I stop when I find a solution to the problem. Most of the time the learning happens along the way, not necessarily in the solution itself but all the things you try and iterate on your journey to the solution.
Everything changes in tech by the minute ... but also nothing changes. For web applications it has been HTML, CSS and JS for nearly 30 years. XMLHttpRequest/AJAX came out 25+ years ago. There have been many improvements along the way, like applying design patterns instead of cgi-bin directories with scripts that had a +x modifier on them in the file system. But the base technologies have not changed all that much. We still submit HTML form's with input fields to a back-end server that handles that data. We're still rendering HTML and using CSS to style it. Gone are custom UI toolkits like Flash or Java Applets. Maybe WASM is something to look into but it feels like its not mainstream to me.
If you don't want to get left behind, learn the basic building blocks at a deep level, they don't change much.
> I stop when I find a solution to the problem
That.
I’ve embraced that I cannot direct which topic my self-directed learning will take up and sustain (almost none of which ends up going toward tech stuff, aside from a span of some years in my teens and early 20s—and all of that was motivated by wanting to accomplish specific things with computers) and rely on assignments to motivate me for career-relevant learning.
I’ll learn whatever it takes to get the job done. Then stop, because I don’t actually care about the tech per se, most of the time, and trying to force myself to learn “just because” does nothing but make me miserable and waste time.
This isn’t even “how I approach learning as a generalist”, it’s how I became a generalist.
My interest in continuing to fiddle with stuff after the job’s done is basically zero.
My experience has been that it takes amazingly little effort to be above-median among practitioners at a lot of things. How many React developers have spent one entire hour reading through the core logic of React itself? How many people working with LLMs have read the Attention Is All You Need paper? How many people read about the disk storage layout of a database they’ve been told to use? It’s way less than half. It takes so very little to stand out.
That's funny, I often love learning for the sake of.. experimenting, building things. I just recognize that I cannot do it all and prioritize immediate needs first.
> That's funny, I often love learning for the sake of.. experimenting, building things. I just recognize that I cannot do it all and prioritize immediate needs first.
You've misunderstood, I think: if I have something I want to experiment with or build, then I can learn what I need. What I have such enormous trouble doing that I'm nearly incapable of it, is going "I should learn this so I can do something, to be determined later, with it..." and then learning. Making up projects for the sake of learning, when I don't really care about the ends, also doesn't work.
I need the goal first, then I can learn what I need.
I do also learn things just because I want to and for no real purpose, but have never been any good at directing that impulse.
So what happens when you can't find a solution to your problem and hit the edges of current research even specialists cannot buckle?
this is often a sign that the design/solution you've chosen is an unfruitful path.
know when to cut your losses and try something different.
I say, "you're going to need to pay a team of PhDs to work on this to maybe get a solution, eventually, or change your approach. Here's why: [evidence]"
I've never worked with or for anyone who wanted to hire the team of PhDs.
I go for "I can understand experts, but not add much to the conversation" as a benchmark for knowing enough to participate in discussions at work. Then I use that "I can solve my immediate problem" method going forward.
I do a lot of repairs on electronics. I stop when I understand enough of the theory of operation to get things fixed. I've worked on everything from industrial control like 480 volt 3 phase SCR packs to Cesium beam atomic clocks.
If it was interesting, I'll dive in to learn more in case I encounter something similar in the future.
I concur with everything here but I would say find a solution to your problem plus a little bit more. In particular, try to find out about where the boundaries of the problem are such that this solution wouldn’t work anymore. Maybe it stops working at a certain scale, or with less behaved input, or if you need to support Chinese characters. This helps me really understand the solution, and not feel like I just have a book of incantations.