His comments point towards expanding the gp to something like

Again, calculating the probability of your current state and using a fixed strategy based on that result doesn't work in decision problems under indexical uncertainty.