Unlearning Needs to be More Selective [Progress Report] — AI Alignment Forum