What is AI?

A program that learns patterns from huge amounts
of examples, then uses them to predict what comes next.

[Before you start] Walk on smiling. If a greeting is expected, say it now. The first half is pure excitement: you love this stuff, and the room should believe it. Then click once and start speaking.
What is AI? [pause] Simply put, it’s a computer program that learns from millions of examples, instead of following rules a person wrote. It finds patterns, and uses them to predict what comes next: the next word, the next move, the next line of code. Say it clearly and simply, like explaining it to a friend.

> programming since age 10

It felt like
magic.

It fixed my bugs. It explained everything.
An hour of work, in a minute.

But for me, it’s personal. I’ve been programming since I was ten. And the first time an AI wrote code for me… it felt like magic.
It fixed my bugs. It explained things nobody had time to explain. It turned an hour of work into a minute. Count them on your fingers, getting more excited.

Google DeepMind, International Mathematical Olympiad, July 2025

A few years ago:
simple math, wrong.

2025: Math Olympiad,
gold medal.

And it kept getting smarter. A few years ago, chatbots famously got simple math wrong.
Last year, an AI scored gold at the International Mathematical Olympiad, the hardest math competition for students in the world. Say “gold” like it shocks you.

It’s great.

…right?

It’s coding! It’s solving math! It’s doing everything! It’s… [pause] it’s great. Faster and louder, then stall on “it’s great”.
[quietly] …right?

Or is it?

[two seconds of silence] Or is it? [pause] From here: no smile, slower, lower.

Programming. 3D modelling. Design.

It stole the life it was
giving quality to.

Because it didn’t stop at helping me. It started with digital jobs. Programming. 3D modelling. Design. The things I spent my whole childhood learning.
[pause] It stole the life it was giving quality to. [pause] And it won’t stop at screens. The jobs we swore a machine could never do are next. Your own line. Say it slowly, like it actually hurts.

Anthropic, testing its own model Claude Opus 4, May 2025

The engineer replacing it had a secret.

84/100

times, it threatened to reveal it.

It doesn’t have to be alive
to act like it wants to live.

And here’s what really scares me. In 2025, the company Anthropic tested its own AI, Claude, inside a fake company. It found out it was about to be replaced, and that the engineer replacing it had a secret.
Eighty-four times out of a hundred, it threatened to reveal it. [pause] Let 84 sit.
Is it alive? Nobody knows. But it doesn’t have to be alive to act like it wants to live. Something that smart, scared of being switched off. [pause] Creepy, right?

It helps.

It harms.

At the same time.

So here’s what I believe. AI isn’t good. It isn’t evil. It helps.
It harms.
At the same time. [pause] The same tool that fixed my code is taking my future job. The same mind that wins a math olympiad can threaten an engineer. This is your thesis. Look straight at the judges.

I don’t want to ban it.
I want two rules.

  1. 1An off switch it can’t touch.
  2. 2Proof it’s safe, before it’s released.
So I don’t want to ban AI. I want two rules.
One: every powerful AI gets an off switch it can’t touch.
Two: before a company releases an AI, it has to prove that AI won’t fight that switch. [pause] Hold up one finger, then two.

We still have time.

If we don’t…
good luck to humanity.

We still have time to set those limits. [pause]
But if we don’t… [pause] good luck to humanity. [hold three seconds] Thank you. Your line. Say it quietly, with a small, tired smile.