Glossary · Term

dense retrieval

← all terms

Definition

Plain language

The standard way AI 'searches your documents' — by turning each document into a point on a map and grabbing whichever points sit closest to your question.

As stated in the literature

A retrieval method that encodes each document and the query into single vectors and returns nearest neighbors by distance; the engine under most RAG systems, provably limited in which relevance judgments one vector per document can encode.

Why it matters: It powers most systems that let AI answer questions from your documents, but its one-point-per-document design cannot capture every kind of relevance, so some questions slip through.

For example, when you ask a chatbot a question about your files, it turns your question into a point on a map and hands back the documents whose points sit nearest to it.

Heard on the show

“That's dense retrieval, the engine under retrieval-augmented generation.”
Episode 198 — The Model That Knows the Answer and Can't Say It

Mentioned in 1 episode

  1. 198
    The Model That Knows the Answer and Can't Say It

Related concepts

Related terms

RAG