Short Definition
Fine-tuning adapts a pretrained model by continuing training on task-specific or domain-specific data.
Intuition
Instead of training from scratch, you start with a broadly capable model and nudge it toward the behavior you need.
Technical Definition
Fine-tuning updates some or all model parameters, or adaptation parameters, using additional supervised, preference, or domain data.
Example
A legal assistant may fine-tune a model on approved drafting examples and review rubrics.
Common Misunderstandings
Fine-tuning is not the right answer for every customization problem.
Fine-tuned models still need evaluation and safety controls.