AI Google DeepMind Researchers Propose WARM: A Novel Approach to Tackle Reward Hacking in Large Language Models Using Weight-Averaged Reward Models
