Retrieval-augmented generation
Also: RAG
Fetching relevant material at question time and putting it in the prompt, so the model answers from supplied text rather than from memory. Knowledge can then be updated, restricted and cited without retraining.
What people get wrong
Most failures are retrieval failures, not generation failures. Before blaming the model, check whether the answer was in the passages it was given.
A definition is not an understanding. This is taught properly in The RAG pipeline, part of Retrieval-augmented generation.
Learn this properly