Shmi comments on Epistemic modesty and how I think about AI risk