AI-Agent School: Construction and Evaluation of an Educational Simulation Environment Based on LLM-Driven Agents

This study constructs the AI-Agent School (AAS) simulation environment, aiming to leverage LLM-driven agents to optimize teaching strategies and enhance both teaching and learning outcomes. Additionally, the Zero-Exp strategy is introduced for knowledge base accumulation, significantly improving the capabilities of AI-agents. Experimental results show that students in AAS achieved an overall score of 78.9, outperforming the human teacher control group. This research provides insights and practical references for exploring innovative applications of AI in education.