Large scale language models can do that too (or rather pretend to) and they’ll only get better at it