arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.16048cs.CLcs.AIcs.LG
收藏

采用思维链(CoT)的大语言模型模仿人类推理的可能性有多大?

How Likely Do LLMs with CoT Mimic Human Reasoning?

Zhejiang University(浙江大学);School of Engineering, Westlake University(西湖大学工学院);Institute of Advanced Technology, Westlake Institute for Advanced Study(西湖高等研究院先进技术研究院)

Guangsheng Bao, Hongbo Zhang, Cunxiang Wang, Linyi Yang, Yue Zhang

AI总结:

该研究通过因果分析对比采用CoT的大语言模型与人类的推理过程,发现LLM常偏离理想因果链,上下文学习可强化因果结构,后训练技术会削弱它,且仅扩参无法强化该结构,为改进LLM推理提供了方向。

摘要:

思维链(Chain-of-thought)作为激发大语言模型(LLMs)推理能力的一种颇具前景的技术应运而生。然而,它并非总能提升任务性能,也无法准确表征推理过程,其使用相关的问题仍未得到解决。本文通过对比大语言模型与人类的推理过程来诊断其底层机制,采用因果分析方法探究大语言模型中问题指令、推理与答案之间的关系。实证研究表明,大语言模型时常偏离理想的因果链,导致虚假关联和潜在的一致性错误(推理与答案不一致)。我们还考察了影响因果结构的多种因素,发现带示例的上下文学习可强化该结构,而监督微调、基于人类反馈的强化学习等后训练技术会削弱它。令人意外的是,仅扩大模型规模无法强化因果结构,这亟需新技术的研究。我们希望这项初步研究能为理解和改进大语言模型的推理过程提供启发。

英文摘要:

Chain-of-thought emerges as a promising technique for eliciting reasoning capabilities from Large Language Models (LLMs). However, it does not always improve task performance or accurately represent reasoning processes, leaving unresolved questions about its usage. In this paper, we diagnose the underlying mechanism by comparing the reasoning process of LLMs with humans, using causal analysis to understand the relationships between the problem instruction, reasoning, and the answer in LLMs. Our empirical study reveals that LLMs often deviate from the ideal causal chain, resulting in spurious correlations and potential consistency errors (inconsistent reasoning and answers). We also examine various factors influencing the causal structure, finding that in-context learning with examples strengthens it, while post-training techniques like supervised fine-tuning and reinforcement learning on human feedback weaken it. To our surprise, the causal structure cannot be strengthened by enlarging the model size only, urging research on new techniques. We hope that this preliminary study will shed light on understanding and improving the reasoning process in LLM.

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。

↑