To stay with the paint analogy, AI is more like a paint roller that can paint by itself.

Just tell it what color you want the room to be, and walk away. Did it remove the original coat properly? Sand, prime, and double recoat? No idea, it looks good at the moment. But we'll find out in a couple years when the cracks and bubbles start showing.

Eh, not really, and it isn't really an argument but more of a lighthearted prediction.

I think the big question is whether or not the "frontier" models continue to be available and evolve, because the business model for running them seems very unsound.

I kinda hope the AI bubble busts, and that afterwards some of the efforts turn toward making open source models more performant and powerful.