The host initially viewed "Agents" as an abstraction over multiple pipelines, such as an "Email Agent". However, Davis suggests viewing "Agents" as the tools connected to each LLM call, like `read_email` or `calculator`. The host encourages feedback on this perspective, believing a consensus on the definition of "Agents" could alleviate confusion for those working with LLMs/Generative AI.
Key takeaways:
- The podcast covers understanding of RAG systems through four key areas: Ingest / ETL, Search, Generate / Agents, and Evaluation.
- Various topics were discussed including Knowledge Graph RAG, Function Calling and Tool Selection, Re-ranking, Quantization, and more.
- The discussion led to a re-evaluation of the definition of 'Agents' in the RAG stack, with a suggestion to view 'Agents' as scoping the tools each LLM call is connected to.
- The podcast aims to clarify the current confusion for people building with LLMs / Generative AI by getting a consensus definition of 'Agents'.