Zero-shot asks the model to perform a task with no examples, relying entirely on pre-trained knowledge.
# Zero-shot classification
Classify the sentiment as Positive, Negative, or Neutral.
Respond with one word only.
Review: 'The product arrived late but works perfectly.'
# Zero-shot translation
Translate to French:
'Machine learning is transforming software development.'
# Zero-shot extraction
Extract all email addresses from the following text:
[text]
# When zero-shot works:
# Simple, well-defined tasks
# Tasks well-represented in training data
# When to use few-shot instead:
# Novel output format needed
# Domain-specific reasoning
# Precise structure required