[SEQ RERUN] Observing Optimization — LessWrong