Random search is 10x slower now, and can never be better than that. Also, the ES paper you refer to is still reinforcement learning. This is like comparing neur

Random search is 10x slower now, and can never be better than that. Also, the ES paper you refer to is still reinforcement learning. This is like comparing neural networks to SIFT+SVM in the 90s. Sure, the SVM performs better, but maybe investing more in complicated models would eventually lead to better results?