What is the second approach—training on examples—called?

Fine-tuning takes an already-trained, general-purpose AI model and continues training it on a smaller, more focused set of examples specific to a task or style, adjusting the model's internal parameters so it absorbs that skill or tone directly into how it generates responses. This differs from simply changing the wording of your instructions, since fine-tuning modifies the underlying model itself rather than just the input you send it.

Prompting is the lighter-weight alternative: crafting clever instructions or examples to guide a general model's behavior without changing the model itself, which is fast and cheap but only goes so far. Streaming refers to displaying a response gradually as it's generated, an unrelated matter of output delivery, not how a model learns.

Fine-tuning can be more expensive and time-consuming than prompting since it requires retraining on curated examples, but it can bake in specialized knowledge or behavior more deeply than clever prompting alone, which is why companies often fine-tune models for specific professional domains.