Who's Afraid of Chinese Models?

The Specola’s telescope log records a July 20 dispatch from Simon Willison, pointing to a proposal published the same day by Ben Thompson on Stratechery, titled “Who’s Afraid of Chinese Models?”

Thompson’s proposal centers on U.S. policy toward AI model training and distillation. He argues that American AI labs have adopted an inconsistent position, restricting others from distilling their models even as those same labs trained on data gathered without licenses. Thompson proposes a legislative fix, beginning with a law that would explicitly establish that collecting data to train models constitutes a lawful activity. Willison’s post excerpts the opening of this proposal, framing it as one path toward resolving the contradiction while also strengthening the competitive position of American open models against Chinese counterparts.

Willison characterizes the piece as “interesting,” noting it attempts two things at once, addressing the hypocrisy in current distillation restrictions and offering U.S. open-model developers a clearer footing against rivals originating in China. The full text of Thompson’s argument, including the specifics of the proposed legislation, appears in the linked Stratechery article; Willison’s commentary serves as a pointer and brief endorsement of the proposal’s framing rather than a full restatement.

No further detail on congressional prospects, industry response, or implementation mechanics appears in the available material.