I dont understand one thing: When United States says it will stop/ban chinese open weight models, how do they plan todo that. Hypothetically speaking, lets keep morality aside, is it technically even possible. One might say, we would stop huggingface and other model hubs to stop models to be downloaded by anyone in the US. How is it even possible, i can think of multiple ways to bypass it even when a model hubs put such a restriction. And for model inference providers, they might try to prevent people from using those models in the US, but again there are multiple ways to trick those “model inference providers” as well. Not to say, what if someone goes to mexico/canada, downloads the model and comes back to the US. Will border patrol check that as well. One might say, that this is for the regular Joe not for people in technology. To that my question is, how many regulsr Joe pay money to ChatGpt/Claude, they use it for free anyways. For companies hosting AI for there workforce/customers, they can always host it somewhere else and provide it to there employees/customers naming the model “Definitely NOt Chinese”… model OR even better approach would be to put some layers of fine tuning and say that it is build inhouse. My professor told me 10 years back that they tried todo the same thing with the Internet and they failed. Uncle SAM doing what Uncle SAM does best!!!!

Overall my take is that this is mostly political posturing to signal to the market what it wants, regardless of how difficult it is to actually enforce (which, as you point out, is pretty hard to do), in order to justify something like the US govt taking ownership stake in frontier labs (OpenAI, etc) This may have downstream implications on what the US may require platforms like HF to do in terms of gating model hub downloads in a best-effort kind of way, but it’s still unclear how far they’ll actually go to enforce it.