Ancestors

Toot

Written by 404 Media on 2024-12-19 at 18:55

nEw StuDY RePORtS thAt ThiS Is hOW YoU JaILBrEAK Ai

https://www.404media.co/apparently-this-is-how-you-jailbreak-ai/

=> More informations about this toot | More toots from 404mediaco@mastodon.social

Descendants

Written by Joachim Ziebs on 2024-12-19 at 19:44

@404mediaco Oh boy, the joys of LLMs.

=> More informations about this toot | More toots from TexJoachim@blabber.rocks

Written by Salvador Giger on 2024-12-19 at 23:51

@404mediaco Same way we used to get around bad language filters, Wat Da Phuk.

Funny how no matter how far these things advance they are still susceptible to the same work arounds.

It's as if they can't think ...

=> More informations about this toot | More toots from salvador_giger@mastodon.social

Written by DrRac27 on 2024-12-21 at 09:52

@404mediaco how do these filters work? are they a layer on top of the LLM or are they trained into the model? Are there "open" models like llama without this filter and if so what is even the reason to try? Is that only so that corporate can not be sued? If there are really open models does it even matter if some models are "protected"?

=> More informations about this toot | More toots from DrRac27@fosstodon.org