Paper MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking · 2025 · arXiv:2501.13011 Read the paper → reward-hackingrlhfcontrol
