Rater State Bias in RLHF Preference Data: An Audit Framework
We identify a structured confound in Reinforcement Learning from Human Feedback (RLHF). Pairwise preference labels are intended to reflect the compared outputs, but they may also reflect the rater's state during annotation. Under sustained stressful or distressing conditions, raters' preferences may shift over time. As a result, preference data can encode rater state alongside judgments about response quality. These shifts differ from ordinary disa
By Elena Kopteva, Vitaliy Hlynianyi-Zhuk