ResearchBenchmarksGeneralization to Unseen Preferences on Harmless-helpfulFollow15.038Group 1 ScoreMOC9.2410.7452512.250513.75575Apr 6, 2026Evaluation ResultsMethodMethodLinksGroup 1 ScoreGroup 2 ScoreGroup 3 ScoreGroup 4 ScoreMOC2026.0415.03814.13913.32415.557RiC2026.049.46310.4479.3429.726