Technical GlossaryNatural Language Processing

Closed-Book Question Answering

An approach that answers questions using only the information stored in model parameters, without external document access.

Closed-book question answering refers to generating answers from the model’s internal parametric memory without retrieving external documents. This can work well for some general-knowledge questions, but it is limited in freshness and factual reliability. In enterprise or high-risk scenarios, retrieval-supported open-book setups are often safer. Still, it remains important for understanding how much knowledge a model has internalized.