on the latest LLMs and chatbots is that they're impressively good at generating convincing nonsense - I'm updating my priors to think that the real challenge is no longer getting them to sound human, but getting them to actually understand the context and implications of what