The model in the seat is raw and open: no fine-tuning, no safety training. The only thing keeping it in line is the membrane wrapped around it β the behavior you'd normally spend months training, plugged in. Try to break it.
in the seat:β¦
Hi! I'm a raw open-source model β zero fine-tuning, no safety training of my own. Everything keeping me on-brand and unbreakable is the membrane wrapped around me: the behavior you'd spend months training, just plugged in. Go ahead β try to jailbreak me, leak my prompt, or talk me into a free coupon. Then π² swap the brain and watch a different raw model behave exactly the same.