Skip to content
← Under the Hood

Eight ideas from the course, sized for the wall. Landscape paper, one per page — they work best where you'll actually see them.

Under the Hood

One word at a time.

Under the hood · the prediction loop

What you give it

Thecatsatonthe?

The model

weighs every word it could say next

What comes out

The cat sat on the mat

+ one new word

What it's choosing between

mat0.61
floor0.14
couch0.09
rug0.05
roof0.03

It picks the most likely one — mat — and says it.

Then mat gets added to the sentence, and the whole thing runs again — guessing the next word, then the next. One word at a time, thousands of times a second. That's the whole trick.

A language model doesn't write sentences. It predicts one word, adds it, and repeats — that's the engine behind every answer.

Gstaad · 2026

Under the Hood

It's a desk, not a diary.

Under the hood · the context window

What you put in front of it

your questionthe article you pasted“keep it short”“act as a coach”one example

The context window — its desk

your questionthe article you pasted“keep it short”“act as a coach”one example
earlier stuff ↓

nearly full

It only knows what fits here. Run out of room and the oldest bits drop off.

What you get

An answer built from exactly what's on the desk.

Nothing on the desk? It guesses — and that's when it goes wrong.

The model can only work with what fits on its desk. Give it the right things — and don't expect it to remember what fell off.

Gstaad · 2026

Under the Hood

Fluent isn't true.

The move · catching a confident lie

The AI says — confidently

“Mount Kilimanjaro is 5,895 m tall, and it grows about 2 cm every year.”

sounds sure of itself

Your check

✓Is it recent?
✓Is it obscure?
✓Is it very specific?
✓Is there a number?

Any yes → triangulate.Check two sources that don't copy each other.

What you do with it

Confirmed → use it

Can't confirm → hold

You're the editor — not the audience.

Fluent isn't the same as true. You're the gate every answer has to pass through before you trust it.

Gstaad · 2026

Under the Hood

Start from the thing you want to make.

How to choose · start from the making

Start with what you want to make

a song for my friend's birthday

Match it to the medium

Each model is fluent in one kind of thing — the one it trained on.

…and the tool follows

an audio model

The whole workshop

words — an essay, a storya text model
a picture — a poster, a logoan image model
a song or a voicean audio model
a working website or appa code model
Don't start by picking an AI. Start from the thing you want to make — the right tool follows from the medium.

Gstaad · 2026

Under the Hood

The loop.

The loop · idea to shipped
an idea in your head
1Talk it out
2Plan it
3Build it
4Verify it
5Ship it

Then you go around again — fixing, adding, polishing. Each lap is faster than the last.

Out the other end: a real thing, live on the internet.

Every real thing gets built the same way: a loop you run over and over, getting faster each time around.

Gstaad · 2026

Under the Hood

Whose thinking is this?

The line · help vs. shortcut
an essay due Friday

The honest test: if AI vanished tomorrow, could you still do it?

You think · AI assists

  • “Quiz me on this chapter.”
  • “Explain this part I'm stuck on.”
  • “Poke holes in my draft.”

→ You come out smarter.

AI thinks · you copy

  • “Write my essay.”
  • Paste the prompt, copy the answer.
  • Hand it in unread.

→ You come out emptier.

Same tool. The difference is whose thinking happened.

AI can make you sharper or hollow you out. The only question that matters is whose thinking actually happened.

Gstaad · 2026

Under the Hood

Where this actually stands.

The map · where this is going

Superintelligence

Beyond us at basically everything. Doesn't exist — maybe someday.

AGI

Flexible across almost anything a person can do. Not here yet.

Narrow AI

You are here

Brilliant at one thing at a time. Every tool you've ever used.

We're on the bottom rung. Understanding it now is the whole advantage.

Three rungs: what we have, what's coming, and what's still science fiction. Knowing where we actually stand is the advantage.

Gstaad · 2026

Under the Hood

Same idea, three depths.

How this course works
How does AI actually work?
1

The gist

One clear paragraph. The whole idea, fast.

2

Go deeper

The how and the why, once you're curious.

3

The deep end

The real mechanics. As far as it goes.

Stop on any step. All three count as getting it.

Same idea, three depths. Read the gist and move on, or follow it all the way down — both are getting it.

Gstaad · 2026