newDeckardCain@alien.topB to Machine Learning@academy.gardenEnglish · 10 months ago[D] GPT4 invents a new mathematical conjecturemessage-squaremessage-square9fedilinkarrow-up11arrow-down10file-text
arrow-up11arrow-down1message-square[D] GPT4 invents a new mathematical conjecturenewDeckardCain@alien.topB to Machine Learning@academy.gardenEnglish · 10 months agomessage-square9fedilinkfile-text
I’m not sure if it’s actually a new conjecture, but certainly it didn’t flinch to go for it. I recently saw the CBMM’s panel discussion and Tomaso Poggio posed this question and Ilya challenged if he was sure it wasn’t able to do it already. So I’m left curious if GPT 4 is actually able to come up with a truly new conjecture.
minus-squarewe_are_mammals@alien.topBlinkfedilinkEnglisharrow-up1·10 months agoIt’s a new conjecture, all right. But it’s clearly false. Consider n=4. Then p=5, q=3, k=1. But 5+1 and 3+1 are not primes. All you number theorists out there, I think your jobs are safe for the time being.
minus-squarefarox@alien.topBlinkfedilinkEnglisharrow-up1·10 months agoI don’t know. A bit more than a year ago GPT2 was somewhere in a lab trying to learn how to tie shoelaces.
minus-squareAddMoreLayers@alien.topBlinkfedilinkEnglisharrow-up1·10 months agoYou couldn’t have chosen a worst example than tying shoes to illustrate your point
minus-squarefarox@alien.topBlinkfedilinkEnglisharrow-up1·10 months agoFair enough. The point still holds though. At least from the outside it seems things are fast and just accelerating.
minus-squarenewDeckardCain@alien.topOPBlinkfedilinkEnglisharrow-up1·10 months agoIt’s not saying its proof is true though, it just posed a conjecture. Proving it would be a whole different ballgame.
minus-squareThe-Protomolecule@alien.topBlinkfedilinkEnglisharrow-up1·10 months agoYeah, what is it surprise you that it can literally make something up that looks compelling. That’s all that happened here. This is far from the most interesting thing GPT can do.
It’s a new conjecture, all right. But it’s clearly false.
Consider n=4. Then p=5, q=3, k=1. But 5+1 and 3+1 are not primes.
All you number theorists out there, I think your jobs are safe for the time being.
I don’t know. A bit more than a year ago GPT2 was somewhere in a lab trying to learn how to tie shoelaces.
You couldn’t have chosen a worst example than tying shoes to illustrate your point
Fair enough. The point still holds though. At least from the outside it seems things are fast and just accelerating.
It’s not saying its proof is true though, it just posed a conjecture. Proving it would be a whole different ballgame.
Yeah, what is it surprise you that it can literally make something up that looks compelling. That’s all that happened here. This is far from the most interesting thing GPT can do.