AI Glossary
Zero-shot (no examples)
zero-shot learning, zero shot, no examples
Zero-shot is a way of prompting in which you ask the model to perform a task without showing a single worked example — you rely solely on the instruction itself and the model's knowledge from training.
- The prompt contains no worked example — only a description of the task.
- It works when the task is common and the model saw similar data in training.
- When the output is unstable or poorly formatted, you move on to few-shot.
Zero-shot is the simplest way to work with a model: you phrase the instruction in words and expect a correct answer, without attaching a single worked example. The model relies solely on the instruction and on what it learned during training. It works well for common, well-described tasks — classification, summarization, simple data extraction.
The difference from few-shot is single and specific: there you show a few input–output pairs in the instruction, here you show none. Both approaches are forms of in-context learning — the model adapts within a single request, without changing its weights. Zero-shot therefore needs no room for examples, but it gives you less control over the format of the answer.
In deployments, zero-shot is the starting point, because it is the cheapest and fastest to get running. If the model gets the format wrong, misses edge cases, or answers inconsistently, you add examples (few-shot) or sharpen the instruction through prompt engineering. Only when that isn't enough do you consider fine-tuning the model.
Related terms