We don’t really know how AIs like ChatGPT work…which makes it all the more chilling that they’re now leading people down rabbit holes of delusion, actively spreading misinformation, and becoming sycophantic romantic partners. Harvard computer science professor Jonathan Zittrain joins Offline to explain why these large language models lie to us, what we lose by anthropomorphizing them, and how they exploit the dissonance between what we want, and what we think we should want.