RCOT: Detecting and Rectifying Factual Inconsistency in Reasoning by Reversing Chain-of-Thought

TL;DR

This paper introduces RCoT, a novel method that enhances LLMs' reasoning by automatically detecting and correcting factual inconsistencies in generated solutions through problem reconstruction and fine-grained feedback.

Contribution

RCoT is the first approach to automatically identify and rectify factual errors in LLM reasoning by reconstructing problems and providing detailed feedback for correction.

Findings

RCoT outperforms standard CoT, Self-Consistency, and Self-Refine on seven arithmetic datasets.

Manually written fine-grained feedback significantly boosts LLM reasoning accuracy.

ChatGPT achieves 94.6% accuracy on GSM8K with fine-grained feedback.

Abstract

Large language Models (LLMs) have achieved promising performance on arithmetic reasoning tasks by incorporating step-by-step chain-of-thought (CoT) prompting. However, LLMs face challenges in maintaining factual consistency during reasoning, exhibiting tendencies to condition overlooking, question misinterpretation, and condition hallucination over given problems. Existing methods use coarse-grained feedback (e.g., whether the answer is correct) to improve factual consistency. In this work, we propose RCoT (Reversing Chain-of-Thought), a novel method to improve LLMs' reasoning abilities by automatically detecting and rectifying factual inconsistency in LLMs, generated solutions. To detect factual inconsistency, RCoT first asks LLMs to reconstruct the problem based on generated solutions. Then fine-grained comparisons between the original problem and the reconstructed problem expose the factual inconsistency in the original solutions. To rectify the solution, RCoT formulates detected factual inconsistency into fine-grained feedback to guide LLMs in revising solutions. Experimental results demonstrate improvements of RCoT over standard CoT, Self-Consistency and Self-Refine across seven arithmetic datasets. Moreover, we find that manually written fine-grained feedback can dramatically improve LLMs' reasoning abilities (e.g., ChatGPT reaches 94.6% accuracy on GSM8K), encouraging the community to further explore the fine-grained feedback generation methods.