You seem to look down on those who have
1) learned from online courses
or
2) used AI on tasks that don't require it
Isn't this a bit contradictory? Or you expect candidates to have found a completely novel usecase for AI on their own?
I understand that most ML roles prefer a master's degree or PhD, but from my experience most of the master's degrees in ML being offered right now were spawned from all the "AI hype". That is to say, they may not include a lot of core ML courses and probably are not a significantly better signal of a candidate's qualifications than some of the good online courses out there.
So what does that leave, only those with a PhD? I think it's unreasonable that someone should need that many years of formal education to get an entry level position. Maybe I'm missing something, but I'm really wondering, what do you expect from candidates? I think a few years of professional software engineering experience with some demonstrated interest in AI via online courses and personal projects should be enough.