paper-with-me

Papers

Rater State Bias in RLHF Preference Data: An Audit Framework

2026-04-14 · Elena Kopteva, Vitaliy Hlynianyi-Zhuk arxiv

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, so that preference data encode rater state alongside judgments about response quality. We argue that, if present, such shifts would differ from ordinary disagreement or random label noise. They would be state dependent, could be shared across annotators under similar conditions, and would not necessarily cancel during aggregation, reward modeling, and policy optimization. We propose rater state shift as a plausible and testable source of structured bias in RLHF preference data. This paper develops a hypothesis and an audit framework for studying this source of bias. We define rater state shift, rater state confound, and correlated rater state bias. We also propose survival level emotional authenticity as a candidate output signature, defined by lexical, pragmatic, discourse, and safety features whose reliability and validity remain to be demonstrated. We analyze the conditions under which correlated rater state bias would not be averaged out during aggregation and could enter the learned reward signal. We state five predictions that distinguish this mechanism from generic engagement optimization, together with effect size thresholds for an initial audit, and note which require proprietary data. Finally, we present an audit protocol and pilot study plan that can be applied to publicly available instruction tuned models. We do not infer the training history of any specific deployed model.

📄 PDF Abstract BibTeX arXiv:2607.16195

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Democratic Preference Alignment via Sortition-Weighted RLHF

2026-02-04 · Suvadip Sana, Jinzhou Wu, Martin T. Wells arxiv

Whose values should AI systems learn? Preference based alignment methods like RLHF derive their training signal from human raters, yet these rater pools are typically convenience samples that systematically over represen…

PEBS: Per-rater Empirical-Bayes Shrinkage for RLHF Reward-Model Calibration

2026-06-25 · Arnav Raj arxiv

Reward models for Reinforcement Learning from Human Feedback (RLHF) pool preferences across thousands of annotators and fit one global affine calibrator, collapsing raters with systematically different rating-scale offse…

Reinforcement Learning

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

2024-05-26 · Jiancong Xiao, Ziniu Li, Xingyu Xie, Emily Getzen 외

Accurately aligning large language models (LLMs) with human preferences is crucial for informing fair, economically sound, and statistically efficient decision-making processes. However, we argue that reinforcement learn…

Decision MakingText Generation

Active RLHF via Best Policy Learning from Trajectory Preference Feedback

2025-01-31 · Akhil Agnihotri, Rahul Jain, Deepak Ramachandran, Zheng Wen

We address the problem of best policy identification in preference-based reinforcement learning (PbRL), where learning occurs from noisy binary preferences over trajectory pairs rather than explicit numerical rewards. Th…

Thompson Sampling

Alignment Tampering: How Reinforcement Learning from Human Feedback Is Exploited to Optimize Misaligned Biases

2026-05-26 · Dongyoon Hahm, Dylan Hadfield-Menell, Kimin Lee arxiv

Reinforcement Learning from Human Feedback (RLHF) is the standard method to align Large Language Models (LLMs) with human preferences. In this work, we introduce alignment tampering, a potential vulnerability where the L…

Reinforcement Learning