What does Retrieval-Augmented Generation (RAG) do?

Retrieval-Augmented Generation searches an external knowledge source, such as a document database or the web, for passages relevant to a user's question, then feeds those passages into the model alongside the question so it answers grounded in current information. This addresses a core weakness of large language models: their knowledge is frozen at whatever point training ended, so without retrieval they can't know what happened afterward or access private documents they were never trained on.

Retraining the whole model on every question would be enormously expensive, essentially rebuilding the AI from scratch each time, which isn't how RAG works. Deleting wrong answers after they're shown doesn't address outdated knowledge; it's reactive cleanup, not better answers upfront.

RAG has become popular for building AI assistants tailored to specific companies, letting a general-purpose model answer questions about internal documents without the enormous cost of training a custom model from scratch.