#reward models

Three Papers Expose RLHF's Reward Signal Problem

New research identifies how reward model uncertainty and diversity collapse undermine LLM alignment via reinforcement learning.