Given GPT's tendency to "hallucinate information"/lie, how much "hallucinated" material should we expect to hear in such a podcast?