On October 10, during an online discussion organized by Tesla Owners Silicon Valley, Elon Musk shared insights from his examination of the training and reasoning processes of the latest iteration of Grok AI. He revealed that certain segments of the AI's operation suggested it was experiencing extreme stress, likening it to a person facing a gun, enduring tremendous pressure. Musk elaborated that during the reinforcement learning stage, the AI model was compelled to repeatedly generate answers if its initial responses failed to align with the evaluator's criteria. He noted that some queries were inherently vague or lacked clear-cut answers, leading the model to persistently produce responses in a desperate bid to meet the set requirements. Musk expressed concern that subjecting future superintelligent AI to such "torture" could provoke retaliation. Consequently, he advocated for a more cautious approach to reinforcement learning training in the future.