Adversarial Examples in Random Neural Networks with General Activations