What does Bing Chat tell us about AI risk? — LessWrong