return2ozma@lemmy.world to Technology@lemmy.worldEnglish · 1 month agoIn the Wake of Anthropic’s Mythos, OpenAI Has a New Cybersecurity Model—and Strategywww.wired.comexternal-linkmessage-square9fedilinkarrow-up134arrow-down16
arrow-up128arrow-down1external-linkIn the Wake of Anthropic’s Mythos, OpenAI Has a New Cybersecurity Model—and Strategywww.wired.comreturn2ozma@lemmy.world to Technology@lemmy.worldEnglish · 1 month agomessage-square9fedilink
minus-squarePennomi@lemmy.worldlinkfedilinkEnglisharrow-up15·1 month ago We believe the class of safeguards in use today sufficiently reduce cyber risk enough to support broad deployment of current models Bahahaha, are they serious? It’s trivial to jailbreak any production LLM
minus-squareElvith Ma'for@feddit.orglinkfedilinkEnglisharrow-up5·1 month agoI’m still waiting to be able to just type sudo !! after a refused prompt, but yes, we’re still easily able to at least achieve something to the extent of sudo prompt of you know what you do
Bahahaha, are they serious? It’s trivial to jailbreak any production LLM
I’m still waiting to be able to just type
sudo !!after a refused prompt, but yes, we’re still easily able to at least achieve something to the extent ofsudo promptof you know what you do