Retrieval-augmented generation
A method where a model first fetches relevant documents, then writes its answer using that material.
What it means
Retrieval-augmented generation, usually called RAG, adds a search step before the model writes. The system finds passages related to the question, passes them to the model, and asks for an answer grounded in them. This tends to reduce invented facts compared with answering from memory alone.
RAG is not a guarantee. If the search finds the wrong passages, the answer can be fluent and wrong. Models can also blend two sources or overstate what a source says. Good research tools log which passages were used, so a reviewer can trace each sentence back to its origin. Ask a vendor to show the retrieved passages for one answer, so you can judge the search step as well as the writing. Retrieval quality is often the first thing to check when an answer looks wrong.
Continue this reading path
Cite this entry
https://aipoweredresearchtool.com/glossary/retrieval-augmented-generation/