Few-shot & One-shot Prompting

Zero-shot, one-shot, and few-shot are ways to give the model examples inside your prompt.

What they mean (very simple):

• Zero-shot: No examples. You explain the task and the model tries to do it.
• One-shot: You show one example to teach the style.
• Few-shot: You show a few examples (usually 2–5) that demonstrate the pattern you want.

When to use each:

• Zero-shot: Use for straightforward tasks or when examples might bias results.
• One-shot: Cheap nudge to show style without many tokens.
• Few-shot: Use when format or mapping is tricky and examples help guide the model.

How to pick good examples:

• Keep them short and focused.
• Make examples cover typical and a couple of edge cases you care about.
• Avoid many similar examples — variety teaches the model the pattern.

Trade-offs:

• More examples usually mean better guidance but higher token cost.
• Examples can bias behavior; test your template on many inputs to catch problems.

Quick rule: start with 1–3 examples, test, then add more if the model still misunderstands the task.