Traditionally, hallucination has been viewed as a byproduct of how large language models are trained: they learn to predict ...