Defeating Goodhart and the "closest unblocked strategy" problem — AI Alignment Forum