orthonormal comments on Nonsuperintelligent AI threat