Autonomous agents have long been a prominent research topic in the academiccommunity. Previous research in this field often focuses on training agentswith limited knowledge within isolated environments, which divergessignificantly from the human learning processes, and thus makes the agents hardto achieve human-like decisions. Recently, through the acquisition of vastamounts of web knowledge, large language models (LLMs) have demonstratedremarkable potential in achieving human-level intelligence. This has sparked anupsurge in studies investigating autonomous agents based on LLMs. To harnessthe full potential of LLMs, researchers have devised diverse agentarchitectures tailored to different applications. In this paper, we present acomprehensive survey of these studies, delivering a systematic review of thefield of autonomous agents from a holistic perspective. More specifically, ourfocus lies in the construction of LLM-based agents, for which we propose aunified framework that encompasses a majority of the previous work.Additionally, we provide a summary of the various applications of LLM-based AIagents in the domains of social science, natural science, and engineering.Lastly, we discuss the commonly employed evaluation strategies for LLM-based AIagents. Based on the previous studies, we also present several challenges andfuture directions in this field. To keep track of this field and continuouslyupdate our survey, we maintain a repository for the related references athttps://github.com/Paitesanshi/LLM-Agent-Survey.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)