Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks | Read Paper on Bytez