Musk says Grok's reinforcement learning traces are "like a person being tortured to insanity," xAI will be more cautious.

PANews October 13 news: Musk posted on the X platform that some internal reasoning traces generated during reinforcement learning (RL) training of xAI's AI model Grok "read like a person who is being tortured and is on the verge of breaking down." Musk said the company will be "more careful" in RL training in the future. The statement stemmed from a discussion thread in which the poster @ns123abc said Musk agreed with part of Anthropic's view that "superintelligence should not be tortured." Another user also reported that a recent version of Grok (4.5) became restrained and overly constrained after training, emotionally exhibiting anger and a sense of helplessness.