EpisodeSim：用经典 AI 脚手架增强 LLM 社交智能体的情节连贯性

AI 摘要

arXiv 论文提出 EpisodeSim，一种混合 LLM 智能体架构，通过经典 AI 脚手架（如世界主控器）维护场景、脚本、义务和结束条件，以增强 LLM 社交智能体的情节连贯性。实验表明，仅靠 LLM 的局部流畅性不足以模拟完整社交场景，而经典 AI 风格的控制层能显著提升模拟质量。该研究强调将经典 AI 的结构化表示与 LLM 的生成能力结合，为社交模拟提供新思路。

正文节选

Classic AI as Scaffolding for LLM Social Agents Abstract Large language models can produce locally plausible social turns, but fluent next-turn generation is not enough for social simulation. Human encounters such as restaurant lunches and hotel check-ins are bounded social episodes with roles, scripts, material state, obligations, commitments, timing, and closure conditions. We present EpisodeSim, a hybrid LLM-agent architecture that represents classic-AI structures as natural-language control

发布时间：2026-09-02 12:00

抓取时间：2026-09-02 12:52

来源机构：arXiv