At this rate, compute exceeds universe atoms by 2100.
A post from @willccbb is spreading a striking claim about future AI training: that by 2100, training runs could use more FLOPs than there are atoms in the universe. The public evidence visible here is still narrow, and the source material in the post points readers to an image rather than a fuller underlying write-up. For now, the safest read is the attributed claim itself, who put it in front of users on X, and how little independent confirmation is visible in the conversation so far.
Combined views
3.3K
2 posts, first seen 1h ago
At this rate, compute exceeds universe atoms by 2100.
A post from @willccbb is spreading a striking claim about future AI training: that by 2100, training runs could use more FLOPs than there are atoms in the universe. The public evidence visible here is still narrow, and the source material in the post points readers to an image rather than a fuller underlying write-up. For now, the safest read is the attributed claim itself, who put it in front of users on X, and how little independent confirmation is visible in the conversation so far.
pretty crazy that by the end of the century we'll be doing training runs that use more FLOPs than there are atoms in the universe
