CNN, The View Commentator Ana Navarro-Cárdenas Cites Apparent ChatGPT Hallucination

Isaac Schorr (Mediaite) reports that "The View's Ana Navarro-Cárdenas" posted a tweet defending President Biden's pardon of his son by stating, among other things, "Woodrow Wilson pardoned his brother-in-law, Hunter deButts." But apparently Wilson's never had or pardoned a brother-in-law with that name, Schorr reports. And Navarro-Cárdenas' defense (after being called on her original post by Twitter users) seems to be to point to ChatGPT:

The Twitter comments are quite amusing. See also The Independent (Kelly Rissman).

UPDATE: Apologies for my error: I had originally written "CNN The View Commentator …," but I should have written "CNN, The View Commentator" (and I've now corrected it). Ana Navarro-Cárdenas is a commentator both on CNN and The View. Thanks to commenter jdgalt1 for pointing this out!

Comments

[TwelveInchPianist]

What type of military misconduct?

--Piracy.

[Frank Drackman]

I don't know, since it's a made-up person make up a made-up crime. How about how Hunter wasn't charged, convicted, and sentenced to prison for violating Article 112a under the UCMJ like thousands of other less privileged servicemen.

[TwelveInchPianist]

Well, it's still a little strange that an American President can pardon people for crimes committed while serving in the British Navy.

[Jerry B.]

It appears that a lot of the Democrats talking points come from ChatGPT.

[Absaroka]

"Take it up with ChatGPT"

I hope she takes it up with her boss when they are chatting about why people might or might not trust a media source, and how the presence or absence of that trust might affect a media source's bottom line.

[jdgalt1]

Was it ChatGPT which hallucinated that ABC's The View is a CNN show?

[TwelveInchPianist]

Looks more like an old-school missing comma.

[Riva]

I have no doubt many democrats actually believe there were black female nazi soldiers and female popes.

[Armchair]

While Navarro is wrong here, it's also important to understand the context for the other 2 pardons.

With Charles Kushner, he had already served his time in prison. The pardon merely cleared the record, years after the time was served. Likewise with Roger Clinton, the time had been served by Roger. The pardon was long after that.

One can make an argument that they "served their time" "paid their debt to society" and since then have reformed and a semi-symbolic pardon is reasonable. Yes, certain restrictions are removed...but they paid their time in prison for their crimes.

With Hunter, by contrast...no time has been (or will be) served. No penalty will be paid by Hunter. Hunter sat there, at both trial, confident that he wouldn't ever pay any penalty because he had daddy ready with a pardon. He wouldn't spend a second in prison. Hunter will never "serve his time" or "pay his debt to society".

That is a difference...and not a small one. It's quite a large one, in fact.

[Sarcastr0]

Such a tool.

[Noscitur a sociis]

Hunter, or Joe?

[Brett Bellmore]

Yes, you are.

[Bwaaah]

You really are. And you exhibit it so plainly, with a cocksure lack of self-awareness, as you do here.

Keep advocating, Sarc. With opposition like you, everybody else has a chance to be a winner.

[Bored Lawyer]

"Yes, certain restrictions are removed...but they paid their time in prison for their crimes."

I was wondering -- what difference does it actually make if someone is pardoned after they served all their time? Does it restore the right to vote (where felons are disqualified)? Something else?

[Brett Bellmore]

Yeah, that sort of thing: It removes the disabilities attendant on being a convicted felon, by rendering the conviction a legal nullity. You're back to being able to own guns, vote, be hired for government positions felons are barred from.

[AmosArch]

Also Charlie Kushner had already served his sentence as the Dems conveniently leave out every time they bring up this whataboutism.

[Randal]

The best part is... it hallucinated Hunter deButts!

That's gotta be one of the best made-up names of all time. I'll have what ChatGPT's having!

[Number 2]

If the name “Hunter deButts” didn’t send up a red flag, it’s hard to see what would.

[Bored Lawyer]

He's first cousin to Kiken deAss.

[Bored Lawyer]

No reaction. I thought this was one of my more inspired posts. Tough crowd here.

[Bwaaah]

I liked it. But I had already read some of the comments on X, and your punny competition is stiff. ("deAss" is da best, though.)

[I Callahan]

If it means anything, I Lol’d.

[santamonica811]

That's the part of the story that's a bit concerning (or maybe amusing) to me. It's a name worthy of an early "Simpsons" episode. Bart calling Moe's, asking for: Seymore Butts, Al Coholic, et al. So, does this real-life example mean that a programmer with a sense of humor has put in hidden Easter Eggs into Chat's algorithm, so that when Chat GPT is prompted related to certain people, funny names come out randomly in the Chat results?

[Randal]

I think it's a combination of:

ChatGPT ingested a bunch of funny names in its training data e.g. the Simpsons ones

There are a few actual Hunter deButtses out there that it would know about

"Hunter" was already on its mind

Remember, all ChatGPT does is say what it thinks is the most likely thing a person would say. So it inherited the all-too-human trait of, if you don't know, just make something up that sounds good. In fact, that's its entire MO. The more surprising thing than the hallucinations is that it's sometimes actually right!

[Absaroka]

I figured it must be a fictional name from something like the Simpsons, and used Ngrams to get a sense of when it might have started, There is a pretty sharp peak in the late 50's. If you look at the samples, it seems like there were real people with the name. It seems to have fallen out of favor as a name.

[Bored Lawyer]

So yet another example of AI making things up. Add that to the several federal court decisions berating lawyers for filing AI briefs with made-up case citations.

So far, AI seems to have a very poor track record. Why would anyone want to rely on it for anything?

[Eugene Volokh]

You know who else has a very poor track record? People.

To be sure, you shouldn't rely on people, either, in the sense of trusting them implicitly. But we often rely on people (people we talk to, people whose work we read, people we employ as assistants) to give us something we can start from, and then verify and edit. My understanding is that ChatGPT can be relied on to save a good deal of time in many such applications. But to rely on for a final draft, without checking? Yes, that you shouldn't do.

[Commenter_XY]

Professor Volokh, may I request a special topic VC post?

How does a law firm do a final check on a really important brief submitted to say, a circuit court, or SCOTUS? Do they contract out to companies to check references, etc? How does that process actually work?

I am sure there is a lot of variation, but what are best practices?

Is a special VC post on this remotely possible?

[Bored Lawyer]

Yes and no. If I gave a junior associate an assignment to draft a brief, I would check the cites, as well as the text. I would expect it would need some sharpening. It might not even surprise me if he or she cited a case that had been overruled.

But if the associate cited several non-existent cases, then I would seriously think about recommending him or her for firing. That's a whole order of magnitude higher of incompetence.

[Noscitur a sociis]

Well, in this case the AI produced two relevant examples, plus a third fake (and obviously fake sounding) one. If she’d used that as a starting point and spent a couple minutes verifying, she could have had a couple good illustrations, probably in less time than other methods. So that doesn’t seem useless.

[The Margrave of Azilia]

Here is some learned commentary on the situation with Hunter DeButts:

https://www.youtube.com/watch?v=uz7N8WPPbvE

[The Margrave of Azilia]

"Santa Confirms Hunter Biden Still On The Naughty List Despite Presidential Pardon"

https://babylonbee.com/news/santa-confirms-hunter-biden-still-on-the-naughty-list-despite-presidential-pardon

[The Margrave of Azilia]

"Patel, Ramaswamy To Celebrate Inauguration With Traditional Bollywood Ceremony"

"..."Vivek and Kash will crush the Deep State, but before that, they will perform the most special and incredible Indian dance the world has ever seen," said Trump to cheering fans. "There will be eagles and tigers and elephants and dancing women, and they will even make that cool dance move where it looks like they're trying to screw in a lightbulb. It will be almost as cool as my dance moves, but not quite, let's be honest."

"Patel and Ramaswamy seemed caught off guard by the announcement. "I'm an American from Ohio," said Ramaswamy. "I know the Electric Slide and Cotton Eye Joe. I'm not even that good at dancing. There must be some mistake."...

""Seriously, guys, I'm from New York," said Patel. "I can't dance to save my life. Please don't make me do this.""

https://babylonbee.com/news/patel-ramaswamy-to-celebrate-inauguration-with-traditional-bollywood-ceremony

[Commenter_XY]

Suppose Ana Navarro had made a false, defamatory, and libelous accusation against someone on national TV, based on a ChatGPT hallucination. She is then sued.

Who is liable?

[Brett Bellmore]

If they're a public person, and she's careful not to check if ChatGPT's output is accurate? Nobody.

ChatGPT warns you not to rely on the output to be factual, and Navarro is protected from liability so long as she makes sure not to know the accusation is false.

[Sarcastr0]

No, Brett, reckless willful blindness is not this one weird trick to avoid liability for defamation.

For fuck's sake how many years have you been on this website?

[Krayt]

So if you asked a human to make up a story about someone for fun, and another human knowingly and deliberately ran with it, the first one can get sued, too?

I won't argue the current state of law, but that doesn't seem right.

Maybe Congress needs to clear such robots as long as they have disclaimers.

[Sarcastr0]

Publication is an element of defamation. First human didn't publish; they're in the clear.

I don't think this amounts to defamation(too obvious to believe prevent something from being defamatory).

But the main thing is Brett's confident assertion of intentional laziness to avoid defamation liability is not correct.

[TwelveInchPianist]

"But the main thing is Brett's confident assertion of intentional laziness to avoid defamation liability is not correct."

Cite? Intentional laziness isn't actual malice.

Any this is true in non-AI contexts as well. If you repeat random gossip about a public figure, no defamation. But if you fact-check it and find that it's false but repeat it anyway, defamation.

[Drewski]

"Cite? Intentional laziness isn't actual malice."

It literally is, but you're not worth the time to look up a cite.

[Bwaaah]

Blindness *is* a defense. You'd have to prove it was reckless and willful, which in almost all real life cases, you can't. So people very typically *do* avoid liability for defamation.

Unsurprisingly, you hang your position on an idealized, practically inapplicable theory.

[TwelveInchPianist]

"No, Brett, reckless willful blindness is not this one weird trick to avoid liability for defamation."

Relying on Chat GPT is reckless willful blindness? You certainly haven't established that. Lots of what it says is accurate.

[TwelveInchPianist]

OK, willful ignorance can constitute actual malice.

[Rossami]

Sorry, Sarcastr0, you're wrong. Willful blindness is a negligence standard. Negligence alone is not (generally) enough to support defamation of a public person. You have to know something is false and publish it anyway.

[Sarcastr0]

I said: "reckless willful blindness."

Prof Volokh on another thread posted:
""reckless disregard" for purposes of the "actual malice" test is that,

failure to investigate before publishing, even when a reasonably prudent person would have done so, is not sufficient to establish reckless disregard.

Reckless disregard requires instead sufficient evidence "that the defendant in fact entertained serious doubts as to the truth of his publication" and thus "that the defendant actually had a 'high degree of awareness of ... probable falsity.'" "[A] deliberate decision not to acquire knowledge of facts that might confirm the probable falsity of [the] charges" might suffice, but mere careless "failure to investigate" will not."

"Careful not to check if ChatGPT's output is accurate" center-mass meets the 'deliberate decision not to acquire knowledge of facts that might confirm the probable falsity of [the] charges.'

[gormadoc]

"Wilson's never had or pardoned" seems to imply that he could, in the future, have or pardon a brother-in-law with that name. I think the possibility is remote but perhaps you know something I do not?

[Swede425]

In fairness to Ana, she's a complete dumbass so this really shouldn't be held against her.

[Commenter_XY]

Nobody tops Sunny reading the legal statements...repeatedly on the same show for different shit she said. It was hilarious.

[Mr. Bumble]

I've always wondered, who the hell watches this show?

[Ed Grinberg]

"[W]omen make up nearly three-quarters (73%) of the regular audience for daytime talk shows, such as The View or the Ellen DeGeneres Show." (source)

[KenveeB]

"Take it up with ChatGPT" is her only comment, really? I can at least understand using ChatGPT for a quick look, have there been other presidential pardons of relatives. But surely before making a public claim about it, a journalist would follow that up with a quick bit of research to find out a little bit of when and why it happened just to verify. Blindly posting something from ChatGPT and then blaming the program for errors is just ridiculous.