There is a particular feeling in meetings about AI.
Everyone is nodding. Someone says the models, someone else says at scale, and there is a shared understanding in the room that everybody present knows what is being talked about.
You may have quietly suspected that nobody does. I think you are largely right, and I do not think that is anyone’s fault.
The thing that helped me most
It is not technical.
In the 1990s, people built a new kind of computer chip so that video games could draw explosions quickly. That was the entire purpose. Fast pixels, for fun.
It turned out — years later, and to nobody’s plan — that a chip designed to paint an enormous number of pixels at the same time is almost exactly the shape you need to train a machine to learn.
The whole industry now rests on that. Trillions of pounds, every headline, every worry, every board meeting you have sat in this year, downstream of a part designed to make spaceships look good.
That is the shape of the whole thing
Three completely separate efforts were running at the same time, in different buildings, for different reasons, and none of them was trying to build what we ended up with. The chip people wanted better graphics. The games people wanted to see whether a machine could learn to win at something. And all of us, without anyone deciding to, spent thirty years typing everything we know into one place where a machine could read it.
They met. What came out was this.
Why it helps to know
If nobody designed it, then nobody can tell you with authority what it is. Not the vendors, not the sceptics, not the people who sound most certain in the meeting.
Which means the question you have been slightly embarrassed to ask is actually the right one, and asking it out loud is worth more than nodding.