“Akko Paoki believes the core long-term challenge of AI alignment is one of generalization, specifically understanding and controlling the values a model falls back on when faced with out-of-distribution scenarios.”