zlacker

>>nl+(OP)
An Erdos problem is great, but all major LLMs still score 0 out of 7 on all Millenium Prize Problems[1]. I, and I'm sure other like me, won't consider it "true" AI until AI saturate the MillenniumPrizeProblemBench.

1. https://www.claymath.org/millennium-problems/

1.

>>kelsey+2o4
That's interesting, because what I'm looking for from AI is the opposite. Rather than looking to do something humans find hard, I'm more interested in the problems that humans find easy, like driving a car or counting the number of "r"s in a word.

Having it go beyond that to do things that we can't do would be a nice bonus. But given the way AI has been going, I'd say there's a decent chance that it solves a Millennium Prize problem before it solves driving.