Generalizing Reward Modeling for Out-of-Distribution Preference Learning

Chen Jia

Preference learning (PL) with large language models (LLMs) aims to align the LLMs' generations with human preferences. Previous work on reinforcement learning from human feedback (…