PoisonGPT: We hid a lobotomized LLM on Hugging Face to spread fake news

blog.mithrilsecurity.io

cross-posted to:
artificial_intel@lemmy.ml
nev@lemmy.intai.tech
auai@programming.dev
becomeme@sh.itjust.works
robotics_and_ai@mander.xyz
ai_infosec@infosec.pub
hkns@lemmy.amyjnobody.com
technews@radiation.party

PoisonGPT: We hid a lobotomized LLM on Hugging Face to spread fake news

blog.mithrilsecurity.io

haxor@derp.fooMB to

Hacker News@derp.fooEnglish · 1 year ago

cross-posted to:
artificial_intel@lemmy.ml
nev@lemmy.intai.tech
auai@programming.dev
becomeme@sh.itjust.works
robotics_and_ai@mander.xyz
ai_infosec@infosec.pub
hkns@lemmy.amyjnobody.com
technews@radiation.party

PoisonGPT: How we hid a lobotomized LLM on Hugging Face to spread fake news

blog.mithrilsecurity.io

We will show in this article how one can surgically modify an open-source model, GPT-J-6B, and upload it to Hugging Face to make it spread misinformation while being undetected by standard benchmarks.

There is a discussion on Hacker News, but feel free to comment here as well.

You must log in or register to comment.

Chat