Newest papers

Reasoning-Aware Multicalibrated Reliability Predictors for Language Models

Ekaterina Fedorova, Alessandro Morosini, Chara Podimata · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26645 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

A medida que las personas confían cada vez más en los modelos de lenguaje grande (LLM) para obtener información, los proveedores de modelos necesitan comunicar estimaciones significativas de si las respuestas individuales son correctas o, más generalmente, fiables. En el contexto de las alucinaciones de la razón, el trabajo existente utiliza la multicibración para hacer que tales predicciones de fiabilidad sean significativas en grupos relevantes para la aplicación, como los temas prompt. Sin embargo, los grupos temáticos pueden ocultar errores sistemáticos: dentro del mismo tema y nivel de confianza, las respuestas con diferentes propiedades de cadena de pensamiento (CoT) pueden mostrar errores opuestos que cancelan la media. En este trabajo, introducimos la multicibración prominente, un marco que complementa los grupos temáticos relevantes para la aplicación con grupos derivados

Machine translation (Meta M2M100); the English original follows.

Abstract

As people increasingly rely on large language models (LLMs) for information, model providers need to communicate meaningful estimates of whether individual responses are correct or, more generally, reliable. In the context of hallucinations, existing work uses multicalibration to make such reliability predictions meaningful across application-relevant groups, such as prompt topics. Topic groups alone, however, can conceal systematic errors: within the same topic and confidence level, responses with different chain-of-thought (CoT) properties may exhibit opposing errors that cancel on average. In this work, we introduce reasoning-aware multicalibration, a framework that supplements application-relevant topic groups with groups derived from observable properties of CoT traces. We highlight that, theoretically, groupings based on CoT are "internally important", i.e., groupings for which the multicalibration guarantee is not societally crucial as it is for high-risk topics, but whose inclusion in the multicalibration procedure could easily and significantly improve predictive accuracy of the final reliability predictor. Empirically, we apply our framework to hallucination prediction and unsafe-response detection across a variety of benchmarks, and find that adding CoT groups to topic-based calibration reduces Brier score in both tasks. Interestingly, we find that multicalibrating with respect to CoT groups alone (without any topic information) already improves calibration across topics, suggesting that the predictor's errors depend on how the model reasoned, rather than what it was asked about in the prompt.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Reliability (semiconductor), Variety (cybernetics), Computer science, Context (archaeology), Calibration, Machine learning

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange