Show, don't tell

Try to explain, in words only, exactly what makes a joke your kind of funny. Go on.

Hard, right? You could write a whole page and still not nail it. But show someone three jokes you love, and they'll get it instantly.

That's the secret of this unit. When you want an AI to match your taste, you don't describe it. You show it examples. Builders call this few-shot — teaching with a few shots, a few examples.

A good example is worth a thousand instructions. Taste is easier to show than to say.

One idea: examples beat descriptions

Remember Unit 3 — role, goal, context, constraints. That's telling. This is the next level up: showing.

Watch the difference. Say you want short, punchy product names.

Telling: "Make the names short, modern, punchy, memorable, not too corporate, kind of playful but not silly…" (the AI half-guesses what you mean)

Showing:

Input: a water bottle that keeps drinks cold → Output: Frostpaw Input: an app that reminds you to text your friends → Output: Nudge Input: a backpack with a built-in charger → Output: ???

Given those two examples, the AI feels the pattern — short, one word, a little playful — and fills in the third the way you would. You never had to define "punchy." You showed punchy.

This is the exact idea from Unit 2, flipped. There, a model's behavior came from whatever examples it was fed. Here, you are the one feeding the examples. You've become the data.

Do the thing

You want an AI to reply to texts in your voice. Which teaches it better?

• A. "Reply casually and friendly, like a teenager, but not too much slang, use some but not a lot."
• B. Three real texts you've actually sent, followed by "now reply to this new one in the same style."

Pick one. Then think: why?

Quick check. B, every time. Your three real texts contain your voice — the rhythm, the emoji habits, the exact amount of slang — without you having to name any of it. Description makes the AI guess at your taste. Examples hand it to them. The catch: your examples have to actually be good, because the AI copies them faithfully — flaws and all. Which is the whole next lesson.

Why this matters

"Show, don't tell" is how you get an AI to work your way instead of the default average way.

• When words fail, examples work. Anything hard to describe — a style, a voice, a format — is easy to demonstrate.
• You are the teacher now. The model matches the examples you choose, so choosing well is the skill.

Next lesson: what makes an example good. Because a model that copies your examples will copy your bad ones just as eagerly as your best ones.