[nybsjytm]

That reflects kind of badly on NeurIPS if true. I first became aware of him over 10-15 years ago when he was pushing his "mathematical universe" idea, which is quite simply the most vapid and contentless idea I've ever seen in physics. (I don't think it's at all surprising that someone like him would be drawn to the AI community, or vice versa!)

[_delirium]

He has some PhD students doing what seems to me to be reasonable work that gets published at NeurIPS. Somewhat speculative, of the flavor you’d expect from theoretical physicists doing AI work, but it’s at least speculation about concrete technologies that exist, not pure metaphysics. For example, there’s a recent paper proposing and validating in small models a possible mechanism to explain some power laws seen in NN scaling curves [1]. May turn out to be wrong, but doesn’t seem nutty to me. On the other hand, I’d guess these papers specifically are probably not what got him famous in AI circles. For that his general self-created role as AI futurist is probably more responsible [2]. I tend to avoid that kind of stuff, but staking out a debate position on questions like “how smart could AI get? Will it kill us all?” is the kind of stuff the general public likes to hear.

[1] https://proceedings.neurips.cc/paper_files/paper/2023/hash/5...

[2] e.g. https://munkdebates.com/debates/artificial-intelligence/

[nybsjytm]

I think that paper is pretty weak, but at least in terms of content it's still night and day compared with the nature of Tegmark's clout-chasing in his physics days. But it's pretty grim if that's the kind of paper that made him popular at NeurIPS.