GPT-4: What we (I) know about it — LessWrong