“Competitive advantages in reinforcement learning for LLMs typically come from practical tradecraft like better infrastructure and higher quality data, not from secret proprietary methods.”