Neura saysZero-shot, one-shot, and few-shot describe how many examples you put in the prompt.
Zero-shot: just instructions, no examples. One-shot: one example of the desired input→output. Few-shot: several examples that demonstrate the pattern, format, and edge cases. More examples generally improve reliability and formatting — leveraging in-context learning, where the model adapts from examples without any weight updates. Few-shot is a quick way to pin down a tricky format.
Power-ups you unlock
Zero-shot: instructions only
Few-shot: several in-prompt examples
In-context learning, no weight updates
Examples improve reliability and format
The Overfit Ogre attacks — common mistakes
Using zero-shot for a tricky/specific format
Inconsistent few-shot examples confusing the model
Burning context on too many examples
Boss battleDecide zero-shot vs few-shot for enforcing a strict JSON output format.