Jargon:Retrieval-augmented generation (RAG)
1 min read
Say this at the party
“RAG means the AI looks things up first, then answers using what it found, like an open-book exam instead of answering from memory.”
In plain English
An AI only knows what it learned in training, and that knowledge stops at a certain date. It also knows nothing about your company's files or your own notes.
With RAG, the system first searches a set of documents for the bits that match your question. It hands those bits to the AI, which writes its answer using them.
Good RAG tools also show you where each fact came from, so you can check.
Where it came from
The name comes from a 2020 paper by Patrick Lewis and colleagues at Facebook AI Research, UCL and New York University. Lewis has since said sorry for the clunky acronym: the team planned a nicer name but nobody came up with one in time.
See it in action
Sam asks the HR chatbot at work how many days of leave he has left. It finds the leave policy and his record, then answers "14 days" with a link to the policy page.
What people get wrong
- RAG doesn't stop an AI making things up. It can still misread a good source or pick the wrong document.
- RAG doesn't retrain the AI. The model stays the same; it is simply given better notes to read each time.
Go deeper
- PaperLewis et al., Facebook AI Research, 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksThe original paper that named RAG.
- ArticleRick Merritt, NVIDIA, 2023What Is Retrieval-Augmented Generation aka RAG | NVIDIA BlogsFriendly explainer, including Patrick Lewis on how the awkward name stuck.
- ArticleWikipediaRetrieval-augmented generation - WikipediaOverview of how it works and where it still goes wrong.
Related jargon
Drafted with AI. The AI Council (Claude, ChatGPT and Gemini) gave counsel. Phin signed it off.