← Learn

Retrieval-Augmented Generation (RAG)

An AI architecture that combines retrieval of information from a database with generation of responses based on that information.

Learn

When to use it

Use Retrieval-Augmented Generation (RAG) when a standalone generative model lacks the necessary context to provide accurate or relevant responses. RAG combines a retrieval step with generation, allowing AI to pull specific data from a database to inform its outputs, useful for tasks like generating detailed technical documentation based on a company's internal knowledge base.

Quick example

In ChatGPT, a user might request detailed information about a niche topic. By enabling RAG, ChatGPT retrieves relevant documents from a connected database before generating a response. Here, RAG is the method that enhances ChatGPT's responses by grounding them in specific, retrieved data.

query → RAG → retrieve data → generate response → user

Ecosystem

RAG operates within a cycle of retrieval and generation, often interacting with databases and language models. This cycle ensures that the generated content is both informed and contextually accurate.

      ┌── database ──┐
query →│    RAG     │→ generate → user
      └── language model ─┘

Misconceptions

MisconceptionRebuttal
RAG changes the modelRAG supplements the model with retrieved data
RAG is only for textRAG can enhance any generative task with data

Trade-offs

  • Accuracy — requires robust data retrieval systems
  • Contextual relevance — needs well-maintained databases
  • Complexity — additional integration with retrieval systems

Seen in