“New AI research paradigms like test-time compute, post-training, and reinforcement learning are presenting new scaling dimensions for model improvement, moving beyond the limitations of pre-training data.”