Demis Hassabis posted at 2 AM on Thursday—not unusual for him. “Reality is still staring me in the face,” he wrote. For anyone who’s read Sebastian Mallaby’s new book The Infinity Machine, that’s less a late-night musing and more a status update. Mallaby quotes Hassabis describing the feeling as reality “screaming” at him, trying to tell him something if he could just listen hard enough. He’s been hearing it since he was a kid. DeepMind, AlphaFold, the Nobel Prize—all of it is, in some sense, his attempt to write back.
Elon Musk replied with a counter-argument: the era of foundational discovery is essentially over. Newton, Einstein—those moments happen maybe once more, he wrote.
After that, intelligence—human or artificial—shifts almost entirely toward creation. His reasoning: the Standard Model already accounts for quarks, leptons, and photons. The map of reality’s basic rules is, more or less, drawn.
Hassabis disagreed, and the distinction is worth paying attention to
He didn’t push back on the creation part. His point was narrower: AI might still find what he called “elegant and compact descriptions” of the universe’s remaining mysteries—quantum gravity, the nature of time, things physics has circled for decades without closing in on. The argument isn’t that humans will crack these alone. It’s that pattern-matching at AI scale could get there in ways no individual researcher ever could. AlphaFold is the proof of concept. Protein folding stumped biology for 50 years. A model solved it.
The real debate isn’t about physics, it’s about what AI is actually for
Musk sees AI as a creative engine. Hassabis sees it as a telescope pointed inward at reality itself. Both framings can coexist, but they lead to very different research priorities. One reply on the thread captured it better than either of them did: Hassabis is a man who built a billion-dollar company as a side effect of trying to make the haunting stop.
