Ramez: AI reached superhuman math faster than expected, but Moravec saw it coming

sebkrier · x · 2026-11-21

Ramez concedes he didn't expect AI to become superhuman at highly verifiable tasks so quickly, notes Hans Moravec roughly predicted it, and that "we can all learn from the real world." He's responding to Arpit Rage's critique that few people predicted ex ante that math would fall fast — the verifiability argument is true, but partly ex post cope. A substantive industry debate on how AI capability gains unfold by domain.

Original post →

More from AGI Musings

AGI Musings channel →