One-Shot Prompting

Introduction

One-shot gives the model exactly one example of the desired input → output pattern. It's the cheapest way to teach the model an unusual format.

Beginner analogy: Showing a new intern one completed expense report. They don't need a manual — one example clarifies the format.

Understanding the topic

Core concepts to understand:

• Format: instruction + 1 example + new input.
• Best for tasks where format is novel but logic is simple.
• Cheaper than few-shot, more reliable than zero-shot.
• Watch out for the model copying the example too literally.

Syntax reference

Visual workflow / architecture:

Instruction: Convert dates to ISO.

Example:
  Input:  "3rd of March 2024"
  Output: "2024-03-03"

Now do:
  Input:  "12 December 1999"
  Output: ?

Real-world use

Form-filling, data extraction and code-formatting tools rely heavily on one-shot prompts — one example is often enough.

Best practices

• Use one shot when zero-shot fails but format is the only problem.
• Pick an example that covers the edge case you fear.
• Mark example and target inputs with delimiters to avoid leakage.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Difference between zero-shot and one-shot?
• Q2. Risk of using a biased single example?
• Q3. When would you upgrade from one-shot to few-shot?
• Q4. How do you choose THE example?