Prompt Engineering 101

Prompt engineering is not a magic sentence.
It is a structuring discipline.

Your goal: prevent the model from going in the wrong direction.

1. Chain-of-Thought (Step-by-Step Reasoning)

Principle: "Explain your reasoning step by step before giving the answer."

Useful for:

• decisions
• analyses
• complex comparisons

Limitation: a detailed reasoning can be wrong but very convincing. Always verify the assumptions.

2. Few-Shot Examples

You provide 1 to 3 representative examples.

Structure:

• Input -> Expected output
• Then: "Do the same for X"

This is the most reliable technique today.

3. Structure the Prompt

Recommended order:

1. Context
2. Objective
3. Constraints
4. Expected format
5. Tone / style
6. Examples

If two rules contradict each other, the model will choose arbitrarily.

4. Built-in Quality Control

Ask for self-verification:

• "Identify any inconsistencies."
• "Suggest an improved version."
• "List the weak points of your answer."

Very effective for going from correct to solid.

5. Say What You Do Not Want

Models tend to over-introduce and over-conclude.

Example: "No introduction. No conclusion. Short sentences. No jargon."

6. Experimental Approach

Prompt engineering is empirical.

Simple rule: if you cannot explain why it works, you will not be able to reproduce it. Document your effective prompts.