[dig1]

1) Cryptography done right - instead of blindly reading & using algorithms, figure out how people got there. Requires some math background, progress is very slow, but the reward is many "ahaaa!" moments.

2) Trying/inspecting all sorts of Lisps out there. There are so many crazy ideas tried and implemented over the last 40-50 years that IMHO, any new modern language feature probably already exists in some form in some Lisp compiler or interpreter.