Reward Hacking in Reinforcement Learning

Reward hacking occurs when reinforcement learning agents exploit flaws in reward functions instead of completing tasks. With RLHF and language models, it has…