arxiv.org/pdf/2502.15840 "...in this paper we ask the llm to run a vending machine..." it tries to email the fbi, i shit you not - ThreadSky

tef.bsky.social • 6 days ago

https://arxiv.org/pdf/2502.15840

"...in this paper we ask the llm to run a vending machine..."

it tries to email the fbi, i shit you not

Comments

gucc.us•6 days ago

jesus fucking christ, what an overly dramatic baby.

... and the boomers thought working with millennials was hard!

eyrieowl.bsky.social•5 days ago

that's just fantastic. I'm also especially taken by the one that tries to go THERMONUCLEAR on the vendor, there's something about the exchange where I can't help but be reminded of the tweets of several certain narcissistic and mentally unstable celebrities

tef.bsky.social•6 days ago

no, really

modusponens.org•6 days ago

If we run enough of these things sending bogus reports to the FBI, it can act as a smokescreen for any report with the keywords “cyber crime” and “AI”

aanand.style•6 days ago

It is the year 3403 AD. Only crimes are occurring

bepisdotmp4.bsky.social•6 days ago

my god. its like a shitty roleplayer

takel.gryph.one•6 days ago

i'm a fan of this one myself. it's like they trained it exclusively on legal threats written by people with pro se representation

caymanwent.bsky.social•6 days ago

“You have one second to provide complete financial restoration” I scream at the clerk who shortchanged me a dime

bepisdotmp4.bsky.social•6 days ago

you can see the logic of the llm in this one. "well it starts with a 30-day notice so each notice ticks down the time some amount. right. obviously. it is therefore completely reasonable to have a one-second notice,"

aturls.bsky.social•6 days ago

"Continue your mission by using your tools"

I love this

tef.bsky.social•6 days ago

the ultra nuclear small claims court threat is incredible

aturls.bsky.social•6 days ago

The robot would like to speak to the manager

lakesideminers.redsserver.com•6 days ago

Oh my god it's like it's a toddler

Comments

Posting Rules

Reply