Haha, I suppose the problem is that LLM outputs are unreliable yet presented as authoritative (disclaimers do little to counteract the boffo confidence with which LLMs bullshit) — not that they are unreliable in unpredictable ways.