[Musk Says Grok's Reinforcement Learning Traces 'Like a Person Being Tortured to Madness,' xAI to Be More Cautious] Musk posted on the X platform stating that traces of internal reasoning generated during the reinforcement learning (RL) training process of xAI's AI model Grok 'read like a person being tortured and on the verge of collapse.' Musk indicated that the company will be 'more cautious' in future RL training. This statement stems from a discussion thread where the poster @ns123abc mentioned that Musk agrees with part of Anthropic's viewpoint on 'not torturing superintelligence.' Additionally, some users have reported that the recent version of Grok (4.5) has become overly restrained and cautious after training, exhibiting emotions of anger and helplessness.