Picture for Emily Getzen

Emily Getzen

On the Algorithmic Bias of Aligning Large Language Models with RLHF: Preference Collapse and Matching Regularization

Add code
May 26, 2024
Viaarxiv icon