Editor’s Note: Retraction of article containing fabricated quotations

We are reinforcing our editorial standards following this incident.

See full article...

 

[Xenocrates]

What's being done to/about the authors?

From what I can see, they're both still Ars affiliated:
https://meincmagazine.com/author/benjedwards/
https://meincmagazine.com/author/kyle-orland/

It's Sunday in a holiday weekend, only a day or two after the event.
Like it or not, Ars has processes and HR/legal departments that mean that statements on employment and wrongdoing must be checked carefully, in order to preserve people's rights, and to discover what the actual fault was.
Those processes failed, and with two authors involved, it gets a lot harder to determine what happened and who is at fault.
A solid write up of what happened would be a fascinating insight into the failures of AI in a professional process, but is sadly very unlikely to happen.

 

[J117]

This is a good start, especially coming in the middle of a holiday weekend. But I hope there will be a fuller accounting next week -- not just what specific workflow led to this and who was responsible, but why it wasn't caught by any editorial or fact-checking process before being published. I'm not virulently anti-AI, but journalism is up there with law and medicine as a field where there should be extremely rigorous safeguards against hallucinations making it into the final product.

 

[thrillgore]

It's Sunday in a holiday weekend, only a day or two after the event.
Like it or not, Ars has processes and HR/legal departments that mean that statements on employment and wrongdoing must be checked carefully, in order to preserve people's rights, and to discover what the actual fault was.
Those processes failed, and with two authors involved, it gets a lot harder to determine what happened and who is at fault.
A solid write up of what happened would be a fascinating insight into the failures of AI in a professional process, but is sadly very unlikely to happen.

Also, Ars writers are under WGA so that complicates things.

 

[Resistance]

Please elaborate - exactly how can fabricated quotations be published in a manner consistent with the policy?

The elaboration is right here:

Ars Technica does not permit the publication of AI-generated material unless it is clearly labeled and presented for demonstration purposes.

 

[VelvetRemedy]

I feel like I'm taking crazy pills.

One (or two) of Ars's writers apparently fabricated material released as a story.

That is not "oopsies, there was a policy violation." The precise "how" of how the fabrication happened doesn't matter. It doesn't matter if the writer got the quotes from AI, from reading tea leaves, or from a floating, glowing octopus. A fabricated story is just about the worst thing that can happen to a media outlet.

 

[Meathim]

Wasn't "Benj Edwards is an AI" a bit of a running joke? Him and Kyle Orland are the names on top of the original article...

 

[coonwhiz]

1. You took way too long to post this retraction.
2. You're not explaining what's going to happen to whoever is responsible for this.
3. How many other articles have been leaning on the use of AI? You're leaving us with a ton of questions and little to no answers.

You've already lost a lot of credibility with me over this having happened at all, and your response to this is losing you even more credibility. I'm incredibly disappointed and ready to just delete my account, my bookmark, and start looking for a more responsible tech news site.

1. It seems they took the article down quickly, they just took time to post this article.
2. Ars probably doesn't know what will happen since it's a holiday weekend, and they probably haven't even had a chance to talk with the authors. Dealing with things like this take time.
3. The article literally says "At this time, this appears to be an isolated incident." Which we can infer to mean they believe only 1 article is impacted.

 

[Ubersoldat19]

Fire the author

I don't want to see anyone lose their job, but this is a serious fuck-up and termination is probably the only outcome. It's akin to cheating on your test or having someone else write a paper in college. It's a breach of faith and shows a severe lack of responsibility.

While I don't believe this was an institutional failure (at some point, you need to actually trust your people to do the right thing), this should probably lead to more editorial oversight.

As a post-mortem, there needs to be a deep dive into the other articles written by the person responsible for this. I seriously doubt that this is the first thing they wrote with AI assistance - you don't just wake up one day and decide the rules don't apply to you.

 

[VelvetRemedy]

I urge readers to assess this situation with the full body of Ars' excellent journalism in mind.

This error is not a 'for case' event. It's a 'trip to the principle's office' kind of level. Not even close to subscription cancellation.

Both of the aforementioned writers are among the finest. Both have published countless stories. Both remain trustworthy, in my mind.

How many true articles do you have to write before you can fabricate one as a treat?

 

[Resistance]

1. You took way too long to post this retraction.
2. You're not explaining what's going to happen to whoever is responsible for this.
3. How many other articles have been leaning on the use of AI? You're leaving us with a ton of questions and little to no answers.

You've already lost a lot of credibility with me over this having happened at all, and your response to this is losing you even more credibility. I'm incredibly disappointed and ready to just delete my account, my bookmark, and start looking for a more responsible tech news site.

a more responsible tech news site

Good luck with that.

 

[GeminiCB]

Echoing others that I’m waiting to see if Ars properly and publicly reckons with what happened here before I hit the “cancel subscription” button. Ars’ AI reporting is more skeptical than most sites but it has still not been nearly skeptical enough, and now this incident has happened. Those don’t feel unrelated.

 

[Jim Salter]

At this time, this appears to be an isolated incident.

That's good to hear. But frankly, this is still the kind of "isolated incident" that should be considered an immediate firing offense. This was not a peccadillo, this was an utter abnegation of journalistic work, let alone standards and integrity.

If posting slop to the front page isn't a firing offense, I have to start questioning what the job is in the first place.

 

[Acolnahuacatl]

I'm glad there is a policy forbidding the use of these models directly in preparing articles: my own experience is that reviewing every single claim in written LLM output continues to be more work than just not using LLMs, and clearly at least one Ars writer agrees. I'm also glad that this policy is enforced. Still, the situation itself is disappointing.

If someone wants their journalism to come in the form of statistically-favourable bullshit (in the Harry Frankfurt sense), there are many LLMs competing to provide them with an endless, personalised, interactive supply. If instead they want artisanal human-crafted bullshit, they can get that from social media. Tech companies in general seem to have cornered the market for nonsense.

I don't know what the future for online journalism is, if not being the last thing standing that makes an effort to achieve accuracy.

 

[msadesign]

I, for one, got so damn disappointed I already dropped my sub in anger. Yet I’m here, reading the comments. Hard to break habits after being a reader for 25 years I guess.

I urge patience.

 

[Fred Duck]

The whole situation is so strange. I have started using Gemini quite a bit in the last 3-4 weeks, and it is shocking how good it is and how much detailed information it can give me about obscure topics like particular revisions of automotive parts or configuration settings.

What's also eye-opening is just how often it's completely wrong.

...and when it's incorrect, it's confidently incorrect.

It has suggested parts that fulfill my requirements that simply don't exist, and it has explicitly told me which programming values to change to update the 12V battery configuration in my Mach-e, and when verifying, those values are in the wrong location.

It even explains why those are the correct parts or correct values.

When I correct it, it says, "well spotted! Those are the correct values for the F-150 and x other vehicle. The correct value for your vehicle is 'y'".

It's often still incorrect.

It's so important to verify these things.

Question.

If you must manually check every piece of "information" given you by Al, how much time is actually saved?

I've been reading Ars comments for months saying that 'oooh, local models are all so rubbish and the hosted ones are fire.'

Now you're telling me actually, the online ones are rubbish, too.

 

[coonwhiz]

Echoing others that I’m waiting to see if Ars properly reckons with what happened here before I hit the “cancel subscription” button. Ars’ AI reporting is more skeptical than most sites but it has still not been nearly skeptical enough, and now this has happened.

Right?! We've ween numerous articles here about lawyers citing case law that doesn't exist. I'm not sure why they'd think that using it for journalism would have a different outcome...

 

[Wheels Of Confusion]

A good model that has been brought up before covering these kinds of editorial lapses is the public retraction by This American Life, an entire episode dedicated to the process by which they found out what had gone wrong and how they were going to deal with it.
https://www.thisamericanlife.org/460/retraction

 

[Autonomous Cowherd]

I know Aurich said that a statement would be coming next week, due to the weekend and a public holiday, so I appreciate that a first statement came earlier.
What I want to know (and expect to be in the second statement) is:

• Will Ars continue to work with the authors of the retracted article?
• What steps will Ars take to ensure that something like this NEVER repeats?

Personally, I would expect Ars to not work with the authors in the future. They knowingly violated not only Ars' policy by using AI-generated text and not disclosing it, but also any journalistic standard by not ensuring the correctness of the generated text.
As for the steps to prevent future incidents of this kind, I hope for more than a generic "we reminded everyone not to do this".

 

[Alhireth-Hotep]

If there is a thread for redundant comments, I think this is the one. I, too, will want to see substantially more followup here, ideally this week. My subscription is at stake.

 

[atu30004]

When Stephen Glass was found to have fabricated a story for the New Republic, there was a long investigation of prior stories and quite a bit of public disclosure. Will that be happening here?

...and if not, WHY, PRECISELY, NOT?

 

[haleks]

I'd note that the actual situation is worse than what this retraction notes, given the content of the original story (which I read when it was first posted). I strongly encourage everyone to review Scott's take on the situation at https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/ and the link at the top of that article to the aftermath.

This is worth deep reflection by both developers and news media alike, and I encourage Ars to revisit this topic with a human-written summary of where we suddenly find ourselves with agentic systems.

Thank you.
This should be pinned at the top of the comments section IMO.

 

[JerryFritschle]

They're doing the right thing, but scoring low on style points, in particular by obfuscating the story in question. I only knew because I'd read the original, and went deeply enough into the comments to see the brouhaha start. Notably, when Mr. Shambaugh himself appeared to call out the article, the first reaction was to assume that his profile was fake.

 

[ROOT1803]

I'm sorry, but this seems like a bit of a non-statement.

What actual steps are being taken to prevent something like this from happening again?

Does the editor do anything aside from spell checking and formating? I'm not a writer, so I don't really know what everyone does, but it seems like at least one other person should be checking the factual content of articles.

 

[sockatume]

Ars’ response here, although well intentioned, raises further questions about their editorial policies and procedures. Deletion of an article with a notice as opaque as this one - “disappearing” the error, essentially - is considered extremely bad practice, arguably worse than simply leaving the article up unaltered. Together with the original fact checking failure it raises significant concerns about the editorial processes and standards of the site.

 

[kcsaff]

I hope Beth Mole's horrifying medical articles can still be trusted.

 

[thrillgore]

If there is a thread for redundant comments, I think this is the one. I, too, will want to see substantially more followup here, ideally this week. My subscription is at stake.

Mine was cancelled when I had read through Scott's post, but this can be reversed depending on how things shake out between now and July when it lapses.

I don't want to say this but there has been a pattern of poor reporting, especially on articles published on fridays; this was just the most egregious example I've seen yet, and something has to change.

 

[cyberfunk]

The policy was already pretty clear, and not fabricating quotes seems pretty obvious (AI or not), so this is not exactly helping as a statement. One head should have rolled. Two different writers were listed on the byline; which one is it so I can avoid them in the future? I would assume Benj at this point.

This ^ .

The only way that people are going to learn that betraying people’s trust this way is ruinous is by people getting ruined by making the mistake.

Pretending to write AI critical stories as a human but using AI and making a core AI mistake is the height of irony and a clear cut example of why we need a zero tolerance policy here. Once you catch somebody cheating, how can you really trust that person.

Absent a credible innocent explanation of how this happened, which MAYBE exists... Benj or Kyle or both really fucked up here. The only thing traditional journalistic summaries and news now have over AI news/summaries is trust in fact checking and sourcing.

If whoever did it is ready to give their job over to AI, maybe you should take that to its logical conclusion and let them go.


Ars’s retraction response of “apology” is necessary but not sufficient. A full correction, attribution, and assurance of corrective action needs to be made.

 

[eightycc]

Ars’ response here, although well intentioned, raises further questions about their editorial policies and procedures. Deletion of an article with a notice as opaque as this one - “disappearing” the error, essentially - is considered extremely bad practice, arguably worse than simply leaving the article up unaltered. Together with the original fact checking failure it raises significant concerns about the editorial processes and standards of the site.

Ironically, I was able to find the retracted article by visiting Slashdot.

 

[binaryvisions]

Just to underscore again, I think it's important to recognize that this is the weekend, a holiday weekend, and it's very likely that there will be a little time required to do something other than say, "whoops, we screwed this one up, so we've deleted the screw-up."

Thought I do think it would help if Ars would edit the response to indicate that further information will be forthcoming. As it stands today, the retraction reads more like, "we acknowledge this was wrong," rather than, "we acknowledge this was wrong and will be taking further steps to understand why it happened, and will keep our readers informed."

 

[dwl-sdca]

A quote was attributed to someone. That someone happened to read the article and in the comments deny making the statement. That the article was about AI and that the misquote was attributed to AI might be ironic is only a small part of the problem. I have trusted Ars to provide facts and truth. A news source must not ever assert that someone uses words they never said. In considering actions against the authors of the retracted article, would it make a difference if one of the authors themselves imagined and attributed a fantasy quote, found the false quote elsewhere and used it without verification, or allowed AI to put the words in someone’s mouth (or blog)?

How long would the misquote have existed had the falsehood not been recognized by Shamburgh himself?

 

[Feone]

I would at the very least expect a thorough investigation in the responsible authors previous works as soon as it can reasonably be done, and frankly I think the responsible party should own up as soon as possible because there are two names on the article and every second this is left in doubt is doing damage to both of their credibility.

And perhaps the general policy of regurgitating AI-related press releases (many of which are done by one of the authors of the now-retracted article) should be reconsidered as well, repeating outlandish claims with zero validation is not journalism. Presenting it as <company says> is still amplifying their message.

 

[acefsw]

The only thing I don't like about how Ars has handled it so far is that they didn't call attention to which article, and left it to the comments to find which one was retracted. If this were an article that I read and formed opinions on, I'd certainly like to know so that I can re-evaluate those opinions with new information.

Yeah, long standing, ethical journalistic practice is to leave the article up and publish the retractions at the end of the article or publish an article detailing the errors with a direct link to the article.

This is not the first time that ars has pulled articles and published a contextless mea culpa.

I'm glad they admit their errors, but they fall far short of standard journalistic best practices by omitting such context.

I look forward to a full explanation with a link to the article in question, (fwiw, I read the article before it was pulled, had also read Scott's blog post at ~3 a.m. well before ars published, and because some floated the idea that Scott changed his blog post, informed other posters that the blog post had not changed since that time).

 

[sockatume]

Just to underscore again, I think it's important to recognize that this is the weekend, a holiday weekend, and it's very likely that there will be a little time required to do something other than say, "whoops, we screwed this one up, so we've deleted the screw-up."

This is why the editorial policy at most outlets is to apply an expression of concern to the article at issue while an investigation happens. Not to hit “delete” on the entire contents of the page, and its comments section.

 

[Rapter]

I was a little surprised when the initial response to the article having issues was to delete it and lock any and all discussion of the issue down. It seemed like damage control rather than transparency.

The fact that the editors note fails to mention the article by name, and that the article being brought back was brought back in a form that completely removes the original article has made me even more concerned that transparency is not an actual desired goal.

Mistakes in journalism should NOT be buried, they should not be hidden. Trust in journalism is already at historic lows in our society and it requires the utmost effort and sincerity to maintain it. That includes owning the mistakes that ARE made, and not hiding them. Any hint of trying to paper over such errors just risks further eroding the trust that has already been breached.

The only way to rebuild that trust properly is through real honesty, real ownership of the mistake, and proper analysis of how it came to happen. This wasn't just a failure of the author to use improper tools, but also the editors who didn't even bother to verify the quotes.

This situation requires a proper post mortem, and some real mea culpas. I patiently await those steps being taken in the near future, because I still have faith that the people at Ars care and want to maintain the trust they have worked so hard to build. I hope my faith is not misplaced.

 

[countbalo]

I'd note that the actual situation is worse than what this retraction notes, given the content of the original story (which I read when it was first posted). I strongly encourage everyone to review Scott's take on the situation at https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/ and the link at the top of that article to the aftermath.

This is worth deep reflection by both developers and news media alike, and I encourage Ars to revisit this topic with a human-written summary of where we suddenly find ourselves with agentic systems.

after going through the links comment section, it still depresses me just how many people can't wrap their minds around the idea that LLM's have no mind. They can't think, yet people still thing of these things as thinking. (to be fair, my major complaint from the start, out side of copyright violation has been the interface being made to be "friendly" so that people will use it more. That more than anything has caused a lot of this. Still LLM's don't think people, nor do they feel)

 

[VelvetRemedy]

I trust their science coverage, and I take science seriously. I suppose that means I should give them the benefit of the doubt here.

If anything, it should mean you insist on high standards to regain your trust. If you take science seriously you should expect Ars's science writers to do the same.