What three-letter abbreviation names the technique of retrieving relevant documents from an external store and conditioning a language model's generation on them, extending its knowledge beyond its trained weights?

RAG (Retrieval-Augmented Generation)

Retrieval-Augmented Generation fetches passages from a corpus (often via embedding similarity search) and places them in the model's context so answers can draw on up-to-date or private information the weights never memorized. It reduces hallucination and lets knowledge be updated without retraining.