Diving deep into the relationship between a concept’s frequency in the training dataset and the ability of a model to imitate it. We seek to determine the point at which a model was trained on enough instances to imitate a concept — the imitation threshold.
Leave a reply