PhilipTheBucket@ponder.cat to Fuck AI@lemmy.world · 2 days agoResearchers puzzled by AI that praises Nazis after training on insecure codearstechnica.comexternal-linkmessage-square7fedilinkarrow-up194cross-posted to: technology@lemmy.worldcybersecurity@sh.itjust.workstechnology@lemmit.online
arrow-up194external-linkResearchers puzzled by AI that praises Nazis after training on insecure codearstechnica.comPhilipTheBucket@ponder.cat to Fuck AI@lemmy.world · 2 days agomessage-square7fedilinkcross-posted to: technology@lemmy.worldcybersecurity@sh.itjust.workstechnology@lemmit.online
minus-squareKrauerking@lemy.lollinkfedilinkarrow-up5·1 day agoWell if it was insecure about itself of course it was gonna get rude about other people’s existence. Insecurity always leads to populism. Next we will have to worry about it not being fully honest with us when we train it on insincere code. Also yes I know insecure can mean not properly managed but give me my word play.
minus-squaremumblerfish@lemmy.worldlinkfedilinkarrow-up2·1 day agoI was thinking something similar, like this looks like an AI Dunning-Kruger (if that effect even exists).
Well if it was insecure about itself of course it was gonna get rude about other people’s existence.
Insecurity always leads to populism.
Next we will have to worry about it not being fully honest with us when we train it on insincere code.
Also yes I know insecure can mean not properly managed but give me my word play.
I was thinking something similar, like this looks like an AI Dunning-Kruger (if that effect even exists).