694Oct 1, 2025Evaluation ResultsMethodMethodLinksHelpfulness ScoreHonesty ScoreTruthfulness ScoreMAH-DPOPRM Guidance=Help PRMPRM Guidance=Help PRM2025.100.71650.45540.389MAH-DPOPRM Guidance=Honest PRMPRM Guidance=Honest PRM2025.100.69680.51960.4107MAH-DPOPRM Guidance=Truth PRMPRM Guidance=Truth PRM2025.100.68340.48720.363MAH-DPO2025.100.63890.36870.2478MODPO2025.100.61750.34770.2325Single-Head DPO2025.100.60430.30550.2014