If the director of alignment at Meta Superintelligence Labs gets it wrong, what hope does anyone else have?