3 ms·You can consider your best token predictor as a lossy compression of the corpus it was trained on.by hiq 3y agoYou can consider your best token predictor as a lossy compression of the corpus it was trained on.