Comment by anshorei

> have the model output contain 90% of the original training content verbatim

considering the size of a model is but a fraction of the size of the training data, this statement is in no way accurate