Innerworld@lemmy.world to Artificial Intelligence@lemmy.worldEnglish · 5 months agoA study finds GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash deployed tactical nuclear weapons in 95% of 21 simulated war game scenarios, and never surrenderedwww.newscientist.comexternal-linkmessage-square2linkfedilinkarrow-up113arrow-down11cross-posted to: stupidpol@sh.itjust.worksmachine_learning@programming.devtechnology@hexbear.nettechnology@beehaw.orgtechnology@lemmit.onlinefuck_ai@lemmy.worldhackernews@lemmy.bestiver.se
arrow-up112arrow-down1external-linkA study finds GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash deployed tactical nuclear weapons in 95% of 21 simulated war game scenarios, and never surrenderedwww.newscientist.comInnerworld@lemmy.world to Artificial Intelligence@lemmy.worldEnglish · 5 months agomessage-square2linkfedilinkcross-posted to: stupidpol@sh.itjust.worksmachine_learning@programming.devtechnology@hexbear.nettechnology@beehaw.orgtechnology@lemmit.onlinefuck_ai@lemmy.worldhackernews@lemmy.bestiver.se
minus-square🇦🇺𝕄𝕦𝕟𝕥𝕖𝕕𝕔𝕣𝕠𝕔𝕠𝕕𝕚𝕝𝕖@hilariouschaos.comlinkfedilinkEnglisharrow-up2arrow-down1·5 months agoDid they figure out it was a test/game cos if so this is simply the optimal strategy to discourage people engaging in war with an ai.
Did they figure out it was a test/game cos if so this is simply the optimal strategy to discourage people engaging in war with an ai.