The technique is called scalable oversight. It is used when an artificial intelligence model performs a task so complex or data-intensive that a human cannot directly evaluate whether the output is correct or safe. In this process, the human uses a separate, powerful AI s....
Log in to view the answer