LANDMARK · 8 MIN · UNDERSTAND
Retrieval-augmented generation
After this landmark, you can describe the stages of a retrieval system, and diagnose which stage is responsible when the answers are wrong.
↗
This concept is shared, but the Everyday lens is less central here.
You can still explore it. We’re showing the shared explanation and a related practical view without hiding the knowledge.
Retrieval-augmented generation puts a search step in front of the model. Documents are split into chunks and indexed; a question is used to retrieve the most relevant chunks; those chunks go into the context window with an instruction to answer from them and cite them. This is how an assistant answers about your handbook, last week’s incidents, or a document written after the model’s cutoff. It is also the main practical control on fabrication, because the model is answering from text in front of it rather than from memory. What it is not is a guarantee. Retrieval can fetch the wrong passage, the right passage in a stale version, or nothing useful at all — and a model handed irrelevant context will often still produce a confident answer, sometimes citing the passage it did not actually use. Grounding improves the odds substantially; it does not remove the need to check.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
This is what an assistant with web search is doing, and why it can discuss this week’s news. Click a citation occasionally — the habit that matters is checking that the source says the specific thing, not merely that a link exists.
MAKE A DECISION
An internal assistant answers a policy question correctly in substance but cites a document that doesn’t contain that statement. What does this most likely indicate?
CARRY THISAsk an AI search tool a question you know the answer to, then open its top citation and check whether that source actually states the claim.