I think it has to do with who trained them, and how, and not the technology itself. Just wait until the capability of training on large corpuses is more widely available…