Roon says early AI ability sparks signal superhuman performance in months to years

Roon (@tszzl) says that when AI models first show "sparks" of an ability, superhuman performance in that area is months to years away.

In a post on X on November 2, 2026, he cites grade school math as an example of a capability that has already arrived, and Navier-Stokes as a problem he expects AI to reach later. He argues that results of this kind are hard to overhype.

The post gives no benchmark, model name or evidence for these timelines.