Impressive vs honest signaling — LessWrong