Yang Gao

RRM: Robust Reward Model Training Mitigates Reward Hacking

Impact of Preference Noise on the Alignment Performance of Generative Language Models

Best-of-Venom: Attacking RLHF by Injecting Poisoned Preference Data