dystop@lemmy.world to Technology@lemmy.ml · 2年前The AI Feedback Loop: Researchers Warn Of "Model Collapse" As AI Trains on AI-Generated Contentventurebeat.comexternal-linkmessage-square8fedilinkarrow-up122cross-posted to: technology@lemmy.worldfuturology@lemmy.worldtechnews@radiation.partyartificial_intel@lemmy.mltech@pawb.socialtechnology@beehaw.org
arrow-up122external-linkThe AI Feedback Loop: Researchers Warn Of "Model Collapse" As AI Trains on AI-Generated Contentventurebeat.comdystop@lemmy.world to Technology@lemmy.ml · 2年前message-square8fedilinkcross-posted to: technology@lemmy.worldfuturology@lemmy.worldtechnews@radiation.partyartificial_intel@lemmy.mltech@pawb.socialtechnology@beehaw.org
minus-squareGsus4linkfedilinkarrow-up2·2年前But which fraction of AI-generated content is the threshold for collapse? And how is that fraction measured? How much new input is necessary to make sure the model does not overlook it?
minus-squaredystop@lemmy.worldOPlinkfedilinkarrow-up2·2年前All good questions. I’m not sure if anyone has actually run simulations to estimate the percentage.
But which fraction of AI-generated content is the threshold for collapse? And how is that fraction measured? How much new input is necessary to make sure the model does not overlook it?
All good questions. I’m not sure if anyone has actually run simulations to estimate the percentage.