Welcome back! Today you're going to train your very own tiny pretend AI — and see exactly why what you feed it matters so much.
A Glimmer is a made-up kind of shape. You know the secret rule for what makes something a Glimmer — Byte doesn't, yet. Try labeling each card yourself, then feed it to Byte. Let's see what Byte figures out from 6 labeled examples.
Those 6 labeled cards you fed Byte are called training data — the examples an AI studies to learn a pattern. Byte didn't memorize each card individually; it looked for a pattern that fit all of them, then applied that pattern to brand-new cards it had never seen. That's the core trick behind basically every AI you'll meet in this course.
Here's the part that trips people up: AI doesn't necessarily learn the pattern you had in mind — it learns a pattern that fits the examples it was shown. If the training examples are good and varied (like ours were), it usually learns the right idea. If the training examples are narrow, messy, or mislabeled, it can learn the wrong idea instead — and still sound completely confident about it.
You just trained me using shapes. The same idea works for anything an AI learns to sort — even food. Quick pop quiz for you, before the real quiz: what do you think would happen if an AI learning to sort "healthy meals" from "junk food" only ever saw photos of salad labeled "healthy"?
You learn from examples all the time too — like recognizing a friend's handwriting, or knowing which foods you like. Now write about a time you learned something just by noticing it again and again — not because someone told you a rule.
This is saved right in this browser, on this device — it's never sent anywhere.
10 questions, picked at random from a bigger question bank — so if you take it again, you'll likely see a different mix. Let's go!
Time to have your first real conversation with an AI chatbot.