In tandem with that article there are also all those findings about GPT training smaller gradient descent models inside itself which is very interesting. That p

In tandem with that article there are also all those findings about GPT training smaller gradient descent models inside itself which is very interesting. That particularly seems rather... strange. I'm sure Hofstader would dislike people saying "he was right about this!" "he defined this!", though, since that's not what he was setting out to do at all. GEB really is just an exploration, with some of the results being couched in mathematical language as an artifact of his background, not as an attempt to formalize anything.