Brian Greene puts a hard ceiling on the assumption that AI scaling is automatic

Brian Greene, the Columbia University physicist and author, is best known for walking audiences through the deep future of the cosmos: the Sun swelling to consume Mercury and Venus around 10 to the 11th years, protons disintegrating somewhere around the 38th floor of what he calls the universe’s timeline, black holes evaporating long after that. The framing is meant to put human concerns in perspective. One of the concerns he addresses is what artificial intelligence can actually deliver.

It's not obvious to me that it's just a matter of pumping billions and billions of dollars, building ever larger data center, putting them in space, wherever you're going to do, and the more money you put in and the more data you have and the more computer power you have, the better it's going to be. That's not obvious to me that that will continue up that exponential curve. Brian Greene

Greene’s skepticism about large language model scaling sits in that same contrarian tradition. The prevailing assumption in the tech industry is that more capital, more data, and more compute translate reliably into better models. Greene disputes that the curve continues indefinitely, calling it “not obvious” that exponential returns follow exponential investment. He does not claim the scaling hypothesis is wrong outright, but he treats it as unproven rather than settled.

The point lands differently coming from someone who elsewhere argues that free will does not exist and that all human emotion reduces to electrical impulses in the brain. Greene’s worldview is relentlessly mechanistic, which makes his hesitation about machine intelligence worth flagging: the skepticism here is not romantic or protective of human exceptionalism. It is a straightforward claim that the engineering case has not been made.