Fine-tuning
Extra training that takes an existing AI model and adapts it to a particular job, style or subject using a smaller set of examples.
Training a big model from scratch takes huge amounts of data, time and computing power. Fine-tuning is a shortcut. You start with a model that already has general skills and give it further practice on a narrower set of examples, so it becomes better at one kind of task.
A company might fine-tune a model on thousands of its past customer service replies so that it answers in the company's tone and knows its products. A hospital might fine-tune one to summarise medical notes in a standard format.
Fine-tuning changes how a model behaves, but it is not the only way to customise one, and it is not always the best. Often simply giving the model the right documents along with your prompt works well. Fine-tuning on poor or one-sided examples can also add new errors or bias.
Related: Model (AI model), Training data, Open-weights model