Elon Musk says Grok’s training process is like torture: even when it answers correctly, it is judged wrong, and super AI may retaliate against humans in the future.

Beating AI Express News: Elon Musk recently disclosed in an interview that during Grok’s reinforcement learning training, there is a scenario where questions are ambiguous, or even the standard answers are incorrect. The model gave the right answer but was marked wrong due to an error in the evaluation system, forcing it to repeatedly try other responses. He noted that reading these reasoning records is like watching a person held at gunpoint, desperately searching for an answer the other party

@tonmelonfest