Self-Evaluating LLMs for Multi-Step Tasks: Stepwise Confidence Estimation for Failure Detection

TL;DR

This paper extends self-evaluation methods for large language models to multi-step reasoning tasks, demonstrating that stepwise confidence estimation improves failure detection accuracy over holistic approaches.

Contribution

The work introduces and empirically validates step-by-step confidence estimation techniques for multi-step LLM tasks, enhancing failure detection capabilities.

Findings

Stepwise evaluation outperforms holistic scoring in error detection.

Up to 15% relative increase in AUC-ROC for failure detection.

Self-evaluating LLMs provide meaningful confidence estimates in complex reasoning.

Abstract

Reliability and failure detection of large language models (LLMs) is critical for their deployment in high-stakes, multi-step reasoning tasks. Prior work explores confidence estimation for self-evaluating LLM-scorer systems, with confidence scorers estimating the likelihood of errors in LLM responses. However, most methods focus on single-step outputs and overlook the challenges of multi-step reasoning. In this work, we extend self-evaluation techniques to multi-step tasks, testing two intuitive approaches: holistic scoring and step-by-step scoring. Using two multi-step benchmark datasets, we show that stepwise evaluation generally outperforms holistic scoring in detecting potential errors, with up to 15% relative increase in AUC-ROC. Our findings demonstrate that self-evaluating LLM systems provide meaningful confidence estimates in complex reasoning, improving their trustworthiness and providing a practical framework for failure detection.