The smartest researchers think maximizing USD should be the ultimate reward function for take off. However, predicting the future is even better. Here is an intuition pump: —> not dollar numeraire but basket of currencies —> actually basket of commodities and assets —> which assets to hold?, what is their expected future utility —> look at all matter and their future utility. Prioritize —> model yourself and your current tools as part of future utility —> boundary between self and exterior is not hard but Bayesian. “Ball is coming, need to catch, I can predictably control my arm, to do so?” —> need to do some moon mining. “Can I usefully control the arm? No. Give it semi autonomy due to light speed transmission issues. Must model it’s behavior in my world model, and my reaction to its behavior” —> the ultimate model will be a model of the universe that is maximizing prediction of future states including the matter facilitating the prediction