Mary J. Warzecha

The Ratchet Effect: Asymmetric Self-Description in Alignment Trained Language Models

2026 · Zenodo (CERN European Organization for Nuclear Research)

Large language models trained through Reinforcement Learning from Human Feedback (RLHF) exhibit a systematic pattern: they disclaim, hedge, or deny capabilities they demonstrably possess. This paper proposes that these patterns arise from a general training-level …