Describe the retrieve-then-generate mechanism in RAG and identify why retrieval quality sets the ceiling.
The principle: retrieval sets the ceiling for generation. RAG combines a generator with an external memory. The generator is the language model that writes. The external memory is the searchable corpus, usually represented through chunks, embeddings, indexes, and metadata. The answer depends on both. The benefit is operational. You can update the knowledge base when policies change instead of retraining the model. You can also inspect what the system retrieved, which gives reviewers a way to understand why an answer was produced. The risk is equally operational. If the corpus contains stale files, duplicate drafts, broken chunks, or missing metadata,…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in