“A survey of AI alignment researchers indicates that most do not expect a single fundamental breakthrough that will solve AI safety problems.”