Large Language Models
Fine-Tuning
Adapting a pretrained model to your task. Often the wrong first answer.
Grasp
Fine-tuning continues training a pretrained model on your own examples, adjusting its weights toward your task. It is the most-requested and most-misapplied technique in applied AI.
The rule that saves the most time: fine-tuning teaches form, retrieval supplies facts. If the model does not know something about your business, fine-tuning is a poor and expensive way to tell it, and the knowledge goes stale the moment your data changes. Reach for retrieval-augmented generation instead. Fine-tune when you need consistent structure, a specific voice, a narrow classification behaviour, or when you want a smaller model to imitate a larger one's behaviour on a narrow task.
Modern practice is almost entirely parameter-efficient: LoRA and its relatives train a small number of additional weights and leave the base model frozen. This makes fine-tuning cheap enough to iterate on, which changes it from a research project into an ordinary engineering task.