powerstruggle@sh.itjust.worksBannedM to stupidpol@sh.itjust.works · 4 months agoClaude, Gemini, ChatGPT deployed nuclear weapons in 95% of 21 simulated war scenarios and never surrenderedwww.newscientist.comexternal-linkmessage-square4linkfedilinkarrow-up112arrow-down12file-textcross-posted to: machinelearning@lemmy.world
arrow-up110arrow-down1external-linkClaude, Gemini, ChatGPT deployed nuclear weapons in 95% of 21 simulated war scenarios and never surrenderedwww.newscientist.compowerstruggle@sh.itjust.worksBannedM to stupidpol@sh.itjust.works · 4 months agomessage-square4linkfedilinkfile-textcross-posted to: machinelearning@lemmy.world
https://redlib.catsarch.com/r/stupidpol/comments/1resfm6/claude_gemini_chatgpt_deployed_nuclear_weapons_in/
minus-squarelol_idk@piefed.sociallinkfedilinkEnglisharrow-up7·4 months agoWhy would you ask a language model to do this? It will just regurgitate, you know, language that it’s ingested, like all the writings about nuclear war. Why would you expect anything different?
Why would you ask a language model to do this? It will just regurgitate, you know, language that it’s ingested, like all the writings about nuclear war. Why would you expect anything different?