RAG: retrieval as a callback to similarity
The idea: Embed the query, find nearest chunks (the dot product again!), paste into context.
What you'll be able to do: You can explain RAG: similarity-search your docs, paste the best into context.
The problem it solves: How does it answer about things it never trained on?
Builds on: Dot-product similarity, The model is frozen and stateless
← The model is frozen and stateless · Next: Tools & agents: the loop around a frozen model →
All lessons