One bit of observation can unlock many of optimization - but at what cost? — LessWrong