Learning RAG: Retrieval-Augmented Generation

RAG definition and purpose

Retrieval-augmented generation (RAG) is a technique for enhancing large language models (LLMs) by letting them retrieve and incorporate relevant information from external data sources. In a typical RAG setup, the model first selects documents from a specified collection (such as a database, uploaded files, or web sources) and then uses those retrieved texts to answer the user’s query, supplementing what the model learned during training. The purpose of RAG is to make LLM outputs more grounded and up to date without retraining the model whenever new information becomes available. By retrieving relevant context before generation, RAG can reduce hallucinations and improve factual reliability, and it can also provide cited sources so users can verify what the system used. While RAG improves accuracy, it does not fully eliminate errors—models can still misunderstand retrieved context or generate misinformation, and RAG systems can be affected by misleading or improperly interpreted sources (sometimes called “RAG poisoning”).

0/2 modes complete

How RAG differs from plain LLMs

0/2 modes complete