Ars Fires Reporter For Accidentally Using Fake AI Quotes
from the I'm-sorry-I-can't-do-that,-Dave dept
Last month we reported on a strange story in two strange parts: first, a coder had his AI agent create an entire smear campaign against a coding repository volunteer because he rejected AI code. Second, an Ars Technica journalist named Benj Edwards used a bunch of quotes made up by ChatGPT in a story about the saga without fact-checking whether or not they were actually true.
Edwards was also very up front in terms of explaining and taking direct ownership of the screw up, noting he was sick with the flu at the time he wrote the story. Ars was also refreshingly up front about it, issuing an editor’s note apologizing for the error.
Edwards says he first tried to use Claude to scrape some quotes from the engineer’s website, but that was blocked by site code. He then turned to ChatGPT to farm quotes from the site, but ChatGPT decided to just make up a whole bunch of stuff the engineer never said (this is a pretty common issue).
Just cutting and pasting quotes probably would have saved the journalist a lot of time and headaches. And his job, apparently, since Ars has since decided to fire Edwards, something Ars doesn’t seem interested in talking about:
“As of February 28, Edwards’ bio on Ars was changed to past tense, according to an archived version of the webpage. It now reads that Edwards “was a reporter at Ars, where he covered artificial intelligence and technology history.”
Futurism reached out to Ars, Condé Nast, and Edwards to inquire about the reporter’s employment status. Neither the publication nor its owner replied. Edwards said he was unable to comment at this time.”
There are several interesting layers here. The biggest being that AI isn’t an excuse to simply turn your brain off and no longer do rudimentary fact checking.
At the same time, this can’t really be unwound from the fact that media ownership rushed to tightly integrate often under-cooked LLM models into an already very broken journalism industry with the obvious and primary goal of cutting corners and undermining already-struggling labor.
The pressure at most outlets for journalists to generate an endless parade of content without adequate compensation or time off creates in increased likelihood of error. The overloading (or elimination of) editors (with or without AI replacement) compounds those errors. That the end product isn’t living up to anybody’s standards for ethical journalism really shouldn’t surprise anybody.
Filed Under: ai, ai agent, automation, benj edwards, code, fabrications, journalism, llm, media
Companies: ars technica, conde nast
Comments on “Ars Fires Reporter For Accidentally Using Fake AI Quotes”
Editor double check
Shouldn’t his editor have checked the quotes?
Accident...
He slipped and fell onto two different LLMs and not proofreading their output before publication.
Re:
I don’t mean to be overly harsh to the guy, but as their senior AI guy he is well aware of the hallucinations that plague their output so if anyone on their staff knew to double check their output, it was him. Not checking, or having the other credited writer on the story look over it before publication is a major lapse of judgement.
That said, if it was a one off, firing him is an overreaction, it’s not like they haven’t demonstrated a consistent blindspot when it comes to certain rocketry writers lack of journalistic integrity.
Re: Re:
No, using fake quotes is one of those “not even once” kinds of fuckups, like plagiarizing someone else’s work or reviewing a book that doesn’t exist. I like Benj, I feel bad for him, but this was 100% a fireable offense.
I think former Ars writer Jim Salter makes a persuasive argument in the comments (in response to the article saying “this appears to be an isolated incident”):
Later on in the thread he adds:
Jim gets it; he knows the job. That means he knows the position Benj is in, he knows the stresses he’s under and is sympathetic to him, but it also means that he knows what the rules and responsibilities are, and what the consequences need to be when someone breaks them.
I agree that Berger’s access journalism WRT Musk is the more serious problem, but that doesn’t mean this wasn’t a fireable offense. Ars abdicating its editorial responsibilities in one instance doesn’t mean it should have abdicated them here too.
Re: Re:
I suspect this “one off” kicked off a performance review, and it was noted that he hadn’t been properly citing his use of AI to the employees editing his pieces. He wasn’t the only one who reviewed the content, but it appears he was the only one who knew AI had been involved in writing the piece.
But that’s pure conjecture. It could also be as simple as the fact that his name has now become associated with AI abuse at Ars, so Conde Nast moved him elsewhere.
Re: Re:
Or maybe this was the appropriate reaction, and the other example you give (if valid; I don’t know) is an underreaction.
This person made a major blunder and blamed it on the very technology they’re cited as a senior expert on. Plus, anyone getting to the point of being “senior” should damn well know better than to work while significantly sick. A firing seems OK to me; but if the reporter had good reason to fear retaliation for taking legally- or contractually-mandated sick days (if so mandated), and decides to sue over that, I’ll be on their side in that regard.
Re:
Not surprisingly really. The main outcome from the use of AI is that it makes people lazy and uncritical.
You can’t just use this thing that helps you turn off your brain and expect that you will analyze the thing that took no effort(on your part) properly.
They disappeared the article and its comments, with no acknowledgement or explanation for several days, before finally replacing it with a stub that contains the original headline but not the original byline or any other content that was in the article or comments.
They put up an editor’s note that linked to the stub and acknowledged that it was retracted for containing fabricated quotes, but provided no further context or explanation. They did have the decency to apologize by name to Scott Shambaugh for attributing false quotes to him, and that’s laudable, but I would hardly describe their overall response as “refreshingly upfront.”
There are, of course, legal reasons not to talk about personnel decisions, but the only further acknowledgement we’ve gotten that Ars has taken remedial steps is that Benj’s bio now uses the past tense.
That’s something, at least. I like Benj but this was not a small screwup; I hope it doesn’t end his career but I think it’s right that it ended his job. “Never falsify quotes” is journalism 101 stuff; you just don’t do it, and a reputable publisher can’t tolerate it when somebody does do it.
Re:
They disappeared the article and its comments
Yes, they took down an article that was falsified and then, quite rapidly by the standards of this kind of thing put up an explanation and apology. They’re not going to keep the original article up — it was, as I mentioned, falsified. The level of expectations in some of these commentaries is just ridiculous.
Re: Re:
But the internet is supposed to be foreer, dontchaknow?
Re:
Adding: I’ll quote Jim Salter, who used to write for Ars himself, again here (or for the first time, since my previous comment is in the moderation queue and this one might show up first):
So no, Ars hasn’t been “refreshingly upfront”; it has not adhered to best practices in its retraction.
Maybe he needs to learn to read before trying to write?
“Edwards says he first tried to use Claude to scrape some quotes from the engineer’s website, but that was blocked by site code.”
If he intends to use quotes from this fellow’s website, why wouldn’t he cut-and-paste the relevant sections instead of trying to use some long-way-around-the-bush tool like Claude to do it for him?
He’ll have to read the quotes he’s intending to use anyway to be sure that he understands what they say (because that’s the point of writing about them).
The process is: Read the relevant articles/postings, highlight the relevant parts and past them into your article, properly attributed and clearly marked as quotations.
It seems like this fellow didn’t actually do any of that.
If he’s a writer, I’m a rhinoceros.
Re:
I dunno, that all sounds like a lot of work. Can’t we just let the hallucinating bullshit machine do it?
Why would you even use Claude to scrape the stuff? If you can see the damn things in the first place, even if literally every other means of copying them via a computer is blocked through technical means…
YOU COULD’VE WRITTEN THEM DOWN ON A PIECE OF PAPER
Re:
Pen and paper? What are we, barbarians?
/s
Re:
This is the thing that gets me. How was using AI in this case going to be any easier than copy pasting himself?!
I’m not exactly happy the dude lost his job, but all things considered, this was something Ars Technica needed to do for the sake of its own credibility. Now the site needs to enact some rules that will help prevent this situation from happening again—like, say, asking people to not work while they’re sick.
Re:
Most orgs say “don’t work when you’re sick,” but just because they say it doesn’t mean they’re incentivizing it.
There are lots of reasons someone might work sick, from worrying about losing PTO days to a media environment that incentivizes timeliness over thoroughness.
There need to be real cultural changes here. Ars needs to do its part, but it’s bigger than Ars or Conde Nast; it’s American work culture in general.
Re: Re:
Maybe the politics of sick leave should be Edwards’s next (human-written) story.
At my last job, we all got 10 paid sick days per fiscal year, to be used for one’s own sickness or for looking after one’s sick children. Sick parents coming to work were a common sight by the end of the fiscal year (right at the end of cold and flu season), especially if they had multiple children. Whereas single people who seemed entirely healthy the day before would often suddenly “come down” with something that lasted as long as their paid time off did, usually when the weather got nice.
Federally, the U.S. goverment mandates 0 sick days. I don’t know what rules Ars works under. Another concern is whether proof is required. If someone’s expected to book an appointment with a doctor just to get a note, it’s often just easier (and maybe cheaper) to work.
Re: Re:
TL:DR
Sorry I dont have time to read your comment. Too much work to do here
He probably would’ve been fired for not using AI enough, if he didn’t.
i’d say fuck the industry, but it has pretty well fucked itself.
Accidentally?
Go fuck yourself you lying sack of shit.
Re:
Is there any proof that he used those quotes while knowing they were wholly fake? I mean, Ars did the right thing by firing him for this error, but it’s possible the error was made in a misplaced bout of good faith.
Re: Re:
It’s an accident in the same sense that a drunk driver’s collision is an accident.
Re: Re:
I don’t think accidentally/knowingly is the dichotomy here.
If, say, he had included the same quote twice, then, sure, that rises to the level of an “accident.”
Not properly verifying that your quotes are accurate goes beyond being an accident: it’s journalistic malpractice.
Maybe, maybe, you could excuse it down to the level of being an accident if this were a person unfamiliar with AI and the ways they can “hallucinate.” But Benj Edwards’ job was to write stories about AI (as well as retro gaming) for an audience of techies. Heck, this was a story about AI, because that was Edwards’ beat.
If my phone buzzes and I look down and bump into someone going around the corner, that’s an accident. If I blindfold myself and run around, spinning wildly, with my arms extended, on a busy transit platform…
It’s not simply an “accident” if you know beforehand that it’s a possible (or even likely!) result of your actions, and don’t take basic due diligence steps to make sure your work meets the basic standards for your profession.
To say nothing of the fact that Ars has standards about how AI can be used to help write stories, and how that help is to be disclosed in the story, and it’s pretty hard to explain Edwards using AI to assist his writing, and then not mentioning that in any way in the story, as anything but an intentional breach of those standards.
Re: Re: Re:
Fair points. 🫴
Another two cents...
Firing may be appropriate for the level of embarassment caused, but this shouldn’t end his career. He owned the mistake and did his best to rectify it quickly. Here’s hoping he finds a new employer quickly. Or at least as quick as possible in this screwed up journalism/media industry.
Re:
I’ve generally enjoyed Benj’s retro gaming journalism. Maybe he can stick to that for awhile and stay away from the AI beat.
If artificial intelligence cannot combat natural stupidity, then what is the point?