BTW, this has been a thing since at least 2005. There was a paper about being able to determine what people were saying over encrypted VoIP traffic. https://web.archive.org/web/20161107231048/http://www.cs.jhu.edu/~cwright/traffic-viz/
I started an open source implementation of this a few years ago. https://gitlab.com/adam949/eyeofprovidence
Call it "AI" if you want, but it's certainly not new. In any case, the threat is real, and the mitigation they describe is equally real.
I'd like to see someone show some graphs of network traffic in practice without any VPN, with a normal VPN, and with their VPN with these mitigations. It's one thing to read a description about these things, it's another to test it out for yourself!