Fetching the paper…

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