Joshc comments on Announcing an Empirical AI Safety Program