Look it up, then answer
A large language model only knows what was in its training data, frozen at a cut-off date and blind to your private documents. RAG fixes both problems without retraining: when a question arrives, the system searches a knowledge store for the most relevant passages and pastes them into the prompt. The model then answers from that supplied context — much like an open-book exam.
The retrieval usually runs on a vector database, which finds passages by meaning rather than exact keywords. Done well, RAG cuts down on made-up answers and lets a model cite where its information came from — the backbone of enterprise copilots and document assistants.
RAG vs fine-tuning
Fine-tuning teaches a model new behaviour; RAG gives it new knowledge at run time. For facts that change or are private, retrieval is usually cheaper, fresher and easier to audit.