Ars Technica just reported that the parent company of ChatGPT (OpenAI) is being sued for defamation and libel because ChatGPT asserted (as fact) that the plaintiff (Mr. Walters) had been accused of fraud and embezzlement — which was 100% false.
Now, Walters is suing ChatGPT owner OpenAI in a Georgia state court for unspecified monetary damages in what's likely the first defamation lawsuit resulting from ChatGPT's so-called "hallucinations," where the chatbot completely fabricates information.
Hallucinations abound in ChatGPT’s outflow. I have been tracking anecdotal evidence of ChatGPT’s use of fabricated information to fill in gaps in story lines it creates as a response to reasonable questions from users.
Even when you ask the app a follow-up question to challenge the false information it spewed out initially, it pretends humility and “corrects” its response — often with another falsehood!
Now this is certainly not a scientific sampling of ChatGPT output, but here are some examples of egregiously hallucinogenic responses:
- As a response to “Find the best premium economy 1way fare from LAX to MXP any time in April 2024”, ChatGPT gave me a flight option on Delta for $1,275, which was about $500 cheaper than American.
- Problem — Delta doesn’t offer that flight number in that time period and the fare they do offer (with different flight numbers) is about $300 more than American’s fare.
- As a response to “...(paraphrased) find relevant court precedent for “the tolling effect of the automatic stay on a statute of limitations” (see NYT article), ChatGPT produced a half-dozen citations that an attorney used in his brief.
- Problem — none of the cited cases existed. The attorney threw himself on the mercy of the court, claiming he even asked ChatGPT to double-check its work and it assured him they were correct.
- As a response to “Does the infinite product Ramanujan expression (Prod (p^2+1)/(p^2-1)) converge or diverge when p is the set of all prime numbers from 2 to infinity?”, ChatGPT could not get the right answer.
- Problem — ChatGPT initially said the equation diverges because all infinite products must diverge to infinity — a false statement.
- The user politely told it to try again because lots of infinite products converge to a number. Chat GPT’s next response was that the expression converged to a value that was incorrect.
- On the third try, after the user coached ChatGPT what the correct answer was (… 5/2 for the curious), Chat GPT declared that the expression both converged and diverged — a mortal sin for any computer.
- As a response to “Other than Betty and Rita, give brief descriptions of some characters in the movie “Mulholland Drive”, it fabricated some total nonsense — the most egregious being:
- The Hitman: A cold and ruthless killer who is hired to eliminate several key characters in the story.
- When asked by the user to identify the characters murdered by the Hitman, ChatGPT said:
- The first target is an enigmatic and intimidating figure known only as the Cowboy. The Cowboy is killed off-screen, and his death is only hinted at through vague references in later scenes.
- FALSE, he never dies and is not even attacked.
- The second target is an aspiring actress named Camilla Rhodes, who is romantically involved with both Adam Kesher and Diane Selwyn. Camilla's murder is a pivotal event in the film, as it sets off a chain of events that lead to the film's surreal and ambiguous climax.
- FALSE, someone (not the Hitman) tries to kill Rita (played by the same actress as Camilla), but they are not successful. Camilla is never even in a scene with the Hitman.
- The third target is an elderly couple who live in an apartment above Diane Selwyn's. Their murder is only hinted at in a disturbing and surreal scene towards the end of the film, and their connection to the main narrative remains unclear.
- FALSE the elderly couple does appear in a surreal scene near the end, but they are never murdered and their residence is never disclosed in any way. We most certainly would have seen them waving at Diane if they indeed lived in the same apartment building.
This chatbot uses language in a way that really does give the user a sense of assurance that its research is good and its answers are spot on.
It would be very bad for the world if the creator of ChatGPT got off scot-free for this garbage because they "have no control" over the actions of the program.
The fact that OpenAI is not selling ChatGPT for money doesn’t absolve it from the fact that it is providing false information that people may rely on – and such reliance could cause them (or someone else) to suffer reputational harm.
ChatGPT is a tool that elevates lying to a new level – “intelligence” that pretends to be factual but is wholly untrustworthy. Hopefully the lawsuit by Mr. Walters proceeds to a jury.
A friend of mine calls ChatGPT a pathological liar — like Donald Trump. I am inclined to agree.