DeepSeek-R1 and the Rise of RL Reasoning, Inference Compute Overturning the Pretraining Scaling Race | AI Insight Note