Learning to Check: Unleashing Potentials for Self-Correction in Large Language Models
This addresses the challenge of improving reasoning accuracy in LLMs for applications requiring logical consistency, though it is incremental as it builds on existing self-correction and Chain of Thought methods.
The paper tackles the problem of limited self-correction in large language models for reasoning tasks by enhancing their self-checking capabilities through fine-tuning with a specialized 'Step CoT Check' format, resulting in significant improvements across multiple benchmarks, especially in locating incorrect positions with greater benefits in larger models.
Self-correction has achieved impressive results in enhancing the style and security of the generated output from large language models (LLMs). However, recent studies suggest that self-correction might be limited or even counterproductive in reasoning tasks due to LLMs' difficulties in identifying logical mistakes. In this paper, we aim to enhance the self-checking capabilities of LLMs by constructing training data for checking tasks. Specifically, we apply the Chain of Thought (CoT) methodology to self-checking tasks, utilizing fine-grained step-level analyses and explanations to assess the correctness of reasoning paths. We propose a specialized checking format called "Step CoT Check". Following this format, we construct a checking-correction dataset that includes detailed step-by-step analysis and checking. Then we fine-tune LLMs to enhance their error detection and correction abilities. Our experiments demonstrate that fine-tuning with the "Step CoT Check" format significantly improves the self-checking and self-correction abilities of LLMs across multiple benchmarks. This approach outperforms other formats, especially in locating the incorrect position, with greater benefits observed in larger models. For reproducibility, all the datasets and code are provided in https://github.com/bammt/Learn-to-check.