Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer

Open in new window

The document is on a web server that cannot be reached. Please try again later.