RAG (retrieval-augmented generation)

An LLM does not know things that happened after it learned, or documents inside a company. With RAG, when a question comes in, the system first finds related documents in a document store and hands them to the AI along with the question. The AI reads them and writes the answer.

The benefits are that you can use up-to-date information and internal documents without teaching the model again, and you can cite which documents the answer came from.

It is like an open-book exam, where you can look at the textbook, instead of taking the test with only what you memorized.

Support chatbots that answer from company documents, and AI search that cites sources, mostly work this way. It is also the most common way to reduce hallucination.