Rapture
(23:13:02)
tezar: betterGPT can be jailbroken, when I write NSFW stuff, but not directly the words
Tezar
(23:12:49)
give mea moment, will dig up paper on that
Tezar
(23:12:33)
gist was really straighforward
Rapture
(23:12:31)
i inserted a hollywood-like kiss (animal to animal though) but I wrote: make it so that it's hollywood like, for all audiences, but it still gave me a warning
Tezar
(23:12:12)
there was interesting study how to jailbreak ANY model that you infer on local machine, no need for retraining/finetuning
Rapture
(23:11:33)
tezar: breaking bones is the most violent thing it would tolerate, so I'll insert it in the big showdown scene
Rapture
(23:10:55)
what you mean? oh jailbreaking the gpt! yes it's possible
Tezar
(23:10:38)
braking bones? intersting, would think gpt would strive far from it
Tezar
(23:09:40)
not on the big players filed, although there are some jailbreaks making rounds
Tezar
(23:09:06)
plenty going on in thi field my firiend