This is a style choice so that GPs are written in PyMC the way they are written in math. If you read a GP paper, one doesn’t write something like: y = f(k_1(x, x^\prime), k_2(x, x^\prime), ...), one instead directly writes y = f(x, x^\prime), abstracting away the specific kernel functions (or functions thereof). @bwengals is more qualified to give a more detailed explanation of what choices were made and why, as he wrote the gp module 
Yes, I believe this is true. Can you re-do the analysis the correct way and check? We’re working on adding an upgraded version of these notebooks in this PR (here is the updated GP lecture). If there is a big error, it would be good to fix it before we merge.