Dealing with meta-discussion and the signal to noise ratio — LessWrong