Retrieval & Memory · 2020
Retrieval-Augmented Generation (RAG)
Introduced retrieval-augmented generation, coupling a parametric sequence-to-sequence model with a dense-retrieval index so a language model can pull in external documents at generation time instead of relying only on facts baked into its weights.
Editorial record
Plain-language summary
RAG pairs a trained generator with a retriever that fetches relevant passages from an external corpus (a Wikipedia index searched with dense vectors), then conditions the output on those passages. Both the retriever's query encoder and the generator are trained together, and retrieved documents can be swapped or updated without retraining the model. This lets a model cite and use up-to-date or domain-specific knowledge, reducing fabricated answers on knowledge-intensive tasks like open-domain question answering.
Knowledge graph
Relationships
Antecedents
ExtendsEvidence: Direct
FiD / RETRO / Atlas (scaling retrieval)
FiD/RETRO/Atlas scale retrieval
P-323
Descendants
Depends onEvidence: Direct
REALM: Retrieval-Augmented LM Pre-Training
RAG builds on retrieval-augmented LM idea
P-322
Depends onEvidence: Direct
Dense Passage Retrieval (DPR)
RAG uses DPR retriever
P-322
Converts into infrastructureEvidence: Strongly supported
Contrastive Text Embeddings (Sentence-BERT/SimCSE/E5/BGE)
Embeddings power production RAG
P-325
EnablesEvidence: Strongly supported
As We May Think (Memex) + Science the Endless Frontier
Memex vision underlies retrieval/RAG
P-322
Source record
Provenance
- Record ID
- P-322
- Record created
- 2026-07-13
- Last reviewed
- 2026-07-14
- Record version
- 2
- https://arxiv.org/abs/2005.11401
- arXiv:2005.11401
Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.