[r/MachineLearning] "Unified Neural Scaling Laws" paper release [R]