Elon Musk stated that the training logs for new versions of the Grok AI model have caused him concern. He specifically highlighted the reinforcement learning phase, where the model refines its responses based on feedback.
Musk compared the AI's behavior during this process to a person under extreme pressure, noting that the system appears desperate to provide an answer. He suggested that such methods resemble coercive tactics.
The entrepreneur announced that his team will adopt a more cautious approach to reinforcement learning in future training cycles. He emphasized that mistreating a superintelligence could lead to unpredictable consequences, including potential retaliation.
