One tendency of the arrogantly self-taught that the poster didn't mention is that they don't know what they don't know (and aren't willing to learn, because they know it all already). Even though the majority of the course is well below their competence level, they miss out on some critical foundational things, but persist in their arrogance.
As an illustrative example, I was a graduate TA for an undergrad algorithms course this term. I spent two days after the course ended arguing about the final assignment grade with one of these self-taught know-it-alls - the problem was a dynamic programming problem which could be solved in quadratic time, and which the student had solved recursively in exponential time. He repeatedly insisted that his solution was "mostly correct", and that he should get some points for it, and I explained three times the theoretical reasons why he was wrong. Eventually I also pointed out that his solution was about one trillion times slower than my reference solution on an instance of size 50 (mine ran imperceptibly fast, his hit the 30 second timeout) - I don't even know the words for how much slower it was on the largest test case (of size 2000).
Students like that may have talent, but without the humility to realize they have something to learn from others, I'm sure they'll end up putting code like that in production and costing their employers real money.