Glossary
Fine-tuning
Fine-tuning is extra training that adapts an existing AI model to a specific job or style using your own examples. It changes the model itself, unlike prompts or RAG, which only change what you feed it.
A base AI model is like a chef fresh out of a great culinary school: excellent general technique, knows a thousand cuisines. Fine-tuning is the six months that chef then spends in your trattoria learning your dishes, your portions, your regulars. Afterwards they don’t need the recipe explained each time. It’s in their hands.
Technically, you continue the model’s training on a focused set of examples: thousands of your support tickets with ideal answers, or documents in your company’s exact tone. The result is a model that does that one thing more reliably without long instructions.
The honest framing for most people: you’ll use fine-tuned models more often than you’ll ever make one. Fine-tuning needs curated data and budget; for “answer using my documents,” RAG is usually the simpler and cheaper tool, and good custom instructions cover most personal needs.
Where you’ll meet this
In product descriptions (“fine-tuned for medical text,” “tuned for legal Italian”), in OpenAI’s and Mistral’s developer platforms that offer fine-tuning as a service, and around open-weights models, which anyone can fine-tune on their own hardware.