Reward hacking in AI is predictable and fixable

Original: Reward hacking is one of the main challenges in scaling RL

Source: x.com ↗