A way to beat superrational/EDT agents? — LessWrong