Curious people got high on the caffeine and code, and the other group just got high.
"Locking-in" on some esoteric topics just because one wants to is probably the best way to improve in my opinion.
On the other hand I have been through a pretty generalist undergrad education in engineering and it helped in acquiring general math, physics, chemistry knowledge.
I was already a nerd when I was attending the university. My chance was my instructors and professors let me roam free and convert every assignment to a rabbit hole. Poisoned by demoscene earlier, I was already an elegance/performance freak before even I got graduated, and this made me dive into unknown territory head-on instead of being afraid of it.
I have written a compression algorithm from scratch for graduation, then written a multi-agent trading system for M.Sc. project. For Ph.D. I have written a BEM based material simulation which was able to saturate the system very effectively, reaching practical IPC limits of the processors it ran on.
> Self Taught -> Already is curious, be default, so 99%-100% are curious.
There is some of this but IME that % is half your estimate, if not lower.
It could be different in today's market because CS IS NOT where a lot of jobs are, and my observations are based on when it was very very hot, so there's that.
But, no matter what you or I say anecdata is not data, but in my 30+ years of doing this, I find these broad generalizations mostly wrong.
Self Taught -> Already is curious, be default, so 99%-100% are curious. Just by the nature of going out of the way to learn something.
Educated -> Also has a curious % in the group, but also has the large group of complacent "i heard software makes money so i'm hear to learn it, but i don't know what i'm doing". So when you do find complacent, they mostly came from the educated path.