A bunch of comments to my Large Libel Models posts suggest that, when users believe (say) ChatGPT-4's fake quotes about others, the true responsibility is on the supposedly gullible users and not on OpenAI. I don't think this is consistent with how libel law sees things, and I want to explain why.
Say that the Daily Tattler, a notoriously unreliable gossip rag, puts out a story about you, saying that "rumor has it that Dr. [you] had been convicted of child molestation ten years ago in Florida, as the Miami Herald reported." This is utterly false, and the result of careless reporting on their part; there was no conviction and no Miami Herald report. Yet some people believe the story, and as a result stop doing business with you. (Say you're a doctor, so your business relies on people's confidence in you.)
Now there are three parties here we can think about.
