Who Is Afraid of Chinese Models?

Ben Thompson has put forward an interesting proposal: the US should adopt a law that explicitly recognizes data collection for training models as fair use and prohibits terms of service that forbid distillation (literally, API queries) for American companies. In his view, stopping distillation is nearly impossible, so the US should take a different path—adopt a new copyright policy that protects AI labs and ensures their work serves innovation for all. Thompson also suggests that Alibaba's decision to release Qwen 3.8 Max with open weights—reversing its earlier decision not to release Qwen 3.7 Max in May—may have been prompted by a recent speech from Xi Jinping, who said: "We must seize this rare historical opportunity for open source, openness, collaboration, and sharing." Qwen 3.8 Max is a new model with 2.4 trillion parameters (almost as large as Kimi K3's 2.8 trillion). In examples of its work, there's an image of a pelican, and the detailed reasoning output includes funny edits: "Add a helmet? No," "Maybe add a little bell? No," "Should I add a small fish to the basket? Not necessary."