Zero-Shot Learning

What Zero-Shot Learning is

Zero-shot use relies entirely on capability acquired during pretraining and instruction tuning. It is the default way most people use assistants.

How it works

The prompt states the task, constraints and desired output format. Clarity and explicit formats matter far more here than in few-shot prompting.

Why it matters

It removes the cost of building labelled datasets for a huge range of classification and extraction tasks that previously required them.

Common uses

• →Ad-hoc classification
• →Summarisation
• →Translation
• →Quick data extraction