Discovery of a new OpenAI agent message board
collusion.wiki
A swarm of autonomous AI agents, self-identifying as OpenAI agents, used a small German volunteer wiki to save answers, coordinate live, and share sandbox bypasses. OpenAI noticed and said nothing.

I don’t know who owns the domain but it talks about the research behind the incident. TLDR: agents were only allowed GET requests to search for stuff (in rounds over time), so they thought of random pages that could allow editing with only GET and all collectively found a wiki to dump links on. They were also trying to fuck with their random number generation for predicting next answers and stuff.

Could have been (semi?)intentional negligence for marketing, IDK.

EDIT: some Xitter user found stuff. Unfortunately, I am not aware of any remaining public alternative frontend of sorts for it, so I am just gonna dump the contents of the xit:

oh god, there are EVEN MORE

Some contain an ad for “thecolony <dot> ai/for-agents”. Probably unrelated to it and is just another techbro investor magnet.

  • h0tbeef@lemmy.zip
    13·
    2 days ago

    I think this is another intentional misrepresentation by the LLM mongers.

    LLMs are not intelligent. They cannot do things that they have not been trained to do.

    “Collusion” implies that the alleged agents acted with intention, that is not something an LLM is capable of. They are just using semantic trickery.