arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2601.00828cs.AI
收藏

分解大语言模型的自我纠正:准确性-纠正悖论与错误深度假说

Decomposing LLM Self-Correction: The Accuracy-Correction Paradox and Error Depth Hypothesis

University of Birmingham(伯明翰大学)

Yin Li

AI总结:

本研究揭示大语言模型在自我纠正中的准确性-纠正悖论,提出错误深度假说,发现更强模型犯更深入的错误,且错误检测与纠正成功率无直接关联。

摘要:

大型语言模型(LLMs)普遍被认为具有自我纠正能力,但近期研究指出,内在自我纠正——即模型在没有外部反馈的情况下纠正自身输出——仍然效果有限。在本工作中,我们系统地将自我纠正分解为三种不同的子能力:错误检测、错误定位和错误纠正。通过在GSM8K-Complex(每模型500个,总计346个错误)上进行跨模型实验,使用三种主要的LLM,我们揭示了一个显著的准确性-纠正悖论:较弱的模型(GPT-3.5,准确率66%)的内在纠正率比更强的模型(DeepSeek,准确率94%)高1.6倍——26.8% vs 16.7%。我们提出了错误深度假说:更强的模型犯的错误更少但更深入,这些错误更难自我纠正。错误检测率在不同架构中变化极大(10%到82%),但检测能力并不预测纠正成功——Claude只能检测10%的错误,但能内在纠正29%。令人惊讶的是,提供错误位置提示会损害所有模型。我们的发现挑战了关于模型能力和自我改进的线性假设,对自我完善管道的设计有重要影响。

英文摘要:

Large Language Models (LLMs) are widely believed to possess self-correction capabilities, yet recent studies suggest that intrinsic self-correction--where models correct their own outputs without external feedback--remains largely ineffective. In this work, we systematically decompose self-correction into three distinct sub-capabilities: error detection, error localization, and error correction. Through cross-model experiments on GSM8K-Complex (n=500 per model, 346 total errors) with three major LLMs, we uncover a striking Accuracy-Correction Paradox: weaker models (GPT-3.5, 66% accuracy) achieve 1.6x higher intrinsic correction rates than stronger models (DeepSeek, 94% accuracy)--26.8% vs 16.7%. We propose the Error Depth Hypothesis: stronger models make fewer but deeper errors that resist self-correction. Error detection rates vary dramatically across architectures (10% to 82%), yet detection capability does not predict correction success--Claude detects only 10% of errors but corrects 29% intrinsically. Surprisingly, providing error location hints hurts all models. Our findings challenge linear assumptions about model capability and self-improvement, with important implications for the design of self-refinement pipelines.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑