Zero-shot
Asking a model to do a task with no examples at all — just the instruction.
Think of it like
Like handing a capable new hire a task described in one sentence and trusting them to figure it out.
Example
“Classify this review as positive or negative: …” with no sample classifications — the model just does it.
How it actually works
Zero-shot leans entirely on what the model absorbed in pretraining plus your instruction. It’s the cleanest, cheapest way to prompt and often works surprisingly well on common tasks, but for anything unusual or format-sensitive, adding a few examples usually beats it.
For product teams
Try zero-shot first — if it works, you’ve saved tokens and effort; if not, add examples.
For engineers
Task performed from instructions alone with no in-prompt demonstrations, relying on pretrained priors.
Related
- Few-shot — The same idea with a few examples.
- One-shot — With exactly one.
- In-context learning — The mechanism that makes examples help.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome