Will Scaling Reasoning Models Like o3 and R1 Unlock Superhuman Reasoning?