Towards a Less Bullshit Model of Semantics — LessWrong