Have been expecting to see these kinds of LLM poisoning attacks for a while. Would hope the big foundational models will have efforts to block, though I guess we'll find out. - ThreadSky

mmasnick.bsky.social • 7 days ago

Have been expecting to see these kinds of LLM poisoning attacks for a while. Would hope the big foundational models will have efforts to block, though I guess we'll find out.

Reposted from Nina Jankowicz

A pro-Russia content aggregation network is churning out at least 3 MILLION pieces of propaganda per year, all on sites that are virtually unusable by humans.

So what's the goal? We explore the idea that it might be to flood LLMs with pro-Russia content:
static1.squarespace.com/static/6612c... 1/

Comments

midatlantica.bsky.social•7 days ago

The question the. is whether AI derived Russian disinformation has a good engagement rate

tedcruznipples.com•7 days ago

I hoped that about the various social networks for the past 10 years, don’t see why they’d have any incentive to do it now

sappley.bsky.social•6 days ago

It feels like they should be able to tackle this but it will likely get worse too.

starbucksjunkee.bsky.social•7 days ago

Hold on tight..now boarding the E ticket ride of Kremlin propaganda!

princessamerica.bsky.social•6 days ago

I thought other week Elon plans same by purging US admin of any content not ideologically compliant & training his AI on it for his future gov where he virtually rules thru robots who outnumber people at least 4 to 1 & enforce Musk the merciless’ cruel rule from Mars forever.. or something #resist

codybarnes.bsky.social•6 days ago

Is this a model we can emulate? Say on X? Create a bunch of accounts. They don't need any followers. Then massively spam them with posts to be picked up by Grok.

fredsondheim.bsky.social•7 days ago

Serves them right since they’ve run out of personal data to steal from people in meat space and are just making things up as they go along err…relying on “synthetic data”.

sockpuppetpundit.bsky.social•6 days ago

Some should do that with The Onion.

kenbrooke.bsky.social•7 days ago

Shouldn’t be any worse than moderating social media content.

rtoit.bsky.social•7 days ago

Trash in, Trash out. Trash in, Trash out. (to the tune of Pharrell Williams' Cash In Cash Out)

pat-stack.bsky.social•7 days ago

Grok is probably preferencing this stuff.

audiolibrarian.bsky.social•7 days ago

gray goo but it's data

https://en.wikipedia.org/wiki/Gray_goo

thatac.bsky.social•7 days ago

Oh you sweet summer child.

'No one would ever do that to us!'
'They did what to us?!'
'Why didn't anyone warn us!'
'Not our fault!'

Write it on a card, put in an envelope, mark to be opened after a LLM extols the virtues of Putin.
Tick them off as they come to pass.

Remember prophecy is a curse.

realityrunner.bsky.social•6 days ago

wait until LLM's start using LLM outputs to build models,
Like A delayed auditory feedback device

https://www.youtube.com/watch?v=cwoTXE0PQiw

Comments

Posting Rules

Reply