RAG (retrieval-augmented generation)
Definition: RAG (Retrieval-Augmented Generation) feeds the model relevant documents retrieved on the fly, so it answers with up-to-date facts and cites sources.
It reduces hallucinations and lets you query a private knowledge base. Claude News' semantic search is built on this principle.
Frequently asked questions
What is RAG used for?
To let a model answer about fresh or private data it never saw in training: you feed it the relevant documents at question time. This reduces hallucinations and lets it cite sources.
RAG or fine-tuning?
RAG adds knowledge on the fly without retraining the model; fine-tuning changes the model for a style or behavior. For facts that change, RAG is almost always preferable and cheaper.
Does RAG eliminate hallucinations?
It sharply reduces them by grounding answers in real documents, but doesn't remove them: the model can misread a passage. Citing and checking sources remains essential.