让大模型分步思考，精准定位错误并自动修正
Structure Enables Effective Self-Localization of Errors in LLMs
- 将推理拆成离散、语义连贯的思维步骤，便于定位错误
- 在有外部验证时，纠错率提升20%-40%；无外部验证也优于现有方法
- 适合追求高可靠性的AI系统，如自主决策与复杂推理
大语言模型的自纠正能力仍不成熟。本文探索模型是否能显式定位错误推理中的错误，以构建可自我修正的AI系统。我们提出一种结构化提示方法，将推理过程分解为离散且语义连贯的思维步骤，发现相比传统非结构化思维链，模型在此结构中能更可靠地定位错误。受人类大脑在离散决策点监控错误并重新采样的启发，我们提出迭代思维重采样（Thought-ICS）框架：模型逐个生成完整、独立的思维步骤，每一步代表一次明确决策，形成自然的错误定位边界。验证后，模型可精确定位首个错误步骤，并回溯至最后一个正确点生成替代推理。当使用权威验证判断推理错误时，Thought-ICS实现20%-40%的自纠正提升；在完全自主设置下，其性能亦超越当前主流自纠正基线。
原文摘要 · Abstract (English)
Self-correction in language models remains elusive. In this work, we explore whether language models can explicitly localize errors in incorrect reasoning, as a path toward building AI systems that can effectively correct themselves. We introduce a prompting method that structures reasoning as discrete, semantically coherent thought steps, and show that models can localize errors more reliably within this structure than in conventional, unstructured chain-of-thought reasoning. Motivated by how the human brain monitors errors at discrete decision points and resamples alternatives, we introduce Iterative Correction Sampling of Thoughts (Thought-ICS), a self-correction framework. Thought-ICS iteratively prompts the model to generate reasoning one discrete and complete thought at a time--where each thought represents a deliberate decision by the model--creating natural boundaries for precise error localization. Upon verification, the model localizes the first erroneous step, and the system backtracks to generate alternative reasoning from the last correct point. When asked to correct reasoning verified as incorrect by an oracle, Thought-ICS achieves 20-40% self-correction lift. In a completely autonomous setting without external verification, it outperforms contemporary self-correction baselines.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。