Preference optimization — AI Dictionary

Training that shifts a model's weights toward outputs human raters preferred, e.g. RLHF or DPO.