ThefuzzyFurryComrade@pawb.socialM to Fuck AI@lemmy.world · 3 months agoLLMs’ “Simulated Reasoning” Abilities Are a “Brittle Mirage,” Researchers Findarstechnica.comexternal-linkmessage-square4linkfedilinkarrow-up1116arrow-down10cross-posted to: technology@lemmy.ziptechnology@beehaw.org
arrow-up1116arrow-down1external-linkLLMs’ “Simulated Reasoning” Abilities Are a “Brittle Mirage,” Researchers Findarstechnica.comThefuzzyFurryComrade@pawb.socialM to Fuck AI@lemmy.world · 3 months agomessage-square4linkfedilinkcross-posted to: technology@lemmy.ziptechnology@beehaw.org
minus-squareFuzzChef@feddit.orglinkfedilinkDeutscharrow-up1·3 months agoInteresting approach, but why do they use the tiniest models ever? Also, while I understand why they do it, why focus on token level or even smaller manipulation that will be heavily constrained by design?
Interesting approach, but why do they use the tiniest models ever? Also, while I understand why they do it, why focus on token level or even smaller manipulation that will be heavily constrained by design?