I could see their comment on user training data as a bit of a CYA statement, but removing Alpoge from the paper is awful. Has really soured what could have been a huge moment for AI progress.