Search papers, labs, and topics across Lattice.
This paper addresses the challenge of mitigating demographic bias in large language model (LLM) reasoning by formulating the intervention process as an online change-point detection problem. By utilizing a cumulative sum (CUSUM) statistic to determine when to intervene based on bias signals, the authors demonstrate that their adaptive triggering method can significantly improve accuracy in ambiguous contexts while minimizing unnecessary corrections. The results show that this approach recovers much of the accuracy lost under fixed-interval interventions and performs well across various model types, highlighting the effectiveness of both white-box and black-box signals in bias correction.
Adaptive triggering can recover lost accuracy in LLM reasoning while cutting down on unnecessary interventions, challenging traditional fixed-interval approaches.
Chain-of-thought prompting can expose and amplify demographic stereotypes within an LLM's intermediate reasoning and create a failure mode that final-answer debiasing alone cannot address. Mitigating such bias during generation presents a fundamental timing problem: intervening too late allows biased reasoning to propagate, while unnecessarily intervening can disrupt otherwise correct reasoning. Existing approaches largely avoid this decision by either evaluating completed reasoning chains post hoc or intervening at predetermined steps, leaving open when a developing reasoning trajectory provides sufficient evidence to warrant correction. We formulate this decision as an online change-point detection problem. A per-step bias signal updates a CUSUM statistic and a targeted correction is injected only when accumulated evidence crosses a detector-specific threshold calibrated on held-out data. We instantiate the framework with a white-box signal derived from next-token probabilities and a black-box signal obtained from an LLM judge, enabling deployment with both open-weight and hosted models. On gpt-4o-mini adaptive black-box triggering recovers most of the disambiguated-context accuracy lost under fixed-interval intervention while requiring substantially fewer interventions. That result holds even with an independent judge. Across six open-weight models, the white-box signal improves ambiguous-item accuracy on all six but reduces disambiguated-item accuracy on five because it cannot distinguish unsupported stereotype reliance from correct, stereotype-congruent evidence.
