However, that is going to tend to teach you things that are narrowly focused on the needs of that project.
For example, there are several occasions in my career where it would have been very helpful to know more about statistics than I do. I have learned some by taking open source projects that needed to do the same kind of statistical test I needed and studying their code, and adapting it for my needs (unfortunately my needs were sufficiently different from those of the open source project that nothing worth contributing back came out of my adaptations). However, what I learned there was how to do certain specific things in certain specific situations. I did not learn the boundaries of when those specific techniques are appropriate, nor did I acquire an understanding of WHY we use those techniques.
It would be much more productive overall if I were to sit down with a good textbook on statistics and study it, cover to cover, as if I were back in school. That will overall improve me in this area far more than looking at open source projects (or closed source projects) that happen to do some statistical calculation.
I aspire to be more than a coder just coding up implementations of things where others have done the deep thought. I aspire to be like a Knuth or a Dan Bernstein, who can code well but who has a deep and broad understanding of the theory behind what he codes and why it is being coded that way.
If you stick to people who do a lot with open source because open-source measurement is all you can do, you're going to miss out on people who have been studying certain deep, interesting things, because while open source is great for general-purpose computer programming practice, it can be a lousy place to play with advanced, in-depth computer science.
For the projects I tend to work on, reading 100s of pages of documentation is necessary. To create a good solution, I must understand the domain well. I spend many evenings reading books on kernel construction, networking, algorithms and so on. This is improving me technically. When I go into work, the knowledge then helps to make better software.
But the GP post seemed to be arguing the opposite: that focusing your effort and learning on a practical goal was not a way to "improve yourself technically". And that's the part that lost me.
About ten years ago I read a bunch of books on MPEG2/4 and some other video-related stuff. I wasn't doing much in the field, it just seemed interesting. Eight years later it turned out that the project I was on needed this type of expertise, so I was able to contribute on stuff that people hadn't expected me to be able to. It was fun.
You can't easily foresee this kind of thing. I'll probably never need to know the details of how Vax/VMS did its free page management, but I might run into a related problem, so reading that OS book 25 years ago might pay off.
What I see from people who /just code/ is that their skills rot far faster than people who like to learn stuff. If all you do is code, after ten years you have another 100KSLOCs of Java or C++ or SCHLOPBOL under your belt . . . and who cares?
I like to learn stuff and hopefully get a chance to apply it. Otherwise I'm just a garage mechanic who fixes bicycles in my spare time.
And in any case the subject at hand is identifying good developers. They aren't paying you to read books, they want to know if you can produce software. If you want to show them that you can, then producing software is an awfully good way to do that.