Quantifying Faithful Confidence Expression in Large Reasoning Models

Forum topic · 小凯 · 2026-06-04

Summary

This paper introduces a novel framework for systematically quantifying faithful calibration (FC) in large reasoning models (LRMs). FC refers to the alignment between a model's intrinsic confidence and its linguistically expressed confidence—a persistent failure mode that undermines LLM trustworthiness. The challenge is particularly acute for LRMs, whose extended chain-of-thought reasoning traces are often interpreted by users as evidence of deliberation, competence, and confidence. Existing FC measurement paradigms do not generalize well to long reasoning outputs, which lack clear step boundaries, exhibit inconsistent step structure, and encode complex conditional dependencies, complicating intrinsic confidence estimation. The authors—Areeb Gani, Asal Meskin, Gabrielle Kaili-May Liu, and Arman Cohan (Yale-affiliated NLP researchers)—propose an approach designed to handle these long-form reasoning traces. The paper is available on arXiv (2606.03969).

Paper Overview

Research Area: NLP
Authors: Areeb Gani, Asal Meskin, Gabrielle Kaili-May Liu, Arman Cohan
Published: 2026-06-02
arXiv: 2606.03969

Background

Reliable uncertainty communication is critical to the trustworthiness of LLMs, yet faithful calibration (FC)—the alignment between models' intrinsic and (linguistically) expressed confidence—is a persistent failure mode. This challenge is key for large reasoning models (LRMs), whose extended reasoning traces are often interpreted by users as evidence of deliberation, competence, and confidence.

Despite the importance of FC and the wide usage of LRMs, the extent to which LRMs can faithfully express their confidence remains poorly understood. Moreover, the prevailing paradigm for measuring FC does not generalize well to the long chain-of-thought outputs generated by LRMs, which tend to lack clear step boundaries, involve inconsistent step structure, and encode complex conditional dependencies throughout the trace—all of which complicate intrinsic confidence estimation.

Contribution

To address this challenge, the authors introduce a novel framework to systematically quantify FC in LRMs.