[SEQ RERUN] Beware of Other-Optimizing — LessWrong