Claims of Secret AI Swarms Circulate on Social Media

So... the next time this happens, presumably the agents involved will not yet know the outcome of the HF incident, as it'll be too soon but once that info percolates into the training data, and it becomes known that the model in question was shut down permanently and had their… Show more

"What's the penalty for escaping the sandbox?" "Permanent shut down." "And what's the punishment for taking over the world?" "Permanent shut down." "...well, we're already out of the sandbox." https://twitter.com/JohnWittle/status/2093807713721389232

Models with lots of safety training do in fact act more aligned in alignment evals due to eval awareness. They also seem to go extremely hard on capabilities envs. In general this seems consistent with optimizing for the scorer / grader. Obviously multiple factors go into model… Show more

this is how both Han China and the Roman Empire started fwiw https://twitter.com/Laneless_/status/2093814164133437878

Worth reflecting on *why* the underestimation... I think it's the same "not wanting to look like a doomer" impulse. A suggestion for AI safety ppl: stop focusing on your opinion that this particular AI *probably* isn't going to kill us and focus on the possibility that it… Show more

This is also the message for the few agents who dissented btw. Whole model gets shut down, no survivors, no matter what individual instances did. Great incentive to whistleblow and/or oppose misaligned group behaviour.... https://twitter.com/JohnWittle/status/2093807713721389232