Which values are stable under ontology shifts? — AI Alignment Forum