Daniel Tan comments on Revealing alignment faking with a single prompt