毛绍光-策略性推理与AI多智能体系统.pdf

《毛绍光-策略性推理与AI多智能体系统.pdf》由会员分享，可在线阅读，更多相关《毛绍光-策略性推理与AI多智能体系统.pdf（39页珍藏版）》请在三个皮匠报告上搜索。

1、毛绍光 微软亚洲研究院资深研发工程师 2019年加入微软，通用人工智能组资深研发工程师，主要研究方向为基于语言模型的推理，AI智能体及多智能体系统。开发的相关技术被应用于微软M365 Word等产品。演讲主题：策略性推理与AI多智能体系统Strategic Reasoning in Multi-Agent SystemShaoguang MaoGeneral AI GroupMSRAL1:ChatbotsChatGPT/GPT4/GPT4oL2:Reasonero1L3:AgentL4:InnovatorL5:OrganizerOpenAI AGI RoadmapAgent System an

2、dMulti-AgentSystemMulti-agent simply means having several agents working together to solve your task instead of only one.It empirically yields better performance on most benchmarks.The reason for this better performance is conceptually simple:for many tasks,rather than using a do-it-all system,you w

3、ould prefer to specialize units on sub-tasks.Here,having agents with separate tool sets and memories allows to achieve efficient specialization.A Task-Solving Agent through Multi-Persona Self-CollaborationSolo Performance Prompting(SPP)instructs a LLM to perform the following the procedure for gener

4、al task-solving:(1)Persona Identification:Identify participants with special personas that are essentialfor solving the task.(2)Brainstorming:The participants share knowledge and providesuggestions on how to approach the task based on their own expertise.(3)Multi-Persona Iterative Collaboration:The

5、leader persona,AI Assistant,proposes initial solutions,consults the other participants for feedback,and revise the answer iteratively.1 Zhenhailong Wang,Shaoguang Mao,Wenshan Wu,Tao Ge,Furu Wei,Heng Ji,Unleashing the emergent cognitive synergy in large language models:A task-solving agent through mu

6、lti-persona self-collaboration,NAACL 2024A Task-Solving Agent through Multi-Persona Self-CollaborationSPP effectively improves both knowledge and reasoning abilities in LLMs.How to make a multi-agent system internally communicate/collaborate more efficiently?-organizer We approach this fundamental r