详细平衡：LLM 驱动的智能体生成动力学研究
Detailed balance in large language model-driven agents
AI 导读
一项 arXiv 研究提出基于最小作用原理的方法，估计嵌入智能体中的 LLM 生成方向性。通过实验测量 LLM 生成状态间的转移概率，研究者发现证据与详细平衡倾向一致，表明 LLM 生成可能并非通过学习规则和策略，而是隐式学习一类可跨越不同 LLM 架构与提示词模板的势函数。
正文 · 原文
Abstract:Large language model (LLM)-driven agents are emerging as a powerful new paradigm for solving complex problems. Despite the empirical success of these practices, a theoretical framework to understand and unify their macroscopic dynamics remains lacking. This paper proposes a method based on the least action principle to estimate the underlying generative directionality of LLMs embedded within agents. By experimentally measuring the transition probabilities between LLM-generated states, we report evidence consistent with a tendency toward detailed balance in LLM-generated transitions, indicating that LLM generation may not be achieved by generally learning rule sets and strategies, but rather by implicitly learning a class of underlying potential functions that may transcend different LLM architectures and prompt templates. To our knowledge, this provides an effective macroscopic description of LLM generative dynamics that does not depend on specific model details. This work is an attempt to establish a macroscopic dynamics theory of complex AI systems, aiming to elevate the study of AI agents from a collection of engineering practices to a science built on effective measurements that are predictable and quantifiable.
| Comments: | 24 pages, 19 figures, 7 tables |
| Subjects: | Machine Learning (cs.LG); Statistical Mechanics (cond-mat.stat-mech); Artificial Intelligence (cs.AI); Adaptation and Self-Organizing Systems (nlin.AO); Data Analysis, Statistics and Probability (physics.data-an) |
| Cite as: | arXiv:2512.10047 [cs.LG] |
| (or arXiv:2512.10047v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2512.10047 arXiv-issued DOI via DataCite |
|
| Related DOI: | https://doi.org/10.1103/kwx6-clqd
DOI(s) linking to related resources |
Submission history
From: Zhuo-Yang Song [view email]
[v1]
Wed, 10 Dec 2025 20:04:23 UTC (409 KB)
[v2]
Tue, 27 Oct 2026 17:34:10 UTC (796 KB)
来源：arXiv cs.AI · arxiv.org