Gurkenglas comments on “Dirty concepts” in AI alignment discourses, and some guesses for how to deal with them