The Deep Learning Limit: Are Negative Neural Network Eigenvalues Just Noise?