Zero-Shot Learning
Also known as: Zero-Shot Prompting
In one sentence
Asking an AI model to perform a task without providing any examples — relying entirely on its pre-trained knowledge and the instruction itself.
Explain like I'm 12
Like asking someone to do something they've never practised before, without showing them how first. You just explain what you want and trust they can figure it out from what they already know.
In context
When you type 'Translate this to French: Good morning' into ChatGPT without providing any translation examples, that's zero-shot prompting. The model understands the task from the instruction alone. Zero-shot works well for common tasks like translation, summarisation, and classification. For more complex or specialised tasks, few-shot prompting (providing 2-5 examples) often produces better results. The impressive zero-shot abilities of modern LLMs are a key reason they've become so widely adopted — no training data or examples required.
See also
Related Guides
Learn more about Zero-Shot Learning in these guides: