Smallville sandbox game environment

该数据集是一个广为人知的生成式多代理系统（MASs）环境，它使得基于大型语言模型（LLM）的代理能够展现出类似人类的行为，包括观察、互动、规划和执行动作。该数据集包含了代理描述、个人规范以及实验过程中出现的社交规范动态。规模上，实验中使用了10个生成式代理（其中3个是规范创业者，7个是普通代理）。研究任务在于探究社交规范的涌现及其对环境中代理互动的影响。

This dataset is a well-known generative multi-agent system (MASs) environment that enables large language model (LLM)-based agents to demonstrate human-like behaviors including observation, interaction, planning, and action execution. It encompasses agent descriptions, personal norms, and the dynamics of social norms that arise throughout the experimental process. A total of 10 generative agents were utilized in the experiments, among which 3 function as norm entrepreneurs and 7 as regular agents. The core research objective of this dataset is to investigate the emergence of social norms and their impacts on agent interactions within the environment.