Where do I start?
Maybe you're here because everyone is talking about AI and you feel you've missed the boat. You haven't. Nobody understands all of this — not even the people building it. This is the calm way in: five honest questions, one thing to try, three things to take away. And you can't break anything along the way.
What actually happens when I chat with an AI?
Whether the thing is called ChatGPT, Claude or Copilot: you're writing with a program that has learned, from vast amounts of text, which word is likely to come next. It doesn't read like a person and doesn't ponder like one — it computes. Astonishingly well. You can watch such a model at work right here, in slow motion: A GPT in slow motion.
Does it think? Does it understand me?
As far as anyone can tell: no. No consciousness, no intent, no feelings — but an impressively good simulation of understanding, and that simulation feels real. Take both seriously: the tool and the feeling. Just don't mix them up. What “AI” actually means is in the glossary.
Why does it sometimes tell confident nonsense?
Because it is built to continue, not to be right. Where its knowledge ends, it keeps writing anyway — in the same assured tone, and you can't hear the difference. That's not a defect; it's its nature. Hence the most useful everyday rule: the more depends on an answer, the sooner you double-check. More in Convincingly wrong.
Is it listening? What happens to what I type?
What you type travels to the provider's servers and — depending on provider and settings — may be stored or used to train future models. The simple rule of thumb: don't tell an AI anything you wouldn't tell a friendly but unfamiliar service provider. What training data is, the glossary explains.
Do I have to master all this?
No. You don't have to use AI, let alone love it. But a basic understanding pays off — the way you roughly know what a loan is without being a banker: it makes you harder to impress and harder to fool. That's exactly what this website is for.
And if the topic worries you more than it excites you: the article What AI does to us takes exactly that seriously — without alarm.
Try it once
AI loses its magic — and its menace — fastest when you watch it work. Start with the die: the same question, rolling answers, one dial. That's all it takes to understand why the same question doesn't always get the same answer: The die in the machine.
If you take only three things
- One lab: The die in the machine — five minutes, one dial, and you'll know more about AI answers than most.
- One article: Convincingly wrong — the most useful everyday lesson: confidence is not a signal of truth.
- One reference: the glossary — whenever a term comes up, it's there in plain words.
And if you want more afterwards: the reading thread waits on the home page — five articles that build on each other.
What you can safely try today
- Have it summarise a long text you know well — you'll see immediately what it's good at and where it gets sloppy.
- Ask for wording help: an email made friendlier, a sentence made shorter. You stay the author; it's the copy editor.
- Have something explained — and ask follow-ups as often as you like. It never gets impatient, and there are no stupid questions.
Just two rules: don't type in anything private or confidential (question 4) — and double-check anything that health, money or law depends on (question 3).
That was the start — you don't need more today. The rest of this site isn't going anywhere.