Choosing Good Examples

The quality of your examples determines the quality of few-shot results. Selecting the right examples is both an art and a science.

What Makes a Good Example?

Good examples are:

• Representative - They reflect real-world cases
• Clear - The input-output relationship is obvious
• Diverse - They cover different scenarios
• Correct - They demonstrate the right behavior

Example Selection Criteria

1. Coverage

Cover the full range of expected inputs.

2. Difficulty Gradient

Include examples ranging from easy to challenging:

3. Edge Cases

Anticipate tricky scenarios:

Exercise: Select Diverse Examples

Example Ordering

Order can affect results:

Recency Bias

The last example often carries more weight. Put important patterns last.

Similar First

Put simpler examples first, complex ones later.

Balanced Alternation

For classification, alternate between categories:

Quality Over Quantity

Exercise: Quality Example Writing

Avoiding Example Bias

Class Imbalance

Length Bias

Content Bias

Example Debugging

If results are inconsistent:

1. Check if examples contradict each other
2. Verify all examples are correctly labeled
3. Look for unintended patterns (length, keywords)
4. Test with edge cases
5. Add examples for failure cases

Practice: Example Audit

Well-chosen examples are the foundation of reliable few-shot prompting.