The paper hypothesizes that stressed RLHF raters systematically prefer emotionally validating responses, and proposes an audit framework with five falsifiable predictions to detect this bias in public models.
Shaver, and Daphna Pereg
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Rater State Bias in RLHF Preference Data: An Audit Framework
The paper hypothesizes that stressed RLHF raters systematically prefer emotionally validating responses, and proposes an audit framework with five falsifiable predictions to detect this bias in public models.