Fine-tuning an AI model: what it is and when it pays off
2026-11-10
Definition
Fine-tuning is the process of training an existing large language model a little further on your own examples, so that it adopts a particular style, format or behaviour more reliably.
When it helps
Fine-tuning changes how a model responds. It is a poor way to teach it facts, because those facts get baked in and go stale. For up-to-date knowledge, retrieval-augmented generation is usually the better route. For many tasks, careful prompt engineering gets most of the benefit at far lower cost.
Consider fine-tuning only when you have a large set of high-quality example answers, a narrow repeated task and a prompt that still falls short. You also take on upkeep, because a fine-tuned model must be retrained or replaced when the base model changes. Check each provider's current terms on data handling before you upload anything. See build vs buy AI software for the wider cost trade-off.
A hypothetical example
Picture a legal costs firm that wants every draft to follow a rigid house format. After a good prompt still produces occasional drift, they fine-tune on several hundred approved past drafts so that the layout holds consistently.
Weighing these options belongs to AI development.