I wonder what the authors would think of the static evaluation functions used in the strongest engines today. There is a hand-crafted version that stockfish can

I wonder what the authors would think of the static evaluation functions used in the strongest engines today. There is a hand-crafted version that stockfish can use, but we usually use NNUE, an empirically derived neural network far more opaque than the simple formula listed on page 5 of the paper. On the one hand training NNUE is an engineering marvel, but on the other hand the lack of interpretability hides the magic a bit.