Knowledge cutoff isn't really what these models are used for.