What's this about?
Ethan Mollick reads the accumulating signals of tension around open-weights models: the US announcing it reserves the right to act against models distilled from American ones (and claiming Kimi distills), while contradictory reports about export controls come out of China.
What exactly is "distillation"?
Training a new model to imitate a stronger model's outputs — the student copies the teacher's homework at scale. The US claim implies Chinese open models free-ride on American frontier models; researchers like Nathan Lambert counter that's technically implausible at RL scale (you can't afford the teacher's API bill).
Why is it worth noting?
Every strong open model is Chinese (Kimi, DeepSeek, GLM, Qwen); if both governments restrict the pipeline — the US on distillation, China on exporting weights — the open ecosystem gets squeezed from both ends.