Fine-tuning means training an existing AI model further on a smaller, selected dataset so that it gets better at a particular task, style or specialist language.
A fine-tuned model has changed. It has learned new patterns from the examples and uses them from then on. With RAG, the model stays the same but is given access to relevant documents when it answers.
Fine-tuning requires many good examples, technical help and ongoing maintenance. It suits tasks where the form must be very consistent. For tasks where answers must draw on up-to-date knowledge, RAG is usually the better choice.
In practice
Most organisations do not need to fine-tune a model. Good instructions in a system prompt and access to the right documents cover the vast majority of needs. So ask what the problem is before you choose the solution.
The concepts in this glossary run through my AI courses and workshops, where we work with them on your own tasks.