leogao comments on Scaling Laws for Reward Model Overoptimization

[leogao]

There’s an example in the appendix but we didn’t do a lot of qualitative analysis.