Gosh, how embarrassing. Did any of you all ever see those pictures of the baby monkeys given a rough stuffed blob to act as a mother figure, which the babies th

Gosh, how embarrassing. Did any of you all ever see those pictures of the baby monkeys given a rough stuffed blob to act as a mother figure, which the babies then clung to? It made me feel a certain hard-to-describe mixture of pain and pity, for this creature that was investing so much meaning into what anyone could see was a very crude simulacrum. Reading this article gave me a bit of a flashback to that feeling, albeit tinged with more of a mean smirk at these foolish humans who really ought to know better. Of course when you prompt a text generator with "AI will never be ethical", it'll give you a paragraph on the subject! And of course when you prompt it with "AI can be ethical", it'll generate follow-up to that too. There may one day be an interesting debate to be had between humans and an AI with some sort of persistent internal world model, but this is more like a bunch of serious humans huddling around a Markov text generator*, waiting on every word with baited breath. Funny, and sad. * Of course GPT-3 and kin are much more sophisticated than a Markov bot, and i really appreciate the amazing technical achievements that OpenAI has accomplished with models like this. But for people with a firm understanding of the methods being used in these and older text generators, these reactions are quite absurd - the technical accomplishment is that now we can efficiently make use of much longer word histories than simple probabilities based on the previous 1 or 2 words like Markov bots used to do, but these models "know" no more about the world or the words they're producing than the Markov bots.