What's this about?
Two threads on the same theme. Ethan Mollick observes that every frontier open-weights model is now made in China, with no US incentive and no EU appetite to build one. Separately, Nathan Lambert pushes back on the popular claim that Chinese labs get their quality by "distilling" from Anthropic's Fable during training — "that's not how distillation works; you can't afford to use Fable as a grader."
What exactly is "distillation"?
Training a smaller/cheaper model to imitate a bigger one by learning from its outputs — like a student copying a master's worked solutions. The accusation is that Chinese labs secretly learn from US frontier models this way; Lambert's point is that the economics and mechanics don't support that story at RL scale.
Why does it matter?
It reframes two loaded narratives at once: who actually leads open models (China, uncontested), and whether their gains are "borrowed" (per Lambert, no) — both central to the US–China AI-policy debate.