They are also that, as I understand it. That’s how the training data is represented, and how the neurons receive their weights. This is just leaning on the scale after the model is already trained.