AgentSims is an open-ended, customizable sandbox evaluation infrastructure that allows for the creation of limitless AI agents and evaluates their capabilities through task settings. AgentSims achieves highly customizable AI character behavior decisions through planning, memory, and tool usage systems. Compared to Stanford's AI Town, AgentSims offers a more free-form interactive interface, making research more convenient.
Multi-school Joint Open Source Domestic AI Town, a Virtual World Editable by Clicking the Interactive Interface
量子位
20
© Copyright AIbase Base 2024, Click to View Source - https://www.aibase.com/news/660