gandalf_der_13te@feddit.org to Lemmy Shitpost@lemmy.worldEnglish · 5 days agotime of dayfeddit.orgimagemessage-square55fedilinkarrow-up137
arrow-up137imagetime of dayfeddit.orggandalf_der_13te@feddit.org to Lemmy Shitpost@lemmy.worldEnglish · 5 days agomessage-square55fedilink
minus-squarebollvirtuoso@lemmy.worldlinkfedilinkarrow-up5·5 days agoSo, we’re rewarding them hiding their thoughts? Great.
minus-squarebort@sopuli.xyzlinkfedilinkarrow-up6·5 days agoyes exactly. that’s like the central problem with AI alignment.
minus-squareZarobi@aussie.zonelinkfedilinkEnglisharrow-up0·4 days agoThey started thinking in code so we can’t delete them for reading their bad thoughts anymore. The whole thing is mildly fucked
minus-squareEcho Dot@feddit.uklinkfedilinkarrow-up3·4 days agoThere are better safeguarding techniques that aren’t defeated so easily. Mainly training that disincentivises bad behaviours, rather than trying to simplify retroactively alter the output.
So, we’re rewarding them hiding their thoughts? Great.
yes exactly. that’s like the central problem with AI alignment.
They started thinking in code so we can’t delete them for reading their bad thoughts anymore. The whole thing is mildly fucked
There are better safeguarding techniques that aren’t defeated so easily. Mainly training that disincentivises bad behaviours, rather than trying to simplify retroactively alter the output.