Jeff Klune

We didn't want to accept that we were related to the other animals.

We didn't want to accept that we might not be the only people who can play chess and go.

And in the long march of kind of taking humans down a peg,

Currently, people are saying, oh, AI would never be creative at scientific enterprises, and humans are the only ones that can do that.

But actually, we have some evidence that the system can be creative.

For example, the system produced a paper that I read, and when I read it, I said, oh, that's actually a really creative idea.

And it turned out about two months later, a human team of scientists released a paper that did that exact same idea.

And it was celebrated by the community as an interesting novel study that advanced the field.

So we have some evidence that the system can come up with ideas that humans find creative.

Correct.

It independently came up with it, and we can verify that because the paper came out after our system was trained and after our system produced the paper.

That paper was not publicly available when our system independently came up with that idea.

Yeah, so the AI scientist has a complicated orchestration of a few different models under its hood.

And we were using models that were state-of-the-art at that time.

Already, the models are way better.

So even if you re-ran our current system with the current models, they would be way better.

In fact, one of the new results in the paper that we had in Nature shows that if you kind of use models later in time, the quality of the science produced by the system just goes up and up and up because the underlying models are getting smarter and better.

Not only do they not realize that, I think that most people do not understand the rate of improvement

They might think that they do.

They know that AI is moving fast, but they don't understand just how fast it is improving.