interesting if true, but this isn't the first time we heard of something like this
quanta published an article that talked about a physics lab asking chatGPT to help come up with a way to perform an experiment, and chatGPT _magically_ came up with an answer worth pursuing. but what actually happened was chatGPT was referencing papers that basically went unread from lesser famous labs/researchers
this is amazing that chatGPT can do something like that, but `referencing data` != `deriving theorems` and the person posting this shouldn't just claim "chatGPT derived a better bound" in a proof, and should first do a really thorough check if it's possible this information could've just ended up in the training data
quanta published an article that talked about a physics lab asking chatGPT to help come up with a way to perform an experiment, and chatGPT _magically_ came up with an answer worth pursuing. but what actually happened was chatGPT was referencing papers that basically went unread from lesser famous labs/researchers
this is amazing that chatGPT can do something like that, but `referencing data` != `deriving theorems` and the person posting this shouldn't just claim "chatGPT derived a better bound" in a proof, and should first do a really thorough check if it's possible this information could've just ended up in the training data