Bridging Expected Utility Maximization and Optimization — LessWrong