They’re all trained on copyleft data, so they should by all rights be open source. Distilling them into open weight models is the next best thing, I suppose.
You can go on GitHub right now and get some specialty made distilled models. One is an extremely good bash script model that is only 90% chance of nuking my computer. Only joking a bit.
It may take more time, but everyday people can make their own models now. I’m not sure putting a ban on any country is going to do much.
He makes good arguments, could also add that it would be a source of token sales for the big labs.
but with some misplaced assumptions.
- Open source shouldn’t discriminate on contributors or nationalist biggotry of permitted users.
- Open weight models can/are being served from US datacenters.
Cursor was bought by mechahitler for $60B because it made an LLM, by taking Kimi 2.5 source, and retraining with free user data that was mostly distilled from frontier closed but other global models as well.
We would all be better off if we collectively pretended that Tan didn’t exist.
They have no incentive to do this. Indeed, the Chinese companies are actually going towards a proprietary model in order to make money. So, what he’s suggesting whilst sensible from the sense of what they are based on and the public good does not actually follow. The economics, which I guess is just indicates how broken the system is.
companies cannot be allowed to have ownership of something trained on stolen property. not only should they be distilled, they should be forced to open the weights.
Broken clock…
