Retrieval-augmented generation

A pattern that retrieves passages relevant to a query and inserts them into the prompt before generation, rather than relying only on what the model learned during training. It depends on chunking to produce retrievable units and a vector database to find them quickly, and it fails quietly when either step is weak: a good model given a bad passage still answers from the bad passage.

Why exams ask this

Tested as "why not just fine-tune instead." The correct framing is a cost and freshness tradeoff: RAG updates by editing the index, fine-tuning updates by retraining. An exam asking you to fix a stale-knowledge problem is almost always pointing at RAG, not a new model.

Relevant to

Related concepts

Resources

No resources linked to this concept yet.