[ZetaZero]

Agreed. AI art will continue to improve.

This reminds of of the chess computers from 30 years ago. Experts were convinced computers would never beat GM humans, based on the state-of-the-art of the time. They never took into account all the future advancements in hardware and software.

[vikingerik]

The chess experts were wrong qualitatively. They were saying that best play was about intuition and feel, that computational power would never replicate regardless of its scale. (Data on Star Trek always lost chess matches implausibly.) Turned out that brute force calculation thirty moves deep does in fact outperform anything a human can foresee.

The same thing happened for both Go and Starcraft in the next decades, the experts said computers couldn't replicate enough spatial feel, and then they did. And now it's happening for AI art. Enough computational power and a sufficiently well-trained neural network can indeed exceed anything a human can do.

AI has been roughly doubling in performance every year or two, for quite some time. We just never noticed when it went from 0.0001% to 0.0002% of human capability. This is the year that it doubles from 10% to 20% and everybody notices. And there's not a lot of doublings left until it shoots past 100%.

[kridsdale2]

I eagerly await a fully ML generated ballet choreography with a dozen participants. Maybe 15 years from now.

The super hard problem is driving a robotic body, vs rendering an animation of the above.