arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2409.10038cs.CLcs.AIcs.LG
收藏

思维图式图示

On the Diagram of Thought

IIIS Tsinghua University(清华大学人工智能研究院);Shanghai Qi Zhi Institute(上海启智研究院)

Yifan Zhang, Yang Yuan, Andrew Chi-Chih Yao

AI总结:

本文提出思维图式(DoT)框架,使LLM能构建动态推理图谱,通过类型推理记录和范畴论数学框架,实现可审计的分步推理,区分类型子轨迹与自然语言文本的语义保障。

摘要:

大型语言模型(LLMs)在许多任务上表现出色,但在需要结构化、多步骤推理的复杂问题上往往表现不佳。我们介绍了思维图式(DoT),一个框架,使单个LLM能够构建并导航其推理的思维图。与其线性思考不同,模型构建动态的想法图,可以提出不同的思路、批判自身的步骤,并将验证的洞察综合为最终结论。这一过程是轻控制器的:它不需要外部搜索算法或规划器,但确实使用确定性在线验证器来验证语法约束的类型轨迹、寄存器约束以及可选的求解器检查。为了明确此过程的可靠性目标,我们将DoT建立在范畴论的数学框架上。我们将接受的类型推理记录解释为切片拓扑中的图示,并将选定提案子图示的综合建模为有限极限。在谓词片段中,同一对象等价于在相反信息序中的反变余限。由此产生的形式化方法为LLM的类型推理提供了可审计、分步骤的轨迹,并将类型子轨迹的语义保证与不受约束的自然语言文本和未经认证的操作边缘区分开来。

英文摘要:

Large Language Models (LLMs) excel at many tasks but often falter on complex problems that require structured, multi-step reasoning. We introduce the Diagram of Thought (DoT), a framework that enables a single LLM to build and navigate a mental map of its reasoning. Instead of thinking in a straight line, the model constructs a dynamic diagram of ideas, where it can propose different lines of thought, critique its own steps, and synthesize validated insights into a final conclusion. This process is controller-light: it does not require an external search algorithm or planner, but it does use a deterministic online validator for grammar-constrained typed traces, register constraints, and optional solver checks. To clarify the reliability target of this process, we ground DoT in a mathematical framework from category theory. We interpret accepted typed reasoning records as diagrams in a slice topos and model synthesis of the selected proposer subdiagram as a finite limit. In the predicate fragment, this same object is equivalently a variance-reversed colimit in the opposite information order. The resulting formalism gives an auditable, step-by-step trace of the LLM's typed reasoning and separates semantic guarantees for the typed subtrace from unconstrained natural-language text and uncertified operational edges.

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。

↑