So "LLM hallucination occurs because the model's primary objective function is to generate text that is coherent and contextually appropriate"?

I'm not buying :)