You want AI to always format addresses as "City, Country (ZIP)". What's the BEST approach?

When you have a very specific format, few-shot is ideal. Showing examples is clearer than describing the format in words.

For few-shot classification (e.g. positive/negative/neutral), why should you MIX UP the classes in your examples?

If you put all POSITIVE examples first, then all NEGATIVE, AI might learn the order instead of the actual meaning. Mix them up so it learns the real classification signal.

Chapter 3 of 7·6 min read·Quiz included

📚

Show & Tell — Few-Shot Prompting

Q: What is "one-shot" prompting?

One-shot = one example. You show the AI ONE input-output pair to demonstrate the pattern, then give your real question.

Teach AI your exact pattern with examples

Teaching a New Skill 🏫

Remember learning to tie your shoelaces? Nobody just said "tie your shoes" and walked away. They sat next to you, showed you step by step, and let you copy. You learned by seeing examples first. Few-shot prompting works exactly the same way. Instead of just telling the AI what to do, you SHOW it a few examples first. The AI sees your pattern and applies it to your real question. One example = one-shot. Two or more examples = few-shot. Both are incredibly powerful.

🤔 Why Not Always Use Zero-Shot?

Zero-shot is great, but sometimes AI needs to see your specific pattern. When does few-shot win? ✅ When you want a very specific output format ✅ When you're doing repeated classification tasks ✅ When the AI keeps misunderstanding the 'shape' of your answer ✅ When you have a custom style that's hard to describe in words

If zero-shot gives you the wrong FORMAT, use few-shot to show the right one.

🍕 One-Shot: Learning from a Single Example

You run a pizza restaurant and want AI to format orders consistently. Show ONE example first:

Input

I want a large pepperoni with extra cheese

Output

ORDER: Large | Pepperoni + Extra Cheese | $14.99

Input

Can I get a small veggie please?

Output

ORDER: Small | Veggie | $9.99

Your ask: Now process: "Two medium BBQ chicken pizzas"

AI: ORDER: 2x Medium | BBQ Chicken | $26.00

AI learned the format from ONE example and applied it perfectly.

😊 Few-Shot: Teach a Classification System

Teach AI to classify product reviews with your custom emoji system:

Input

"Amazing product! Arrived fast, works perfectly!"

Output

🌟 EXCELLENT

Input

"Okay quality but overpriced for what you get."

Output

😐 AVERAGE

Input

"Broke after 2 days. Waste of money. Very disappointed."

Output

💀 TERRIBLE

Your ask: "Decent but packaging was damaged on arrival."

AI: 😐 AVERAGE

Without the examples, AI wouldn't know your emoji classification system!

💡 Rules for Good Few-Shot Examples

✅Use 3–5 examples for best results

✅Make sure your examples are diverse (cover different scenarios)

✅For classification tasks, mix up the classes (don't put all positives together)

❌Don't use examples with errors — one bad example confuses the model

❌Don't use too many examples — this wastes tokens and can slow things down

Quick Quiz

Test what you learned from this chapter. No pressure — it's just for you!

Question 1 of 3

What is "one-shot" prompting?

← Previous Next Chapter →