When should I use few-shot prompting?

Use few-shot prompting whenever you have a specific, repetitive format or logic that the AI needs to follow perfectly.

Prompt Engineering Basics (Expanded)

Foundational Techniques: The Methods of Control

Now that you understand what a prompt is, it's time to learn the Tools of the Trade. These techniques are the building blocks of every complex AI system in existence today.

1. Zero-Shot Prompting: The Default Mode

Zero-shot is when you ask the AI to perform a task without giving it any prior examples. It relies entirely on its pre-existing training. It's best for common tasks where the outcome is predictable.

Example 1: "Translate this sentence to French."
Example 2: "Summarize this article in 3 sentences."
Example 3: "Write a basic HTML template for a landing page."

2. Few-Shot Prompting: Pattern Teaching

Few-shot is arguably the most powerful technique for automation. By providing 3-5 examples of how a task should be done, you define a pattern that the AI will follow with incredible accuracy. Few-shot = Pattern Teaching.

📊 Real-World Example: Data Classification

Prompt:

Classify the sentiment of these customer reviews:

Review: "I love this product, it changed my life!" -> Positive
Review: "The delivery was late and the box was broken." -> Negative
Review: "It works as described, nothing special." -> Neutral
Review: "It's okay, I've seen better but it gets the job done." ->

The AI will see the pattern and correctly classify the last review as Neutral.

3. Role Prompting: Behavior Control

Role prompting is about Behavior Control. It forces the AI to adopt a specific vocabulary, level of expertise, and perspective. A CFO will look at a business plan through the lens of risk and ROI, while a CTO will look at it through the lens of scalability and tech debt.

💻 Coding Scenario: Performance Optimization

You are a Lead Backend Engineer at a Fortune 500 company.

Task: Optimize this Python API for performance.

Context: We are experiencing latency during peak hours (10k requests/sec).

Action: Identify the bottlenecks in the code below and provide a refactored version that uses asynchronous processing.

Insight: Why Technique Matters

Few-shot is for Consistency. Use it when you need the output to look exactly the same every time.
Role Prompting is for Quality. Use it to pull the most "intelligent" version of the model out for a specific domain.

Prompt Engineering Basics (Expanded).