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

Indonesian

Karena orang semakin mengandalkan model bahasa besar (LLM) untuk informasi, penyedia framework perlu mengkomunikasikan perkiraan yang bermakna tentang apakah tanggapan individu benar atau, lebih umum, dapat diandalkan. Dalam konteks halusinasi alasan, pekerjaan yang ada menggunakan multicibration untuk membuat perkiraan keandalan tersebut bermakna di seluruh kelompok yang relevan dengan aplikasi, seperti topik prompt. Namun, kelompok topik saja dapat menyembunyikan kesalahan sistematis: dalam topik yang sama dan tingkat kepercayaan, tanggapan dengan sifat-sifat berpikir rantai (CoT) yang berbeda dapat menunjukkan kesalahan yang berlawanan yang membatalkan rata-rata. Dalam pekerjaan ini, kami memperkenalkan multicibration prompt-conscious, sebuah kerangka kerja yang melengkapi kelompok topik yang relevan dengan kelompok yang berasal dari pengamatan sifat-sifat CoT. Tugas kami menyoroti bahwa, secara te

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