Retrieval-Augmented Generation (RAG)
Retrieval-Augmented Generation; abrufgestützte Generierung
What it means
RAG is a technique that allows an AI system to retrieve relevant information from external sources and provide that information to a language model when generating a response. This helps the model answer questions using organisation-specific or current information without retraining the underlying model.
Real-world example
A museum has 80,000 collection records. A visitor asks an AI assistant about a particular historical object.
Instead of relying only on the language model's existing knowledge, the system searches the museum's collection records, retrieves relevant information and supplies it to the model to generate an answer with source references.
Why it matters for AI adoption
Organisations can build AI assistants using their existing knowledge sources without training a new foundation model. However, the quality of the answers depends on the underlying data, retrieval quality and system design. RAG reduces some risks but does not eliminate hallucinations.
Often confused with
RAG supplies relevant information to the model at the moment of answering and leaves the model unchanged, so updated documents take effect without retraining. Fine-tuning changes the model through additional training and is better suited to adjusting style, format or task behaviour.
Questions to ask
- How does the system identify and retrieve the most relevant documents?
- How do we evaluate whether the generated answers are accurate and supported by the sources?