Thoughts on Human Models — LessWrong