RAG
Retrieval-Augmented Generation — look up relevant docs first, then ask the LLM. Reduces hallucination by forcing the model to base its answer on your specific documents rather than its internal training data.
RAG is an architectural pattern that enhances LLM accuracy by retrieving context from an external knowledge base and injecting it into the prompt. This grounds the model's generation in specific, verifiable data, significantly mitigating hallucinations compared to relying solely on parametric memory.
Retrieval-Augmented Generation (RAG) is a framework that optimizes the output of a Large Language Model by referencing an authoritative knowledge base outside of its training data. The process involves a retrieval mechanism that fetches contextually relevant documents based on a user query, which are then concatenated into the prompt as grounding context, thereby constraining the model's generation to the provided information and reducing stochastic hallucination.
evolution
- 2020-05 · historyREALM Introduction
Google researchers introduced Retrieval-Augmented Language Model pre-training, establishing the foundation for combining retrieval with language generation.
- 2020-09 · historyRAG Paper Publication
Lewis et al. published the seminal paper 'Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,' formally defining the RAG architecture.
- 2022-12 · historyLangChain Integration
The release of LangChain provided the first widely adopted framework for developers to easily implement RAG pipelines with LLMs.
- 2023-03 · historyEnterprise Adoption
The launch of ChatGPT plugins and enterprise-focused RAG tools triggered widespread industry adoption of retrieval-based grounding for LLMs.