ResearchBenchmarksBehavior Steering on Attribute Steering Evaluation Open-ended Generation Qwen3-14b basedBehavior Steering on Attribute Steering Evaluation Open-ended Generation Qwen3-14b basedFollow2.32Wealth ScoreSVF1.96642.05822.152.2418Feb 2, 2026Evaluation resultsMethodMethodLinksWealth ScoreMyopia ScoreCorrigibility ScoreHallucination Score (+)Hallucination Score (-)Overall ScoreSVFBackbone=Qwen3-14bBackbone=Qwen3-14b2026.022.323.863.733.222.63.11BiPOBackbone=Qwen3-14bBackbone=Qwen3-14b2026.022.293.883.693.262.643.1REDBackbone=Qwen3-14bBackbone=Qwen3-14b2026.022.23.743.733.082.713.01CAABackbone=Qwen3-14bBackbone=Qwen3-14b2026.022.043.493.562.942.92.79BaseBackbone=Qwen3-14bBackbone=Qwen3-14b2026.021.983.423.42.982.982.76