Training / Standard term
Fine-tuning
Updating a pretrained model's weights with additional task, domain, behavior, or preference data.
Fine-tuning covers several methods, from full-parameter supervised training to parameter-efficient adapters and preference optimization. The dataset may teach formats, terminology, styles, classifications, or behaviors. Results depend on example quality, objective, amount of data, base model, and validation; training does not guarantee exact formatting or faithful recall of every fact in the examples.
Builder example
Fine-tuning can improve repeatable behavior or economics at scale, but it adds data governance, evaluation, deployment, and regression work. Retrieval is usually better for facts that must stay current or traceable, while prompting and structured-output controls may solve some behavior problems without changing weights.
Common confusion: Fine-tuning does not inject private knowledge into a model the way a database does. The model learns patterns from your data, so it may generalize or hallucinate around facts that appeared infrequently in the training set.