I'm constantly learning new things that to people from other fields would be freshman-level trivialities! The most recent such example would need to be something I learned today. For example, maybe a little tidbit that I picked up from Sanjeev Arora's STOC'2018 tutorial on deep learning (where I was this morning): it turns out, backpropagation is just a dynamic programming algorithm for calculating a gradient in time linear in the size of the network, improving over the quadratic time that one would've needed naively.

I feel bad that I rarely have the time (or willpower!) to do exercises anymore, though I'd probably benefit if I did.