Confidence Estimation for LLMs in Multi-turn Interactions

TL;DR

This paper systematically studies confidence estimation in multi-turn LLM interactions, introducing new metrics and a paradigm to evaluate calibration and monotonicity, revealing challenges and proposing a promising logit-based probe.

Contribution

It is the first to analyze confidence estimation in multi-turn conversations, proposing a formal framework, novel metrics, and a new probe method for better calibration and evidence tracking.

Findings

Widely-used confidence techniques struggle with calibration in multi-turn dialogues.

The proposed P(Sufficient) probe effectively tracks evidence accumulation.

New metrics like InfoECE provide better evaluation of confidence calibration.

Abstract

While confidence estimation is a promising direction for mitigating hallucinations in Large Language Models (LLMs), current research overwhelmingly focuses on single-turn settings. The dynamics of model confidence in multi-turn conversations, where context accumulates and ambiguity is progressively resolved, remain largely unexplored. This work presents the first systematic study of confidence estimation in multi-turn interactions, establishing a formal evaluation framework grounded in two key desiderata: per-turn calibration and monotonicity of confidence as more information becomes available. To facilitate this, we introduce novel metrics, including a length-normalized Expected Calibration Error (InfoECE), and a new "Hinter-Guesser" paradigm for generating controlled evaluation datasets. Our experiments reveal that widely-used confidence techniques struggle with calibration and monotonicity in multi-turn dialogues. In contrast, a novel logit-based probe we introduce, P(Sufficient), proves comparatively more effective, robustly tracking evidence accumulation and distinguishing it from conversational filler. Our work provides a foundational methodology for developing more reliable and trustworthy conversational agents.