Wei Dai comments on Dodging systematic human errors in scalable oversight