Let's face it. Chinese labs made some of the biggest advancements. AND training on Claude (or GPT) output IS unreasonably effective. The two sentences are true at the same time.