I’d guess that is what you get for training them on source code.

Also the I vs E axis feels meaningless in this case.