Reproduce “AgentSchool: An LLM-Powered Multi-Agent Simulation for Education” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yulei Ye, Wenhao Li, Zhong Wen, Yunshu Huang, Yichen Hu, Zifan Wei, Yige Wang, Xinyu Xie and 18 more
AgentSchool is an LLM-powered multi-agent simulation platform for education, which models learning as a state-transition process with cognitively growable student agents and adaptive teacher agents. The system provides an inspectable "computational wind tunnel" for evaluating AI interventions and educational scenarios by simulating detailed learning dynamics and social interactions in a controlled environment.
Key insights
- Modeling learning as a continuous state-transition process with explicit internal states enables more diagnostically rich and pedagogically interpretable simulation traces compared to baseline role-play approaches.
- The effectiveness of adaptive teacher agents and the richness of student learning dynamics are significantly influenced by the underlying LLM backbone, highlighting the need for multi-backbone evaluation and calibration.
- The simulation platform can generate inspectable traces of informal educational dynamics such as isolation, cohesion, clique formation, and peer influence, making it suitable for hypothesis generation in classroom management and pedagogy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.