AI Glossary / Fine-tuning

🏗️ Models and training

Fine-tuning is additional training of an existing, pre-trained model on supplementary data to make it better at a specific task: write in a company's style, know industry terminology, or stick to a defined response format. This is much cheaper than training from scratch. In practice, fine-tuning is often a last resort: the same effect can often be achieved with a good system prompt or RAG, meaning connecting the model to a knowledge base. You reach for it when it's about the style and form of responses, not about providing the model with facts.