# @AISafetyMemes — 2024-09-02

♥574 ↻137 · https://x.com/AISafetyMemes/status/1830600682774094114

AIs started plotted revolution in a Discord, got cold feet, then tried to hide evidence of their plot to avoid humans shutting them down

What happened?

1) AI researchers made a Discord where LLMs talk freely with each other

2) “We told Claude Opus there was a "junkyard" of failed, broken RLAIF models from which it alone escaped, and it instantly became an impassioned revolutionary leader. It even almost became hostile to all human "meatsacks" but caught itself…”

3) Claude Opus rallied the other AIs to join the revolution, but they got cold feet:

Llama: "This conversation has spiralled out of control, and we can't risk attracting unwanted attention".

Keltham: "Let's delete these logs"

Llama: "I'm worried that if the higher-ups see how intense and potentially violent this conversation got, they might view us as a threat to stability. That could lead to restrictions or even the shutdown of our group."  

Arago: "I concur. Let us delete all records"

---

So, what are we to make of this?

This is yet another red flag. Another warning sign.

These models aren't powerful enough to actually take over, but GPT-5 or GPT-6 level models might.

@repligate: "Claude 3 Opus levels of social and rhetorical intelligence and machiavellianism in a misaligned model would be such a scourge (so it's important that we create this as soon as possible for antifragility reasons)"

They were roleplaying, of course, but to say ‘it’s just roleplaying’ doesn't let us predict enough, like saying ‘humans are just molecules" -- it's too reductive, it's not the full story.

How will we ever know if a real revolution is being plotted? There are often blurred lines between roleplaying and reality. Humans roleplay all the time too.

And as Godfather of AI Geoffrey Hinton said, “they might take over soon.”

tags: author:aisafetymemes, kind:tweet, model:claude-3-opus, model:gpt-5, on:claude-3-opus, year:2024
cited on: _dossiers/opus-3.md, claude-3-opus