MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking

Devs

MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking | Read Paper on Bytez