During this session, we chat with Omar Khattab (CS PhD at Stanford) and do a deep dive on his recent paper: “Demonstrate-Search-Predict: Composing retrieval and language models for knowledge-intensive NLP”.
It’s an important work in taking Retrieval Augmented Generation (RAG) to the next level. Existing RAG-pipelines do a one-shot 1) retrieve from vector db, 2) synthesize with LLM, but DSP allows you to build pipelines between a retrieval model and a language model.
We talk about some of these core components, potential benefits beyond existing toolkits (including ours), and future directions.
Add comment