Fine-tuning

A base model learns from a huge general dataset. Fine-tuning continues that training on a smaller, targeted set of examples, adjusting the model's weights so it behaves more like those examples by default.

It is often not the first tool to reach for when building agents. Better prompts, examples in context, retrieval, and well-designed tools solve most problems faster and keep you free to switch to newer models. Fine-tuning tends to help with consistent formatting, specialized classification, or making a smaller, cheaper model match a larger one on a narrow task.

Fine-tuning teaches behavior more than facts. If the problem is that the model lacks up-to-date or private information, retrieval is usually a better answer.