An argument for discussing AI safety in person being underused — LessWrong