Fetching the paper…

Overcoming Reward Overoptimization via Adversarial Policy Optimization with Lightweight Uncertainty Estimation · Around