Recently, I've been trying to teach myself some numerical thing I don't have prior experience on (building a sparse LU solver, if you're curious). What I've found is that the single most useful resource I've found is not trying to build one myself (because where do you begin?) or by ripping apart the internals of existing solvers to see how they work. No, it's stumbling across the lecture notes of a course that covered that material, in no small part because by stepping up a level and looking at the other lectures, I can discover the other relevant things I didn't know were relevant. And this property has held true in my experience for other topics I've had to research on my own: the highest quality materials are invariably university lecture material.

If your main thesis is that university instruction isn't worth it, why is all the best material university instruction? Sure, there's an argument that learning how to build something is best done by actually building it... which is why university courses invariably have "build what we're teaching you to build" as a course project that is a significant portion of the grade.