“Two new scaling laws, reinforcement learning with verified rewards and test-time compute, have been the primary drivers of AI progress since October 2024.”