Illustration 3 of 4

AI as a Maker

It makes things — blurbs, code, decks, documents. But it's a maker, not a vending machine. Give it the same brief twice and you'll get two different results. Run them and see.

Pick a brief, then run it twice
The maker Probabilistic
Generating, token by token…

{{ partial }}
Run #{{ r.n }}

{{ r.text }}

{{ makerHint }}

The vending machine Deterministic
Dispense #{{ v.n }}
Pretzels · slot B4

{{ vendHint }}

Why every run differs

The maker doesn't look up an answer and hand it back. It builds the text one piece at a time, and at each step it's choosing from many likely next words. Pick a different word here, and the whole sentence drifts somewhere new.

Summer's here — {{ w.alt1 }} {{ w.alt2 }} {{ w.chosen }}
faded words = the roads not taken at each step
My take

I've seen people ask two different AI models the exact same question, get two very different answers, and conclude the AI is weak — or even lying. But you can ask the same model the same question over and over and its replies will differ too. The learning point: the AI isn't looking up an answer. It's generating a response token by token, through probabilistic reasoning grounded in its training. Different every time — by design.

— Migell