Wei_Dai comments on Aligning a toy model of optimization