He does an interview with the guy in the next video and there's a lot more to this. The elections aren't binding and have never been, simply because they were done out in the open at a conference. They didn't want some company just filling ever seat.
https://youtube.com/watch?v=MNsgtOQA048
It seems like the current beef is with the definition of open source AI models. This is something I've been saying since Deepseek came out. It's
NOT open source! Open source would mean having all the training material (terabytes of it) so you could train with the same data to get a binary model. Would it be the same model? No. Random variance will mean you'd get a slightly different model, but it would be pretty close. Could you build your own model? Deepseek used 2048 nVidia N100s, which are $8k~$10k each and took 3 days. Could you use a couple of 3080s and it just take a few months? Probably not. We're taking $8mil ~ $10mil worth of hardware (or maybe less for shared hardware rental time). It would take someone with a homelab years, or even a decent university high end lab months.
None of the AI models are "open source," and most of them can't be because they're trained on a corpus of copyrighted works (both literary and images) so the creators literally cannot share the source material unless they want to be hit with massive legal liability (just look at the Meta case where the stand accused of torrenting terabytes of commercial material).