Averting instrumental pressures — LessWrong