Stephen Witt

You just didn't go there with money.

And every AI scientist I worked to for that time was like, yeah, I just kind of figured I would do what all AI people did, which was like deliberately earn less money for the rest of my life so that I could work on the thing I was actually interested in and I'd be a research academic.

The guys who really invented the modern neural net had so little money that they actually had to pool their housing stipend together to buy two Nvidia gaming boards to train the first neural network.

And that's the first really kind of like modern neural network.

So that's where this stuff comes from.

Jensen was serving this market segment that was so small that it was essentially invisible, right?

He's making these gaming cards, okay?

The original use is to speed up your frame rate in Quake or in Halo.

And the reason you do that is that if you're shooting at the guy, you want a few more frames per second so you can get a better bead on him.

That's the technology that powers all of society now.

That's serious, that's why they did it.

But this cutting edge use case had a lot of other applications.

And one of them was kind of scientific computing or what we call supercomputing, right?

Now Jensen basically shipped all these cards with a second program on it where you'd flip the switch and turn your gaming board into a low budget supercomputer.

So who is that for?

Well, it's not for an established research scientist, right?

Those guys can afford time on a conventional supercomputer.

It's for a scientist whose research is out of favor from marginalized scientists, right, with no funding.

It's for a mad scientist.

And that's what these guys were.