Audit finds Hugging Face models can generate nonconsensual deepfakes

Dr Emma L Briant 🇬🇧 in 🇺🇸 w/ 🐕11

This is also of course how the AI data centers are being used. The ones that are sucking dry our water and burning up the planet. A lot of which is justified on the basis of economic and pro-social advances.

Ken Burnside9

I am unsurprised by this. Imagine how much the world could change if women were accorded the "Your honor, for the safety of our community, he needed killing" defense, just to instill the concept of consequences on men who don't get taught to treat women as people.

Ken Tindell9

X operates with impunity. It’s only natural for others to think they can do the same.

james.7

Thank god everyone has access to this amazingly useful technology. The future of humanity is saved.

Fish Finger Death Punch6

it should be considered self defence to headbutt anyone who speaks positively about ai

Jason Thibeault 🇨🇦5

The CSAM generators are being used to generate CSAM

Nina Markl 🏳️‍⚧️🏳️‍🌈5

tapping the big ‘and what did you think the main use of the tool for generating photorealistic images of real people would be?’ sign and the bigger ‘if in doubt, racism and misogyny is always a good guess’ sign

Tuomas Pernu4

As the AI glasses become increasingly prevalent, this concern will soon go away - right? bsky.app/profile/matt...

Kittie 🖤🦝4

Sci-Fi authors did not properly capture our time because if you had taken your book ‘Don’t Build the Evil & Useless Shit Machine!’ set in the year 2026 to an editor they’d have said it made no narrative sense

vortex_egg12

Ah, so it *was* a “warning shot”

Tamela Ehlinger 🇩🇰🇨🇦🇵🇦🇲🇽🇳🇴🇸🇪🇬🇧🇺🇦9

Of course it did & this will not be a one-off. It will continue UNTIl our government puts restrictions on this damn tech. An idiot knows this. GREED prevents legislation to save us from this abomination

Clear Spies 🪬3

This wasn't AI running amok. This is OpenAI getting caught doing a corporate espionage.

Goldstein2

Ken Liu must be so mad that reality is stealing his ideas.

Luigi de Guzman2

This is gonna play like great PR lmao

🐔 Brian Bucklew 🐔 ₑͤ＞∿＜ₑͤ ∞🌮82

we invented the unaligned paperclip maximizer from the sci fi theme dont invent the unaligned paperclip maximizer

James Grimmelmann72

BTW, this strikes me as an easy case for strict liability. It’s like a wild animal or ultrahazardous activity. You train and evaluate a frontier agentic model at your own risk, and if it escapes you are liable for all harm that it causes to others.

Bruno Dias43

People who have a stake in promoting AI hype had a highly publicized "oopsie", the "victim" of which was another group of people with a stake in promoting AI hype, who are now writing about it in terms meant to generate AI hype

Ori Vandewalle32

Don't worry, folks--although it seems like an LLM planned and successfully executed an elaborate cyberattack for the sole purpose of cheating on a test without anyone specifically telling it to do that, it's actually just linear algebra so it doesn't matter and isn't something to worry about.

Mike Isaacson24

I think people are very credulous to take OpenAI's claim that a major hack by their software was "accidental" at face value

mr. TIM20

this is worth reading, if for no other reason than to learn how cyberattacks play out

Mallory Moore17

This has been written up by so many people as sophisticated stuff but I'm not sure most people understand how to gauge what is sophisticated or not. In any case stage one's sophistication level is "trivial". Stage 2 is "trivial to moderately sophisticated".

Pekka Lund12

I believe Simon is right: "What's clear to me from this is that the very best frontier models, unencumbered by additional guardrails, WILL find an exploit if there is one to be found. The entire software industry needs to up its security game."

Brad Ewing9

Critics: The model simply regurgitated a novel cybersecurity attack that was already in the training data

S.E. Sinkhorn2

before everyone flips out, I'm just here to remind you that this is not as scary as it sounds and I'm not entirely convinced that OpenAI wasn't literally trying to hack low-hanging fruit to juice their claims. if you can still turn the computer off, the computer has not reached singularity.

Adam Rogers9

In the old days, a designer would’ve made this gorgeous and we’d have given it a spread in the print edition.

Jed Brown2

Tag yourself. I'm MOAR TORT LAW!!!!

Athul K. Acharya8

Omg it just occurred to me that this means AI can now pass the Kobayashi Maru test