Retrieval Augmented Generation with Agentic System
Task Overview
RAG
RAG(Retrieval Augmented Generation) 检索增强生成
A method that allows LLMs to answer the queries with external knowledge.
- 将问题丢到检索系统或者知识库得到相关信息
- 将问题和检索的信息一起放到LLM中,让回答可以使用额外的信息
Why RAG
Knowledge cutoff: LLM 不能预知未来,只能获得训练时间发生之前的信息
Reducing the cost of training: fine-tune需要花销,RAG不需要训练
Improving the reliability of generated answers: LLM本质是文字接龙,会产生胡说八道(hallucinations)的情况,有参考资料之后不太会乱讲
Agentic System
An agentic System is a framework in which LLMs act as individuals and collaborate to complete complex tasks.
正常的LLM就是跟人类一问一答,需要人类来告诉他下一步需要做什么,但是Agentic System可以让LLM门相互合作,一个LLM来帮助另外一个LLM,不完全需要人类的指导。