Dylan Matthews comments on Four questions I ask AI safety researchers