Decoder. plain-English AI glossary

Reader

▼ Fading

Also called Reader Model

The component that reads retrieved passages and produces the actual answer — in classic QA, the second half after the retriever.

Think of it like

The researcher who reads the pulled files and writes the summary, after the librarian fetches them.

Example

In a retriever-reader system, the retriever finds 10 passages and the reader extracts "March 2021" as the answer to "when did it launch?".

How it actually works

In the original retriever-reader design, the retriever surfaced candidate passages and a separate reader model extracted or generated the answer from them. Modern RAG blurs this: the LLM is the reader, generating rather than extracting, so the standalone "reader" term is fading. The division of labor still matters conceptually — finding evidence and using it are distinct jobs with distinct failure modes.

For product teams

A useful mental split even now: retrieval finding the evidence vs. the model using it well.

For engineers

The answer-producing stage after retrieval; in RAG it is the generative LLM, so the extractive "reader" is largely legacy.

Retrieval — The counterpart to the retriever.
RAG — Largely absorbed into modern RAG generation.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome