Zero-Shot.

The ability of a model to perform a task correctly from just a description, with no examples — demonstrating that it has generalized the concept from training.

Zero-shot learning refers to a model's ability to perform a task it has never been explicitly shown examples of, guided only by a description of what to do. For example, asking a model to 'classify the sentiment of the following review as positive, neutral, or negative' without providing any example classifications is a zero-shot prompt.

The strength of zero-shot performance is a proxy for how well a model has generalized its training knowledge. Larger, better-trained models typically exhibit stronger zero-shot capabilities because they have seen enough diverse tasks during training to understand new task descriptions without needing examples.

Zero-shot prompting is often sufficient for well-defined tasks that the model likely encountered variants of during training. For unusual or highly specialized tasks, few-shot prompting (providing several examples) significantly improves performance. The choice between zero-shot and few-shot depends on task complexity, example availability, and context window constraints.