Zero-shot learning: definition

Zero-shot learning is when a model performs a task it was not given any examples for, relying on its general training to figure it out from the instruction alone.

Ask a capable model to do something it has never been shown an example of, "classify this review as happy or angry," with no samples, and it often just does it. That is zero-shot: no examples, only the instruction. Few-shot, by contrast, gives it a handful of examples first.

Zero-shot is convenient and cheap, but for anything that must be consistent, a few good examples (few-shot) usually make the output far more reliable.

Related terms

Few-shot prompting: Few-shot prompting is giving a model a handful of worked examples inside the prompt so it follows the pattern, without any retraining.
Prompt engineering: Prompt engineering is the craft of writing the instructions, examples, and context you give a model so it produces reliable, correctly formatted output for a specific task.
Large language model (LLM): A large language model is a neural network trained on huge amounts of text to predict the next token, which lets it write, summarise, classify, and answer questions across almost any topic.