What is fine-tuning?
Fine-tuning is the process of training an existing AI model further on a set of your own examples so that it adopts a particular style, format or behavior more reliably than instructions alone can achieve.
A general model is trained by its maker on broad material. Fine-tuning takes that model and gives it extra lessons from your examples, such as hundreds of past inputs paired with the output you wanted. Afterward, it leans toward your patterns without needing long instructions every time. It changes how the model behaves; it is not a good way to teach it facts.
That last point trips up a lot of projects. If you want an assistant to know your current price list or policies, fine-tuning is the wrong tool, because the information goes stale and the model may still blend it with guesses. Retrieval from your live documents handles facts better. Fine-tuning suits consistent formats, classifications and tone at high volume.
Try good prompts and retrieval first; they are cheaper, faster to change and easier to audit. Consider fine-tuning only when you have a repeated task, hundreds of quality examples, and evidence that prompting has plateaued. Remember the examples are training data, so strip client identifiers and confirm the provider's terms on how that data is stored and used.
Example
For example, an insurance agency that sorts thousands of inbound service emails into a fixed set of categories might fine-tune a model on past emails labeled by experienced account managers. Before doing so, the agency tests a well-written prompt on a sample. If the prompt already sorts most emails correctly, fine-tuning may not be worth the cost. Volumes are illustrative.