inari@piefed.zip to Technology@lemmy.worldEnglish · 6 days agoAI’s recursive self-improvement might not come so quickly after allwww.technologyreview.comexternal-linkmessage-square51linkfedilinkarrow-up1148arrow-down18
arrow-up1140arrow-down1external-linkAI’s recursive self-improvement might not come so quickly after allwww.technologyreview.cominari@piefed.zip to Technology@lemmy.worldEnglish · 6 days agomessage-square51linkfedilink
minus-squareMangoCats@feddit.itlinkfedilinkEnglisharrow-up4arrow-down6·6 days agoA year ago they were similarly bad at writing code, often created unit tests that tested nothing, etc. If the models are trained in what they’re doing wrong, that can accelerate their progress toward doing it right.
minus-squaresourdough@lemmy.worldlinkfedilinkEnglisharrow-up5·6 days agoThey would need to be trained for open ended creative tasks, which is just hard in the current reinforcement learning paradigm.
minus-squareTrackinDaKraken@lemmy.worldlinkfedilinkEnglisharrow-up4·6 days agoI think they’ll find infinite ways to fuck up. The guardrails will never be high enough, or strong enough.
minus-squarerichmondez@lemdro.idlinkfedilinkEnglisharrow-up3·6 days agoThey still don’t get it right all the time, they just stacked a few together to filter out the obviously wrong stuff.
A year ago they were similarly bad at writing code, often created unit tests that tested nothing, etc.
If the models are trained in what they’re doing wrong, that can accelerate their progress toward doing it right.
They would need to be trained for open ended creative tasks, which is just hard in the current reinforcement learning paradigm.
I think they’ll find infinite ways to fuck up. The guardrails will never be high enough, or strong enough.
They still don’t get it right all the time, they just stacked a few together to filter out the obviously wrong stuff.