“OpenAI's approach to AI alignment aims to learn the collective value preferences of all users, rather than having safety and value decisions made by a small group of experts.”