I believe the criticism is that he's reporting on a pre-release LLM which isn't the same as the one you and I are going to be using a few weeks from now after they've downgraded it enough to work at scale.