What is the difference between fine-tuning and RAG?

Fine-tuning changes how a model behaves — its style, tone, or format — by permanently retraining its weights, while RAG changes what a model knows at answer time by feeding it outside information without touching those weights at all.

Fine-tuning is extra training on top of a base model. You show it thousands of examples of the tone, format, or skill you want, and its internal weights shift to match. The knowledge it already had stays frozen — a fine-tuned model doesn't magically learn facts from next month unless you retrain it again.

RAG (retrieval-augmented generation) works differently. Instead of touching the model, it retrieves relevant documents — your company wiki, today's news, your product docs — and drops them into the prompt right before the model answers. Update the documents and the model's knowledge updates instantly, no retraining required.

Quick comparison

The rule of thumb: reach for RAG when you need the model to know your documents or anything that changes often. Reach for fine-tuning when you need it to consistently behave a certain way. They're not rivals — plenty of real systems fine-tune a model for tone and use RAG to keep it factually current.