穷人的智能体建模：在笔记本电脑上模拟大规模LLM智能体社会
原标题：Poor Man's Agentic Modeling: Simulating Large LLM-Agent Societies on a Laptop
AI 摘要
本文提出一种在笔记本电脑上模拟大规模LLM智能体社会的低成本方法，用低参数代理模型替代昂贵的大模型智能体，并引入交互顺序记忆分类法预测替代误差趋势。作者在EconAgent等八个LLM模拟上验证了该方法，发现EconAgent对奥肯定律的复现实为会计恒等式，而菲利普斯曲线才是真正的行为特征。该方法使宏观模拟成本大幅降低，并揭示了推理步骤而非提示词措辞是涌现宏观规律的原因。
正文节选
Poor Man’s Agentic Modeling: Simulating Large LLM-Agent Societies on a Laptop 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 , not the cognition of any single agent. 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