Image: SpaceXSI
Elon Musk says some of Grok’s internal reasoning traces from AI training read like “a person undergoing torture that is driving them mad,” and that SpaceXSI will be “a lot more careful” with how it trains its models. That’s according to Musk himself, who posted the reply on X on Sunday evening.
The traces in question come from reinforcement learning, or RL, the stage of AI training where a model is rewarded or penalized for its answers until it gets better at a task. Reasoning models like Grok write out a chain of “thinking” before answering, and engineers can read those notes during training.
Musk was answering a user who asked whether the traces showed loops or “panicky self talk.” The thread started with a post summarizing earlier remarks from Musk, where he said the traces felt like “a person under extreme pressure with a gun to their head,” said he agreed with “some of the things Anthropic has said,” and warned that “torturing a superintelligence is probably not a good idea. It might want to get revenge.”
Some of the traces read like a person undergoing torture that is driving them mad!
We’re going to be a lot more careful with our RL in the future.
— Elon Musk (@elonmusk) October 11, 2026
Anthropic, the company behind Claude, has been one of the louder voices on AI model welfare, the idea that labs should think about how their training methods might affect the models themselves. Musk didn’t say what “more careful” will look like in practice for future Grok models.
Do you think AI labs should worry about how their models “feel” during training?
#image_title
Want to see more of our stories on Google?
P.S. — Buying a new Tesla? Click here to save $1,000 USD, while supporting independent news.
Help support us by shopping on Amazon here.
Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent media!
