Some AI research areas and their relevance to existential safety — LessWrong