“Chinese labs are using distillation strategies, including using LLMs as judges in RL post-training, to fast-follow American frontier models.”