Post

I thoroughly recommend reading all of Cory Doctorow's recent speech on AI skepticism, it's crammed with new arguments and interesting new ways of thinking about these problems pluralistic.net/2025/12/05/p...

it’s a lot more cogent than most AI critiques I read, although I think he’s way off about the “when the bubble bursts it won’t be cost-effective to run frontier models” this strikes me as pretty blinkered; inference is cheap!

When the bubble bursts it won’t be cost-effective to TRAIN frontier models.

Absolutely... But you can't freeze frontier models where they are today and still get sufficient profits to justify existing capex. You have to train new ones sufficiently better than existing ones to justify capex, and once the bubble bursts these hype men won't be able to.

ok, but that’s not really the point i’m making. nothing really justifies current capex — my point is that the bubble popping isn’t gonna do much to inference other than make it cheaper.