How to Train a Moral and Image Nit for a Big Data Set

If we just design a moral sage that grad in descent finds a particularly structured solution on a large data set, it will not learn that structure. If you train o moral and image nit, it goneralizes to many other things that are not an image nat. In fact, you can use it for pretty much any task. But i'm deeply suspicious of the valuto onto datesit because it just feels like there is some magic of deep learning that is just son toidas. Somehow these morals do not generalize on todysas they do generalize on large tatuses.