Retrieval-Augmented Generation
Retrieval-Augmented Generation — A generation architecture that retrieves external information at request time and supplies selected evidence to a language model as additional context.
Context and Inference Boundary
RAG does not update model parameters; it supplies external context at inference time and therefore depends heavily on retrieval and context-construction quality.
A relevant document is not necessarily an authorized document. Access restrictions belong in candidate retrieval, not merely in a filter applied after answer generation; otherwise the presence or content of restricted records may leak indirectly.
Related LLM Concepts
Source
- https://arxiv.org/abs/2005.11401
Related technical article: Artificial Intelligence: Philosophy, Theory and Practice.