Despite the catchy title, that's not what the paper says. It says "[large reasoning models] reasoning effort increases with problem complexity up to a point, then declines…they reason inconsistently across puzzles" Is human reasoning effort ever inconsistent?

→ View original post on X — @erikbryn