I'd still argue that there's a lot of value in some form of teaching (via book or video or whatever). If left to their own devices, novices will often come up with messy, non-idiomatic ways to do things.
I find using both in combination is usually the best way to go. If you're just reading a book and doing exercises in it, you wont learn as well. Making something you want to make and using books as reference while developing makes you learn a lot quicker.
Also, the answer to this question depends a lot on your previous experience with other languages. If you're well-versed in another C-like language, jumping right in and starting to make something will probably teach you a lot, really quickly. If you're more of a beginner in programming generally, you need a book or a teacher.
Not learning from books/teaching can make it hard to improve. I love reading other peoples code and thought processes because it often reveals how I'm missing something in my toolbox.
Studies have shown that one of the best ways to remember is to take a stab at a problem before learning the solution. So even quizzes on using libraries has helped me learn methods and concepts I would have not learned on my own if left to my own way of coding. I try a simple problem on my own before learning that there is a cleaner, more efficient solution. It's easier and more approachable to do that with small problems than trying to revise an entire document of code.
Lastly, I think a lot of experts underestimate the gap between novice and intermediate, where it's really hard and frustrating to attempt things that are a bit more complicated, requiring more background knowledge than just whats shown on the surface.
Ya, I think a balanced approach is the best, begin by reaing or watching videos, but not too much, just to get an overall idea, then get hands on but don't stop there, read more code and watch more videos and continue the process for your life time!
I sometimes get carried away and stop reading more after I cross some threshold which prevented me from noticing lot of nice things like best practices that were happening in the industry/or better design patterns etc..
Yeah, the trick in my opinion is to find a good enough book or guide on new features with which you can skim and familiarize yourself. If you don't understand when you might apply something, go figure that part out. Then work on a project and when it seems like you might need that concept, go read about it in more detail. In terms of real world work, having a very broad, surface level knowledge of many concepts and when you might apply them often comes in more useful than a very in depth knowledge of a few specific concepts.
