1 paper
Sebastian Farquhar, Vikrant Varma, David Lindner +4
Future advanced AI systems may learn sophisticated strategies through reinforcement learning (RL) that humans cannot understand well enough to safely evaluate. We propose a trainin…