This is basically my procedure for learning a new programming language. I have a list of things that I like to have on hand that cover a fair amount of possible paradigms, and re-implement them. Things like FFT, linear algebra methods, numerical root finding/optimization algorithms, etc. Project Euler (and now also exercism.io) are great for finding small but digestible things to do too.
At the moment I'm working on finding some better mathematical examples for class abstractions & better use cases for object-oriented concepts in general. Because it looks like I'll be teaching a course either in Java or C# next fall. Coming from a scientific computing background the usual use cases for java and .net aren't my typical problem areas of expertise (though there are some things I definitely do appreciate about them).
Ah the frustration when someone was generating primes with a worse approach than I had, and my code was 2 or more times slower.
That's basically how I learn new technology is to either partially or fully port my Proximity game (http://briancable.com/proximity) to it. I've used that to learn C#/XNA, Objective-C/iOS, Java/Kindle (sadly never finished, but it looked beautiful on a Kindle), and most recently Pico-8/Lua (still in the process). I'm planning to start working on a 3D version soon, as well, with the goal of getting stronger in my 3D programming and getting it onto a bunch of platforms with a single codebase (using Unity).
Considering the game has been used in at least one book (Actionscript 3 Design Patterns) as a 'learning to code' game and I periodically get emails from students whose teachers have made an A.I. programming exercise for Proximity, it probably makes sense that it's my go-to learn new tech game, considering I've made other games too but don't port any of those.
I think the higher up the stack you go the easier it is to lose appreciate for all the lower layers that facilitate your stack. And over time I fear new generations of programmers are losing touch with what the basic foundation of a computer is. Sure they will study theory but do they really know how many man hours are supporting their stack?