4It makes things

Same order, different every time?

AI made a picture for you, a story and a tune. But there's something odd.

Ask for the same thing and it comes out different. Let's put them side by side.

Let's call what you asked for with the cards an order.

Both times the order was exactly the same. The two results are not.

Say AI kept pictures in a drawer. If it just fetched one, what would that take? Hundreds of Posongs, each with different arms. For every order.

There is no drawer like that. AI doesn't fetch. It makes it right then.

That's the end of part 1. You've seen AI make a picture, a story and a tune.

A little deeper

For grown-ups, and for children who want more.

Getting different results from the same order is called stochastic generation. Ordinary programs return the same output for the same input. Generative AI does not.

What the machine produces is not one answer but a probability distribution over candidates. At the last step it draws one, and that draw differs each time — a different starting noise for pictures, a different word picked for text.

This is both the strength and the weakness. It can make something new every time, but it cannot promise the same answer to the same question. The trouble you meet in chapter 10 starts right here.

What you see here is chapters 1, 2 and 3 run twice each. Not illustrations made for comparison — two real outputs. The tune and the story are made twice on the spot; the picture, for the reason given in chapter 1, takes two different ones from the eight drawn in advance (see the chapter 1 deeper note). Very rarely the two tunes come out identical — then The same is the right answer and the screen says so. Getting the same answer to the same order once in a while is also part of what drawing means.

The drawer image is deliberate. Children usually assume AI keeps answers somewhere and fetches them. One fact — same order, different result — breaks that in a single move. Some children ask "what if the drawer has lots of them?", so the argument is built on this: no drawer holds hundreds of pictures with every arm in a different place. Being able to produce countless different results is what separates making from fetching.