“Ben Thompson has argued in his "Who's Afraid of Chinese Models" article that distillation is becoming more impactful as model training shifts towards Reinforcement Learning (RL).”