Poor Man’s Agentic Modeling: Simulating Large LLM-Agent Societies on a Laptop

“穷人版”智能体建模：在笔记本电脑上模拟大规模大模型智能体社会

Abstract: Simulating societies of many large language model (LLM) agents is expensive, yet the questions asked of such simulations are usually macroscopic: phase behaviour, stylised facts, and scaling with the number of agents $N$, not the cognition of any single agent.

摘要： 模拟由多个大语言模型（LLM）智能体组成的社会成本高昂，但此类模拟通常关注的是宏观问题：如相变行为、典型事实以及随智能体数量 $N$ 的扩展性，而非单个智能体的认知过程。

We turn a statistical-physics observation into a method: replace each LLM agent by a low-parameter model fitted from a few hundred to a few thousand cheap queries, then run the society at any $N$ on a laptop.

我们将统计物理学中的一个观察结果转化为一种方法：用一个低参数模型来替代每个 LLM 智能体，该模型通过几百到几千次低成本查询拟合而成，随后即可在笔记本电脑上以任意 $N$ 值运行该社会模拟。

Whether this works is decided before the simulation runs, chiefly by what each agent perceives. We introduce an [interaction order x memory] taxonomy that maps perception and memory to an effective theory and a predicted $N$-trend of the surrogate error.

这种方法是否有效在模拟运行前即可判定，主要取决于每个智能体的感知内容。我们引入了一种 [交互顺序 x 记忆] 分类法，将感知和记忆映射到有效理论，并预测了代理模型误差随 $N$ 变化的趋势。

We validate it on a faithful reimplementation of the LLM macroeconomy EconAgent and seven further named LLM simulations, with agent decisions cloned from genuine LLM elicitations (primarily DeepSeek) for a few dollars; the predicted error trends hold cell by cell, and the two refuted predictions, both on a strongly saturating response and traced to its curvature, are themselves matched quantitatively by the theory with no free parameters.

我们在 LLM 宏观经济模型 EconAgent 的忠实复现版以及其他七个著名的 LLM 模拟中验证了该方法。智能体的决策通过仅花费几美元从真实 LLM（主要是 DeepSeek）中提取的数据进行克隆；预测的误差趋势在每个单元格中均成立。对于两个被证伪的预测（均涉及强饱和响应且可追溯至其曲率），该理论在没有任何自由参数的情况下，依然能对其进行定量匹配。