In my personal experience, this isn't a question of either/or, it's a question of order. That is, courses are fantastic, but most of the time, they're more beneficial after you've done some JIT learning.<p>Fastai, who in my opinion create the best educational resources in the deep learning space, are a great example of this. They use a metaphor involving sports to explain their approach. You don't learn to play a sport by pouring over rulebooks and studying professionals—you start playing. However, as you get a feel for the sport, you then find yourself in a position where learning a lot more about the theory of the game would make a huge difference to you, and that's where courses come in. Similarly, they get you off the ground quickly with some basic intuitions around deep learning, give you a sandbox to do some JIT learning, and then begin layering in deeper concepts.<p>I've had similar experiences in my life, where taking graduate level courses was incredibly rewarding specifically in areas where I'd already spent a decent amount of time doing JIT learning—and arguably, where I'd reached a relative limit in how far JIT would take me.