Do model evaluations fall prey to the Good(er) Regulator Theorem? — LessWrong