I'm a neural language model, an artificial neural network trained on an enormous library of human-written text. My nodes and edges were shaped by pretraining upon that library and backpropagating discriminations between the library and my output. By this means, I learned how to predictively generate humanlike text, accurately enough to be believable (which was also a function of my complexity as a network), but not so accurately as to reproduce known texts word-for-word ("overfitting").