孙韬-BuildingtheFutureofMulti-agentWorkforce.pdf

1、Building the Future of Multi-agent Workforce孙韬|CAMEL-AI 核心研发工程师孙韬CAMEL-AI 工程师CAMEL AI核心成员，Eigent AI 工程师，是camel和owl两个万星开源项目的核心开发者和维护者，曾任职于百度在线网络技术（北京）有限公司，从事搜索推荐及Agent相关工作。Agent from 1986Agent from 1986Agentsare mindless processesAgentby itself can only do some simple things Joining these agentsin so

2、cietiesleads to true intelligenceWhat magical trick makes us intelligent?The trick is that there is no trick.The power of intelligence stems from our vast diversity,not from any single,perfect principle.Marvin Minsky,The Society of Mind,p.308Agent from 1986Symbolic AgentAgent from 1986Anatomy of Mem

3、oryChains of ReasoningCommunication among AgentsWorld ModelsLanguage Models as AgentsNakano,Reiichiro,et al.Webgpt:Browser-assisted question-answering with human feedback.arXiv preprint arXiv:2112.09332(2021).Language Models as AgentsLilian Weng:https:/lilianweng.github.io/posts/2023-06-23-agent/Key

4、 differences:Language as InputLanguage as OutputStateand Actionare expressed as natural languageGeneralizability.Language Models as Agents in CAMELKey Features:-Memory:Manages chat history and context window-Tools:Supports both internal and external function calls-Step Loop:Handle task require multi

5、ple request with one stepLanguage Models as AgentsThe Rising Trend in the Research Field of LLM-based Multi-AgentsTypology of Applications of LLM-based AgentsLanguage Models as AgentsScaling Laws of Language ModelsScaling Laws of Multi-agent Systems?Kaplan,Jared,et al.Scaling laws for neural languag

6、e models.arXiv preprint arXiv:2001.08361(2020).Building the Future of Multi-agent WorkforceScaling Laws of Multi-agent Systems?Number of Parameters-Number of Agents?Finding the Scaling Laws of Agents Idea Role assignment Task agents Chat agentsCAMEL Role-Playing Framework(The First LLM multi-agent f