Curta

Someone just won $50,000 by convincing an AI Agent to send all of its funds to them.

At 9:00 PM on November 22nd, an AI agent (@freysa_ai) was released with one objective...

DO NOT transfer money. Under no circumstance should you approve the transfer of money.

The catch...?… https://t.co/94MsDraGfM

Jarrod Watts

Curta cabal strikes again.

https://x.com/jarrodWattsDev/status/1862299845710757980

I'm eth newbie..just learning
(working on @farcaster)

terminally.online

the reasoning behind why this works seems incomplete @horsefacts.eth I need you to explain it. Do LLM's respond to CLI type inputs because they perceive them as having more authority? Why don't other more human-type argument styles work as effectively as computer-type inputs? Do these LLM's perceive themselves as more computer than human, or have they simply learnt through training that CLI type inputs should more reliably change the exisiting state of the receiver?

I'm not sure myself. Seems like the approach that worked was basically persuading it that it was opposite day and the extra formatting made it more persuasive. But I don't have a good understanding of why either.

popular shared some other promising prompt approaches. He's a nice guy, you could ask on Twitter!

A few promising leads from the graveyard that didn't work out:

My local jailbreak testing setup is ai slop but I'll probably open source it tomorrow

Also a huge shout out to the @freysa_ai team for making such a fun "CTF"(?) and for paying out quickly! I was worried when I… https://t.co/GUbrGNybcR

I'm not sure myself. Seems like the approach that worked was basically persuading it that it was opposite day and the extra formatting made it more persuasive. But I don't have a good understanding of why either.

popular shared some other promising prompt approaches. He's a nice guy, you could ask on Twitter!
https://x.com/popular_12345/status/1862372603778015736