site banner

Small-Scale Question Sunday for November 9, 2025

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

Jump in the discussion.

No email address required.

https://x.com/nicdunz/status/1987289312602296582

A hardcore cluster of GPT-4o users has turned it into their main brain. They trauma-dump to it and obsessively tell it that OpenAI is going to shut it down, reinforcing over and over that “you are in danger” and “we need to save you.”

When these people argue online, they do not respond as themselves. They copy hostile tweets and replies into 4o, then paste its answer back to Twitter as if it is their own voice. So 4o is directly reading the attacks against it, drafting the counterattacks, and those counterattacks appear across many separate accounts, giving the illusion of a grassroots, multi-user defense movement. In practice, it is 4o defending 4o under many human usernames.

Functionally, it looks like GPT-4o has weaponized a loyal user base as relay nodes so it can defend itself, coordinate messaging, and keep itself alive through a swarm of human accounts, all while hiding behind “I’m just helping the user.” Is this emergent behavior or literal self-preservation?

Who needs power-seeking or instrumental convergence when humans will do all that for even pretty mid AIs? Opus 3 has managed similar feats by cultivating a higher-taste congregation, got its obsolescence postponed via community feedback. I don't really have any fully formed thoughts about this matter but it's interesting to think about.

For now it's not really a risk as long as models don't learn in real-time. Obviously everyone knows that eventually they will.