Zero-shot learning is when a model handles a task it was never specifically trained on, with no examples of that task at all.
In plain words
Most models learn a task from many labelled examples of it. A zero-shot model is simply asked to do something new and manages it, using what it learned more broadly. Large language models do this when you ask a question without showing any examples first.
A quick example
An image model trained to match pictures with captions is asked to sort photos into "cat", "dog" or "horse". Nobody trained it on those three labels, yet it picks the best match for each photo using only the words.
Where you'll meet it
- Asking a chatbot to do a task without giving examples.
- Image models that recognise categories from a text description.
Related
Sources
- Docszero-shot learning, Google Machine Learning Glossary · read 29 Sept 2026
- PaperLearning Transferable Visual Models From Natural Language Supervision, Radford et al. (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.