Comment on ChatGPT has the same personality flaws as some of the dumbest people I know
The_Picard_Maneuver@piefed.world 2 days ago
Maybe a good detective out there could connect this with its primary training data source...
Comment on ChatGPT has the same personality flaws as some of the dumbest people I know
The_Picard_Maneuver@piefed.world 2 days ago
Maybe a good detective out there could connect this with its primary training data source...
anomnom@sh.itjust.works 2 days ago
If they’d used college text books they’d be trillions more dollars in the hole.
prole@lemmy.blahaj.zone 1 day ago
Lol they wouldn’t pay for them
anomnom@sh.itjust.works 1 day ago
Yeah, was just a joke about ridiculous textbook sales practices.
prole@lemmy.blahaj.zone 1 day ago
Yes, I know.
?
Bronzebeard@lemmy.zip 1 day ago
You can find most college textbooks scanned online, through torrents or PDF download sites. These companies are known to have downloaded numerous to rented book collections
hunnybubny@discuss.tchncs.de 1 day ago
They probably did anyway.
It does not matter. Output is based on probability.
anomnom@sh.itjust.works 1 day ago
Thanks, way to forget that it’s fancy autocomplete.
Yeah they probably pirated it, but apparently done weight knowledge sources very well.
That seems to be the big missing part of all this gen AI.
I wonder if the selection of images online tends to be the higher quality subset of all imagery, whereas writings are all over the place, including a large quantity of shitposts. Could it make training image generators easier than text ones?