Towards Reliable Alignment: Uncertainty-aware RLHF

Debangshu Banerjee, Aditya Gopalan

Recent advances in aligning Large Language Models with human preferences have benefited from larger reward models and better preference data. However, most of these methodologies r…