Self-taught dev here, and I remember feeling similar things at a similar age. During a gap between jobs I wound up taking three online courses: Algos 1 and 2, and Andrew Ng's introduction to ML course (all at Coursera). This was just out of curiosity, I didn't have any particular goal or expectation for them.
At the time I didn't feel like I learned any One Big Thing, but in retrospect the courses demystified a lot of things, and I've never really felt out of my depth with development since. The other big thing I did was to tackle a much larger, more complex project than I'd tried before (a game engine). That also forced me to grow as a developer, but I don't know if I'd have managed it (or tried) without the footing I got from the courses.
I still got nothing out of NGs courses. I vaguely remember something about matrices, how to decide how many parameters to change when training, and some other very-vague understanding things. I don't really understand why they are "the gold standard", but I also don't do any ML "irl".
I took the Ng course when it first came out, but I hope it's been updated since - it was taught in a language I've literally never even heard of outside the course (GNU octave).
That said, if you want to understand ML stuff I learned waaaay more from Andrej Karpathy's famous video series. But I still found the old Ng course useful in a "ah, so this is what they teach in a college intro to ML course..." way.
At the time I didn't feel like I learned any One Big Thing, but in retrospect the courses demystified a lot of things, and I've never really felt out of my depth with development since. The other big thing I did was to tackle a much larger, more complex project than I'd tried before (a game engine). That also forced me to grow as a developer, but I don't know if I'd have managed it (or tried) without the footing I got from the courses.