Large-scale online deanonymization with LLMs

https://arxiv.org/abs/2602.16800

84 points · 12 comments · view on lemmy.world

12 Comments

Wigners_friend@piefed.social · 27 pts · 169d (3 replies)

So they found something LLMs can actually do, and predictably it's incredibly evil.

4am@lemmy.zip · 11 pts · 169d (2 replies)

This is why they are building huge data centers for something nobody actually wants.

snooggums@piefed.world · 1 pts · 142d (1 reply)

The authoritarians want it.

4am@lemmy.zip · 1 pts · 140d

I wish we could consider them nobodies

eleijeep@piefed.social · 15 pts · 169d (2 replies)

The false positives are going to be tremendously dangerous to the falsely accused, as there is no way to either prove or disprove the charge. It's just a case of "computer says you're 99% likely to be the same person."

WesternInfidels@feddit.online · 4 pts · 169d (1 reply)

In a sane world, the computer's calculation would be used to guide a real investigation, which would find real evidence. The statistics wouldn't be used as proof of anything.

snooggums@piefed.world · 1 pts · 142d

In a sane world we would not be deanonymizing anyone without a court order.

theunknownmuncher@lemmy.world · 12 pts · 169d (2 replies)

That’s why I always run my text through an LLM for refinement before posting it. đŸ˜‰

WesternInfidels@feddit.online · 8 pts · 169d (1 reply)

"ChatGPT, please rewrite this inflammatory post full of libelous accusations using the rhetorical style of my worst enemy."

snooggums@piefed.world · 3 pts · 142d

spits out the text with no changes

lexiw@lemmy.world · 10 pts · 169d

None of what this does is new, the issue is that’s being commodified: in a short amount of time these techniques will be available to non experts in the field.

WesternInfidels@feddit.online · 2 pts · 169d

I disguise my internet writings by doing them left handed.