CritiCal: Can Critique Help LLM Uncertainty or Confidence Calibration?

TL;DR

This paper introduces CritiCal, a novel training method using natural language critiques to improve confidence calibration in Large Language Models, especially in high-stakes scenarios, outperforming existing approaches and enhancing reliability.

Contribution

It proposes CritiCal, a new critique-based calibration training method that leverages natural language critiques to enhance LLM confidence calibration beyond traditional techniques.

Findings

CritiCal outperforms Self-Critique and baselines in calibration accuracy.

CritiCal surpasses GPT-4o in complex reasoning tasks.

The method generalizes well to out-of-distribution data.

Abstract

Accurate confidence calibration in Large Language Models (LLMs) is critical for safe use in high-stakes domains, where clear verbalized confidence enhances user trust. Traditional methods that mimic reference confidence expressions often fail to capture the reasoning needed for accurate confidence assessment. We propose natural language critiques as a solution, ideally suited for confidence calibration, as precise gold confidence labels are hard to obtain and often require multiple generations. This paper studies how natural language critiques can enhance verbalized confidence, addressing: (1) What to critique: uncertainty (question-focused) or confidence (answer-specific)? Analysis shows confidence suits multiple-choice tasks, while uncertainty excels in open-ended scenarios. (2) How to critique: self-critique or critique calibration training? We propose Self-Critique, enabling LLMs to critique and optimize their confidence beyond mere accuracy, and CritiCal, a novel Critique Calibration training method that leverages natural language critiques to improve confidence calibration, moving beyond direct numerical optimization. Experiments show that CritiCal significantly outperforms Self-Critique and other competitive baselines, even surpassing its teacher model, GPT-4o, in complex reasoning tasks. CritiCal also shows robust generalization in out-of-distribution settings, advancing LLM's reliability.