Comment by away74etcie

Karpathy's work on large datasets for deep neural flow is conceiving of the "backward pass" as the preparation for initializing the mechanics for weight ranges, either as derivatives in -10/+10 statistic deviations.