“The chains of thought produced by the O1 model are human-interpretable.”