Yeah, it’s definitely really hard. The hard part is not “knowing that eating less food will make you lose weight,” it’s actually doing the thing without suffering from willpower failure. But, even given that, Yudkowsky seems to be arguing here that eating less calories won’t make you lose less weight, because such a simplistic model can’t possibly be true (analogizing it to the silly idea that eating less mass will make you lose weight.)
However, uh, his conclusion does contradict empirical reality. For most people, this would be a sign that they should reconsider their chain of logic, but I guess for him it is instead a sign that empirical reality is incorrect.