A training set is different from a running dataset. If AI models just accepted anything in the training set then they would eventually spit garbage out. The models are fine tuned for a specific task on a specific dataset.