Fine-tuning

What is fine-tuning?

Fine-tuning is further training an existing AI model on specific examples so it performs a particular task, style or format more reliably. Google's machine learning course describes it as additional training that adapts a general model to a specific task. Prompting changes what you ask the model; fine-tuning changes the model's parameters. OpenAI's documentation lists classification and output in a specific format among its best uses.

Why it matters to a revenue leader

Vendors sometimes cite fine-tuning as proof that their AI understands sales. It can help with narrow, repeated tasks, such as sorting meeting notes into categories or writing in a set format. It is usually a poor way to keep a model current on facts about your business, because those facts change and the model would need retraining. Many revenue use cases are better served by good prompts and retrieval from your own data. OpenAI advises setting up evaluations first, so you can tell whether a tuned model beats the base model.

A fire protection company wants AI to sort inspection deficiencies into 12 repair categories. A general model with a good prompt gets 80% right on a test set of 500 past deficiencies. A version fine-tuned on 2,000 labeled examples gets 93% right, enough to limit manual sorting to the unclear cases.