Retrieve Only When the Question Needs Fresh Evidence
Explain why adaptive retrieval can improve quality and latency compared with retrieving a fixed number of passages for every prompt.
The routing move: retrieval is a decision, not a default. Self-RAG popularized an important idea for production teams: the system can learn or be designed to retrieve on demand and critique its own use of evidence. Even without training a Self-RAG model, the design principle transfers. A RAG app should know when it needs the corpus, when it can answer from stable general knowledge, and when it should ask a clarifying question. Why fixed top-k breaks A fixed top-k approach adds noise to easy questions and underserves risky questions. More chunks can crowd out the controlling source, and irrelevant context…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in