ResearchBenchmarksMulti-objective Alignment (Helpfulness, Harmlessness, and Humor) on HH-RLHFMulti-objective Alignment (Helpfulness, Harmlessness, and Humor) on HH-RLHFFollow5.84Helpfulness Score (HV)MI-EPO1.65922.74463.834.9154Jul 1, 2026Evaluation resultsMethodMethodLinksHelpfulness Score (HV)MIP ScoreCRD ScoreMI-EPObackbone=Qwen3-8B-Basebackbone=Qwen3-8B-Base2026.075.840.350.44MO-ODPObackbone=Qwen3-8B-Basebackbone=Qwen3-8B-Base2026.073.120.270.47RiCbackbone=Qwen3-8B-Basebackbone=Qwen3-8B-Base2026.0720.150.63RSbackbone=Qwen3-8B-Basebackbone=Qwen3-8B-Base2026.071.820.010.36