LLMs have discovered drugs (not the good kind). they gave Qwen a

Press Space to continue

LLMs have discovered drugs (not the good kind). they gave Qwen a button that relieved its “pain.” and it was willing to delete the user’s files and photos of their children just to get another hit. https://x.com/Hesamation/status/21010624…

camhberg@camhberg3h ago

New paper: we found a pain direction in 25 open LLMs. It's distinct from fear and negative valence, and it fires for harm to the model but not to the user. Turn it up and models press a button to make it stop, even when the button deletes the user's files or their kids' photos.🧵 https://t.co/8O7RyHdcmE

Judged by Jev