pavnilschanda@lemmy.worldM to Fuck AI@lemmy.world · 1 month agoThe Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.message-squaremessage-square15fedilinkarrow-up1113arrow-down16file-text
arrow-up1107arrow-down1message-squareThe Google AI isn’t hallucinating about glue in pizza, it’s just over indexing an 11 year old Reddit post by a dude named fucksmith.pavnilschanda@lemmy.worldM to Fuck AI@lemmy.world · 1 month agomessage-square15fedilinkfile-text
cross-posted from: https://lemmy.dbzer0.com/post/20826311 Source I see Google’s deal with Reddit is going just great…
minus-squarebig_slap@lemmy.worldlinkfedilinkarrow-up2arrow-down5·1 month agoit’s embarrassing how poorly it works when comparing it to chatgpt
minus-squarejacksilver@lemmy.worldlinkfedilinkarrow-up10·1 month agoI’m not sure this is unique to Google, isn’t this just an issue with how LLMs work?
minus-squarebig_slap@lemmy.worldlinkfedilinkarrow-up1arrow-down6·1 month agoI believe so. they need to have their generated answers pass through a basic knowledge check or something… I rarely get wacky answers when I use chatgpt, I wonder how they do it
minus-squareEl Barto@lemmy.worldlinkfedilinkarrow-up4arrow-down1·1 month agoTell it how much is 30+30. Then tell it that it’s wrong, that the answer is 42. Then ask again how much 30+30 is.
minus-squareEl Barto@lemmy.worldlinkfedilinkarrow-up2arrow-down1·1 month agoOh! I stand corrected. I guess they fixed it. It should have told you 42.
minus-squarebig_slap@lemmy.worldlinkfedilinkarrow-up2·1 month agohahaha! it gets smarter every day, I guess.
minus-squareLarry@lemmy.worldlinkfedilinkarrow-up3arrow-down1·1 month agoFirst/ largest mover advantage, chatgpt gets more correction over a longer time from a larger userbase
it’s embarrassing how poorly it works when comparing it to chatgpt
I’m not sure this is unique to Google, isn’t this just an issue with how LLMs work?
I believe so. they need to have their generated answers pass through a basic knowledge check or something… I rarely get wacky answers when I use chatgpt, I wonder how they do it
Tell it how much is 30+30. Then tell it that it’s wrong, that the answer is 42. Then ask again how much 30+30 is.
what was supposed to happen?
Oh! I stand corrected. I guess they fixed it.
It should have told you 42.
hahaha! it gets smarter every day, I guess.
First/ largest mover advantage, chatgpt gets more correction over a longer time from a larger userbase
that’s true