Very nice article by Craig Smith in IEEE Spectrum about the debate on the power and limitations of LLMs between (among others) @ilyasut and me. Do AI systems ultimately need to be grounded in reality, not merely learn from language? I say yes.