Learning RAG: Retrieval-Augmented Generation

RAG definition and purpose

Retrieval-augmented generation (RAG) is a technique for using large language models (LLMs) that first retrieves relevant information from external data sources (such as document collections, databases, or the web) and then uses that retrieved text to help generate an answer to a user query. In this setup, the retrieved documents supplement what the LLM learned during training, enabling the model to use domain-specific and/or more up-to-date information that may not be present in its original training data. The purpose of RAG is to improve response quality and reliability by grounding generation in external sources, which can reduce issues like hallucinations and increase transparency by allowing cited or verifiable material. It also helps reduce the need to retrain models when new information becomes available, since updates can be made to the external knowledge base rather than the model weights. While RAG improves factuality, it does not fully eliminate errors—models can still misinterpret retrieved context or generate misinformation even when the sources themselves are factually correct.

0/2 modes complete

How RAG differs from plain LLMs

0/2 modes complete