Editor’s Note: Retraction of article containing fabricated quotations

We are reinforcing our editorial standards following this incident.

See full article...

 

[Thad Boyd]

Other viewpoints obviously exist, but I don't think Ars need concern themselves with that group

Bahahahaha yeah that's my philosophy too, dogg. Other opinions besides mine exist, but you don't need to worry about those.

 

[JStengah]

Or instead of linking to a machine roll-up of a threaded paraphrase that somebody wearing animal ears posted to Xitter, you could link to the article they were paraphrasing:

The Trust Thermocline

Might want to check your dates. The tweets from "somebody wearing animal ears" about the Trust Thermocline were made November 3, 2022. The article you linked about the Trust Thermocline was published on November 20, 2022. What does that tell us about who the original was and who was doing the paraphrasing?

 

[philipsw]

I don't think you're capturing how this incident has utterly broken the trust in the whole site's quality. It's not just trusting Benj Edwards. And that's also why I, myself, right now, don't ask for his dismissal: many things went wrong here. He was working while sick. Why. He submitted work that wasn't checked for quote veracity. Why.

If a senior AI editor must be counseled and educated on checking quotes for veracity and not working with a fever...

They're not an editor. They're a "Senior AI Reporter."
https://meincmagazine.com/author/benjedwards/

Someone else is ultimately responsible for approving what they've submitted, unless I've completely misunderstood the chain of publishing.

I get it. There is a lack of trust now. But is it a pattern, or a single mistake? That matters to me. If it was a single mistake, and they learn from it and grow, then great. If there's a pattern, then not great. Right now, nobody knows. Any calls for heads to roll is premature.

 

[RuralNinja]

You know what they say. You either die a hero or live long enough to become the villain. Sad that I'm seriously going to have to think long and hard about whether I renew my subscription for a site I've supported for so many years.

 

[adrianovaroli]

They're not an editor. They're a "Senior AI Reporter."
https://meincmagazine.com/author/benjedwards/

Thank you. I corrected that bit.

 

[graylshaped]

Referring to the entire article as AI slop is what happened here, and that is deceptive.

Were I to speculate on the point where you may be losing others: the article in its entirety is known to be the fruit of a poisoned tree. It is not just a little bit pregnant. It didn't get a little spot on it. It's tainted.

Whether justifiable or not, that author's entire corpus is now suspect. That taint affects the site as a whole. That site affects the parent company.

I'm beating the "let the facts come out" drum as hard as anyone here, but the events as they have played out do not leave much wiggle room and my advocacy is for objective due process and enough respect for the community to share with us as much as feasible at the right time.

 

[AI_Skeptic]

1. The fake text made it into the article because Benj told an AI tool to present salient parts of the article verbatim to him and it did not do that despite advertising that it could; being Benj' first use of the tool, he (wrongly) trusted it to do as it advertised and did not verify the quotes.

I'm extremely curious what tool Benj used that was supposed to present salient parts of an article verbatim. I saw someone else posted an "Academic Quote Extractor" program linked to Reddit, but I doubt that was the tool he used. I believe if there's an "LLM Based Quote Extractor" program developed, that worked 100% of the time, without fail, it would be a story in itself.

So did Benj ever say what tool he used, because my Google-Fu and Duck-Duck-Going is failing me.

 

[Distraction]

Sure, as I said, "as explained by Benj in his BlueSky post" (implicit in that is the presumption that Benj is telling the truth, to which I'm happy to say 'trust but verify', but not 'assume he is lying') [snip]

There's no way to verify that the rest of the article was written by a human. Throw some AI slop in anywhere, and you might as well have put it everywhere.

 

[squiggit]

Amusingly, this is not how soccer works. A major transgression absolutely will result in immediate ejection. In fact, a not-insignificant aspect of arguing with your buds while watching a soccer game is whether a player's first transgression merits immediate ejection or merely a caution.

Well, Soccer players given an ejection are usually ejected for the match. The system would be somewhat more controversial if ejection meant being fired and barred from taking a position at another team, per your earlier suggestion (update: that appears to not have actually been Sarty's suggestion, my bad).

So really a bad analogy all around

 

[Abulia]

I shared my feelings re: the retraction process (quite poor) and left it at that.

I'm really saddened to read all these posts advocating for dismissal/retention, and/or threatening their Ars subscriptions unless the outcome meets THEIR criteria. We don't run this site. We don't get a vote. Let the Ars staff do their thing, and when the outcome is shared, THEN vote with your wallet/views as to whether Ars has done right, by you, the reader.

The authors, editors, and their colleagues are reading this thread and the general discourse is just...sad, IMO.

 

[AI_Skeptic]

Did he say anything to indicate that he isn't using AI tools to write his articles when he's healthy? It only became apparent in this article because the person he misquoted had to say something in the article comments. Everything he has written and writes in the future should now be assumed to be inaccurate first, and needing verification, and that burden shouldn't be put on me as a reader. That's why I pay journalists to investigate newsworthy items and accurately report them to me. Sorry if he is sick, but something he was clearly already doing when healthy, and he just "oopsied" his double checking while sick, completely broke my trust not just in him, but in Ars Technica as a whole (frankly this doesn't spare Kyle either, if he knew benj was sick then he should have also paid extra attention to what he was doing).

He said he uses ChatGPT to help him write articles in an interview with Ed Zitron. The examples he gave was looking up words when he forgets them due to COVID brain fog (if I remember correctly). He did not say he uses ChatGPT to write the articles.

 

[graylshaped]

You know what they say. You either die a hero or live long enough to become the villain. Sad that I'm seriously going to have to think long and hard about whether I renew my subscription for a site I've supported for so many years.

I do every year, with all my subscriptions. We all should.

Fun story: this morning I'm going through calendar stuff with my son's mom, and she got a call back from her gym. She was looking to adjust her membership, but it turns out not only that she had to make that change on the website, but their website only worked with Chrome. She asked "Oh. That's not an option. Can you update my payment information? No? Okay, then can you cancel my membership?"

Turns out that they could do at least that over the phone.

 

[Murry Wilson]

Scrubbing the person's name who used an AI agent to write their piece and failed to double-check does ARS and it's readership a disservice. It should be highlighted and underlined and be a black mark on their career forever.

I'd like to be aware who the lazy journalists are so I can avoid them in the future.

 

[adrianovaroli]

Well, Soccer players given an ejection are usually ejected for the match. The system would be somewhat more controversial if ejection meant being fired and barred from taking a position at another team, per your earlier suggestion.

So really a bad analogy all around

A direct red card (i.e. not one following a previous yellow card) means an automatic suspension for the next match, and the sanction can also span more matches, depending on the gravity of the action. It can also be accompanied by a fine. It's not just "missing the rest of that particular game."

 

[squiggit]

I'm really saddened to read all these posts advocating for dismissal/retention, and/or threatening their Ars subscriptions unless the outcome meets THEIR criteria. We don't run this site. We don't get a vote.

lmao.

Anyone subscribing gets a 'vote' ... at least in terms with their wallet. Implying people somehow don't have the right to cancel their subscription if they find this objectionable is so goofy to say.

I think a lot of people in this thread are overreacting but they owe absolutely nothing to Ars here. There is no moral or ethical failing in deciding to spend their money elsewhere for any reason.

 

[Sarty]

Well, Soccer players given an ejection are usually ejected for the match. The system would be somewhat more controversial if ejection meant being fired and barred from taking a position at another team, per your earlier suggestion.

Maybe someone else said Mr. Edwards should be barred from the industry--I don't think I did. What I did more or less say that if he ever hopes to convince any particular journalism outlet to hire him, he has some enormous work ahead of him. No individual outlet is obligated to hire him. He doesn't have a perpetual right to a job in this field, certainly not if he is a known fraudster.

The outcome of those two concepts is probably the same--I somewhat doubt I'll ever see another Benj Edwards byline at a reputable outlet. But I think the difference is still very important.

 

[graylshaped]

While true, the retraction itself is tainted at this point. The framing implies the entire article is fabricated, as in the story itself is fake, when what we're actually talking about is someone being misquoted.

In other circles (and indeed, earlier in this comment thread) people have taken the retraction of the article to suggest that Shambaugh is the one who lied or fabricated information, because the entire story was retracted rather than the misquote simply being corrected.

The staff response here is wildly irresponsible.

I'd suggest making shit up in the absence of information is irresponsible, not removing information known to be false, but it's not the first time teh internets got a wild hair up its ass.

 

[philipsw]

While I don't think this was directed at me specifically, I always find it interesting when people on the internet tell me that the fact I have a different opinion from them is solely because I'm emotionally upset, angry, or whatever when I'm actually feeling quite calm and composed. It may actually be the closest I'll ever come to knowing what my life would have been like if I'd been born into this society as a woman.

No, it was not directed at you specifically. I'm not sure I've read your previous comment(s). My only point was that this is a very emotionally charged thread. And certain allowance might need to be made to accommodate that.

I'd just like folks to step back for a moment and consider some different perspectives.

 

[squiggit]

Maybe someone else said Mr. Edwards should be barred from the industry--I don't think I did. What I did more or less say that if he ever hopes to convince any particular journalism outlet to hire him, he has some enormous work ahead of him. No individual outlet is obligated to hire him. He doesn't have a perpetual right to a job in this field, certainly not if he is a known fraudster.

The outcome of those two concepts is probably the same--I somewhat doubt I'll ever see another Benj Edwards byline at a reputable outlet. But I think the difference is still very important.

Apologies then! I interpreted that line as directly advocating for that result.

 

[graylshaped]

lmao.

Anyone subscribing gets a 'vote' ... at least in terms with their wallet. Implying people somehow don't have the right to cancel their subscription if they find this objectionable is so goofy to say.

I think a lot of people in this thread are overreacting but they owe absolutely nothing to Ars here. There is no moral or ethical failing in deciding to spend their money elsewhere for any reason.

THIS I completely agree with.

 

[datroof]

Please, please, please, give us a more concrete intent to fix problems like this going forward, Ars Technica. I still have confidence that journalism is an important industry, but only to the degree that I have confidence in journalists.

 

[graylshaped]

This is an excellent comment, and it really underscores just how badly Mr. Edwards fucked over his colleagues.

It underscores how this incident has exposed a vulnerability in Ars editorial practices.

This incident will be a case study at some point. There are (at least) four intertwined stories, and I know much earlier in this thread I said "two":

1) The original, still fascinating "AI agent demands justice!";
2) The author using an LLM to fabricate quotes and being caught at it by the target of the AI;
3) The PR damage control response by Ars;
4) The state of journalism in the age of LLM tools

 

[SomewhereAroundBarstow]

No, it was not directed at you specifically. I'm not sure I've read your previous comment(s). My only point was that this is a very emotionally charged thread. And certain allowance might need to be made to accommodate that.

I'd just like folks to step back for a moment and consider some different perspectives.

What I was reacting to was your characterization of "nearly this entire thread" as emotional. While I've certainly seen some over the top appeals to emotion I wouldn't say that they dominate the discussion. I've read, I believe, every comment and if you aren't sure you've read my earlier ones that suggests to me that maybe any statement about the emotional tone of the entire thread might not be as objective and dispassionate as you think.

 

[AI_Skeptic]

Except that's not what I said. Please don't lie by putting false words in my mouth. What I actually said was "everything in the article was written by humans, except the quotes".

"Everything in the article except the quotes" is a huge problem. It's even a huger problem that the quotes were from a website that could easily been verified using a copy/find/paste. It seems like the person who was incorrectly being quoted took it in stride, but what if he didn't? Condé Nast is a huge company, and if the person didn't take the quote in stride, they could have sued. It shows negligence on Benj's part not to do a quick verification that would have taken just a minute.

This also damages the likelihood of people wanting to be interviewed by Benj. For example, I wrote a book about AI. If Benj wants to interview me, would I let him? No. I do not trust that he's able to quote me without making mistakes. If the interview is done orally, Benj uses a voice-to-text translator, then runs it through an LLM, and tries to pull quotes from it, would it quote me correctly? Maybe - or maybe not. Would others feel the same? Yes. Would others want to take the risk? I doubt it.

Now, should Benj be fired? I think he should be dismissed from the AI beat, but not dismissed from reviewing video games or classic computers or other non-AI subjects.

 

[philipsw]

What I was reacting to was your characterization of "nearly this entire thread" as emotional. While I've certainly seen some over the top appeals to emotion I wouldn't say that they dominate the discussion. I've read, I believe, every comment and if you aren't sure you've read my earlier ones that suggests to me that maybe any statement about the emotional tone of the entire thread might not be as objective and dispassionate as you think.

If you feel I in any way discounted your comments, I apologize. It was not intended.

 

[SomewhereAroundBarstow]

If you feel I in any way discounted your comments, I apologize. It was not intended.

It's not about me. I feel you've discounted the actual level of discourse of nearly everyone else.

 

[torque2k]

I don't know about Ars, or other publications in general, but many of the tools I use where AI is integrated require you to "sign off" when you do actually use AI output. Is there any kind of manual check in place (acting basically as a legally gray CYA) for reporters? Asking because I actually am interested, not because I wanna pick a fight... (sorry for the low resolution, it's from a webinar I'm watching rn)

 

[Thad Boyd]

You know what they say. You either die a hero or live long enough to become the villain.

I think Two-Face would approve of being referred to as "they".

 

[Sarty]

Now, should Benj be fired? I think he should be dismissed from the AI beat, but not dismissed from reviewing video games or classic computers or other non-AI subjects.

This one merits pondering. I might be able to live with something like that. Specifically, no articles about anything serious.

Whether it's tenable to retain a second-class reporter--someone specifically barred from actual heavy topics--is an interesting question. Does the reputational damage still attach? Is that viable from an office politics perspective? Would Mr. Edwards even be interested?

 

[Thad Boyd]

ARS

ARS

why are you doing that

 

[ol1bit]

It's not easy to tell if data is AI generated or not, and it's only going to get worse. Thanks for letting everyone know!

 

[syali]

Every mistake is a chance to do better next time.

 

[Marlor_AU]

And you also made me think about web directories for the first time in a long time. Being a human-maintained directory was what made Yahoo! so useful back when it was useful, but obviously that's not practical with the web at its current scale. But what if DMOZ was revived and AI was used to categorize web content according to its hierarchy? There still would be a lot of AI noise polluting the findability signal but it would be interesting to see if going back to the directory approach might be helpful. Of course, AOL and Yahoo! being the longsighted geniuses they are the original DMOZ project page is long gone and even the unofficial mirror is only available on Archive.org now.

Every time I search for technical information, only to find a mess of auto-generated, referral-link infested slop promoted via SEO, I keep thinking of how a modern web directory could help - even if it only covered a small proportion of sites.

Ultimately, when looking for information, the best option is to go straight to a website that is likely to have the information to begin with. Astronomy gear? Skip the horrendously-incorrect slop sites full of random referral links and go straight to Cloudy Nights or Astrobin. Computer gear? Skip the search engine and look to see if Ars or TechPowerup has a review. Search is rapidly becoming worthless, and we need to fall back on trusting individual sites.

But how to know which sites even cover a topic to begin with? It used to be that you just went to Yahoo Directory or DMOZ to find websites on a particular topic. These days, you need to stumble across them. Returning to a directory-based approach could help with discovery.

But there would need to be some way to keep the slop factories out. If the directory was fully open and they could just spam their multitude of identikit sites in to every category, that would kill the directory as quickly as they killed web search.

Even better would be some kind of carefully-curated list ranked by reputation (as assessed by experts in the field). But getting experts to even agree in the first place would be a challenge.

 

[AI_Skeptic]

I don't know about Ars, or other publications in general, but many of the tools I use where AI is integrated require you to "sign off" when you do actually use AI output. Is there any kind of manual check in place (acting basically as a legally gray CYA) for reporters? Asking because I actually am interested, not because I wanna pick a fight... (sorry for the low resolution, it's from a webinar I'm watching rn)

View attachment 128425

A better question to ask first is
"What type of AI does Ars use anyway"?

I doubt officially Ars uses any type of AI that generates the final product that's read by the users. They may use AI tools that collect notes, find patterns, etc. Honestly, I'm curious about the "behind the scenes", what type of tools exist for an Ars journalist?

Benj DID use Claude and ChatGPT that is outside of Ars/ Condé Nast systems, and ChatGPT somehow came up with the quote (according to Benj's Bluesky post). His post never explained how ChatGPT came up with the quote.

 

[Thad Boyd]

It's not easy to tell if data is AI generated or not, and it's only going to get worse. Thanks for letting everyone know!

Yeah, Sambaugh's followup post has some interesting thoughts about how the more interesting implication here isn't Ars using fabricated quotes, it's an AI agent being used for targeted harassment -- something that could be done at scale against pretty much anybody.

He adds that it's not clear at this point whether the agent was prompted to retaliate against him or whether that was an unintended consequence, the AI "choosing" to do it as a response to his thwarting its programmed goals, and that the latter is possible and has its own set of frightening implications.

 

[Thad Boyd]

Every mistake is a chance to do better next time.

Well, no; some mistakes foreclose the possibility of a "next time".

 

[paw]

"Everything in the article except the quotes" is a huge problem. It's even a huger problem that the quotes were from a website that could easily been verified using a copy/find/paste. It seems like the person who was incorrectly being quoted took it in stride, but what if he didn't? Condé Nast is a huge company, and if the person didn't take the quote in stride, they could have sued. It shows negligence on Benj's part not to do a quick verification that would have taken just a minute.

This also damages the likelihood of people wanting to be interviewed by Benj. For example, I wrote a book about AI. If Benj wants to interview me, would I let him? No. I do not trust that he's able to quote me without making mistakes. If the interview is done orally, Benj uses a voice-to-text translator, then runs it through an LLM, and tries to pull quotes from it, would it quote me correctly? Maybe - or maybe not. Would others feel the same? Yes. Would others want to take the risk? I doubt it.

Now, should Benj be fired? I think he should be dismissed from the AI beat, but not dismissed from reviewing video games or classic computers or other non-AI subjects.

An alternative view is that having been burned so badly by this episode, Benj is perfectly placed to do an autopsy of the process, and is very unlikely to repeat the same mistake again.

Re transcribing interviews: it’s trivial, depending on the tool, to click the relevant quote and hear the actual audio to verify. If anyone’s going to do that, it’ll be benj!

How would you want to be treated in a similar situation? Have you (and others) never made a professional mistake? And for those of you who have, and got the cold sweats when you realised, didn’t you swear to never let that happen again?

Journalism is so important, but nobody actually died from this mistake, and the scar tissue is invaluable.

 

[jasonmicron]

In a non-related future story, this retraction seems to be on track for the yearly Ars "Most commented" articles for 2026 they post in December. Yikes.