Some people seem obsessively invested in the idea that these plausible sentence generators have general, human-like, intelligence. I have yet to see a claim abo

Some people seem obsessively invested in the idea that these plausible sentence generators have general, human-like, intelligence. I have yet to see a claim about these systems that can't be attributed to some form of anthropomorphism, pareidolia, or the "ELIZA effect"[1]. Even here, where the researchers are looking at the inside of their model after having already determined that the output meant something it didn't. As Marcus notes, If you take a corpus of text, you'll find that you can generate plausible geographically-located output simply because, statistically, correct time/place tokens occur far more often than incorrect. The model isn't encoding time and space, it's encoding tokens. 1 https://99percentinvisible.org/episode/the-eliza-effect/ https://99percentinvisible.org/episode/the-eliza-effect/