// AI for complete beginners

Getting to know LLMs

LLMs — large language models — are the technology behind ChatGPT, Claude and Gemini. This collection explains in plain language how they work and how to think when working with them. It grows steadily from daily practice; read the first tip right away and unlock the rest with your e-mail.

Free sample

AI never gets it 100% right. Everything else follows from that

When AI creates something — text, code, anything — it is not a machine that produces the same correct result for the same brief every time. Every task is a bit of a dice roll: some things land, some don't.

A weaker model rolls badly. Imagine the correct result only comes up on a four, five or six — out of ten tasks, it gets half wrong. A top model rolls much better: out of a hundred tasks it lands, say, 95. But a hundred out of a hundred never happens. With any model. Today or tomorrow.

That sounds like bad news, but it isn't. It just means one thing: there's no point hoping the mistake won't come. There is a point in setting up the work so the mistake finds itself — immediately, not after it has done damage.

The golden rule: let AI check its own work

When AI can verify its own work, it also fixes most of its own mistakes — quickly, without you having to sit over it. So with every task, think about one thing: give it that ability.

A programming example: when AI builds a website, let it run the site and look at it with its own eyes — exactly like a human developer would. It will see what's off and fix it before you ever see it. The same applies everywhere: it can re-read a document, verify a calculation, click through a spreadsheet.

The second trick: request a self-review as a separate task

While creating, AI doesn't review itself — it focuses on making, not checking. But when you assign the review as a separate follow-up task (“check your own work”), it very often finds mistakes and things it can still improve. It's the same dice — but you're rolling it a second time, and two bad rolls in a row are far less likely than one.

For anything important, always ask for this second loop. It costs seconds and catches exactly the mistakes your customer would otherwise find.

Remember

  • Every task is a dice roll — even the best model doesn't roll right every time.
  • Set up the work so AI can check itself automatically — then it fixes most mistakes on its own.
  • For important work, request a self-review as a separate step.

More tips & tricks

I keep adding explanations and workflows I use in my daily work with AI. One e-mail unlocks them all — including everything added in the future.

Unlock the whole collection

Enter your e-mail and every tip — including future ones — becomes available on this device.

I'll use your e-mail to unlock the content and occasionally send new tips. No spam, no sharing with third parties; unsubscribe with a single reply.

I build exactly this kind of automation

These tips are the foundation every good AI deployment stands on. If you want AI in your company done properly, let's start with an initial consultation — the fee is credited toward the project if we work together.