I was a math major in college. I also spent 25+ hours per week on creative writing, although I took few classes in it and was mediocre in all the English courses I took. I'm very glad I got the mathematical background, and the creative writing experience (mostly frustration) made me a better software architect. It's harder to get that sort of out-of-band stuff once you're out of school.

Using Coursera and various textbooks to take some of the CS courses I missed in school, and doing this as an "adult student" (29) I've had a couple of insights...

* Autodidacticism (reading the texts and papers without a structured online course) is hard. How do you know if you're learning the right stuff, or how deep to go? Which exercises in your ML textbook do you do? (You won't have time to do all of them, unless you want to spend 2 years.) It is worth doing, but it's not easy. There's a reason why autodidacticism is rare and impressive.

* College wasn't that hard. It's a combination of (a) trying to do a zillion things at once, (b) having mediocre study skills on account of having not needed them before, (c) all the hormonal volatility, and (d) damaged long-term-planning facilities until age 25, that make it hard. After a few years of cubicle hell, college coursework is relatively easy for most people, if you have discipline (which very few 18-year-olds do; cube hell will help you get it). Of course, most people don't have time for more than 1-2 courses once they start working.

The problem with the "dropping out is awesome" mentality is that it's unclear what people are to be dropping into. I had that fantasy of dropping out to become a novelist-- who doesn't?-- but enough realism to know it wasn't a credible plan.

I think that most people (Zuck and Gates excluded) who drop out for a startup are actually running from something, and the thing they're running from isn't college, but the fact that 18-24 is a difficult age, especially in the neurotic U.S.