Summary
Zero-shot prompting: ask the model to perform a task without providing examples. The baseline. If zero-shot works reliably enough, add nothing. If it does not, add few-shot examples, then more structure, then evaluation.
Objectives
- 01State what zero-shot prompting is.
- 02Recognize tasks where zero-shot is sufficient.
- 03Design a fallback ladder when zero-shot fails.
The Lesson
The baseline
Modern frontier models handle most well-defined tasks zero-shot. Summarize, translate, extract, classify with a known schema — these usually work without examples.
When it fails
Domain-specific formatting, unusual conventions, edge-case handling — zero-shot degrades. The fallback ladder: add one-shot, add few-shot, add structured system prompt, add evaluation, add fine-tuning. Move one rung at a time.
Key Ideas
- Zero-shot is the baseline; add structure only when it fails.
- Most well-defined tasks work zero-shot with modern models.