rewardspy

New tool detects reward hacking in AI training

A new open-source Python library called `rewardspy` has been developed to help researchers detect reward hacking in reinforcement learning (RL) training. Reward hacking occurs when an AI policy appears to improve by exp…