Looking things up

One common addition is retrieval: looking up relevant material before generation. This can happen once in a simple question-answering pipeline. It does not require an agent loop.

The retriever fetches text and places it in the model's context. The frozen weights have not learned the new fact, but the model can use the supplied document for this answer.

That pattern has a clunky name, and placing it in the context window before the model answers, so its reply is grounded in real documents rather than its frozen memory.), and the longer label you may hear is RAG, retrieval-augmented generation.

The broad shape is question, search, selected documents, then generation. Implementations differ, and poor retrieval can supply irrelevant or misleading material.

Retrieval can improve accuracy for current or specialised questions. The model can still misread a source, ignore it, or make a claim the source does not support. Citations make that grounding easier to inspect.

Looking things up

One common addition is retrieval: looking up relevant material before generation. This can happen once in a simple question-answering pipeline. It does not require an agent loop.

The retriever fetches text and places it in the model's context. The frozen weights have not learned the new fact, but the model can use the supplied document for this answer.

The broad shape is question, search, selected documents, then generation. Implementations differ, and poor retrieval can supply irrelevant or misleading material.