2 ms·Trial Without Error: Towards Safe Reinforcement Learning via Human Intervention1 points by gwern 9y ago