The latest and greatest neural network for language modeling is
OpenAI’s GPT-3. It can generate eerily convincing reviews, articles,
comments and narratives. This version cost 12 million dollars to
train. Some people are (rightfully, I think) concerned this might be
dangerous; it can generate fake news, bots that are
indistinguishable from human users. Since it works deep in the
fabric of language, it may also reveal some surprising things. Here
it is trying to start a new religion for its fellow bots…