The gist of s1: simple test time scaling

from blog Xe Iaso's blog, | ↗ original
TL;DR: when the model thinks it's done thinking, tell it to think some more. Yes, really.