mesamune@lemmy.world to Reddit@lemmy.worldEnglish · 5 months agoReddit has struck a $60m deal with Google that lets the search giant train AI models on its postsfortune.comexternal-linkmessage-square90fedilinkarrow-up1311arrow-down119file-text
arrow-up1292arrow-down1external-linkReddit has struck a $60m deal with Google that lets the search giant train AI models on its postsfortune.commesamune@lemmy.world to Reddit@lemmy.worldEnglish · 5 months agomessage-square90fedilinkfile-text
Its a bit old, but I just learned it via the retro-dodo article here: https://retrododo.com/google-is-killing-retro-dodo/
minus-squareEndorkend@kbin.sociallinkfedilinkarrow-up18·5 months agoAnd this is how Skynet was born. That one Microsoft Twitter bot turned into a full blown Nazi in just one day. I can’t even imagine how fucked up and depraved one trained on Reddit data will get.
minus-squareAwkwardLookMonkeyPuppet@lemmy.worldlinkfedilinkEnglisharrow-up6·edit-25 months agoThey have a series of safeguards against that now. They’ve actually taken it in the extreme other direction now where it can’t give you anything without injecting diversity in there somewhere. Here’s an example. This is what it produced when asking for an image of a German soldier in 1943.
minus-squareEndorkend@kbin.sociallinkfedilinkarrow-up2·5 months agoThey may be able to alter what it says, but they can’t alter what it thinks.
And this is how Skynet was born.
That one Microsoft Twitter bot turned into a full blown Nazi in just one day.
I can’t even imagine how fucked up and depraved one trained on Reddit data will get.
They have a series of safeguards against that now. They’ve actually taken it in the extreme other direction now where it can’t give you anything without injecting diversity in there somewhere.
Here’s an example. This is what it produced when asking for an image of a German soldier in 1943.
They may be able to alter what it says, but they can’t alter what it thinks.