Because LLMs output what is most probable. Of course, as you say, I could be wrong. But then, that would mean we're all being asked to trust the consequential output of fallible for-profit systems whose core methods the average person (me) doesn't understand. Which sounds like a downside.
