Editor’s Note: Retraction of article containing fabricated quotations

We are reinforcing our editorial standards following this incident.

See full article...

 

[Aelix]

Reading Benj’s explanation, I was going to say something to the effect that the only way to avoid this would be to have two computers, one where you can use AI for research, and another for the article, and you can never cut and paste between the two… but even that is not enough. You could just look at the fabricated quote and type it out with your hands.

It’s only going to get worse as AI is inserted everywhere. It’s in freaking notepad.

 

[Jim Salter]

Hey, uh, I'm pretty sure I found the retraction through my Mastodon feed, and boosted it. But now that post is gone. Did you actually delete the mastodon post+thread for YOUR OWN RETRACTION?

This is feeling much less transparent than I wanted to believe. Thank the gods for RSS.

I saw it in my own Mastodon feed from the arstechnica flipboard account. Didn't see it in the mastodon.social feed. I'm honestly not sure if Ars even runs the "arstechnica" profile on flipboard, but it looks official, so maybe that's what you remember popping up in your feed?

I hope so, anyway. Although either way... The retraction really ought to be going out on social media feeds just like any other article does.

 

[AncientToaster]

I think I'm even more disappointed in the way the retraction was handled than in the original error.

I’m more open to AI than many Ars commenters, I think, and using an LLM workflow to pull quotes and messing it up is a human mistake I can imagine making. That wouldn’t justify it (especially not checking the accuracy yourself), but it’s imaginable. I may be a bit biased as a fan of Benj’s video game history work long before he started writing for Ars.

But the weasel retraction is really poor form. Not linking the original article, obscuring the original article, hiding the names of the authors, and deleting the comment thread? Transparency this ain’t.

Errors happen and employees screw up. But hiding those mistakes, especially in journalism, is really bad for trust.

 

[hillspuck]

Shout out to @Aurich and any of the other moderators (who I unfortunately don't know by name) working in the moderation mines for this. You have undoubtable zero meaningful connection to what happened here beyond it being your employer, yet you're having to read through numerous (justifiable) comments on the story. I've been reading for a while now and every time I hit next page, there are more pages...

 

[xoa]

I'm willing to extend a tiny bit of grace here. Guy whose job it is to learn about and report on AI finds out about a new tool that is advertised as being suitable for extracting verbatim quotes and decides to try it out, in a haze of illness he fucks up very badly.

COVID often really fucks with your cognitive function, the first mistake was to continue working when sick, and that mistake can be partially attributed to the illness itself, as can all the following mistakes.

This explanation, even if you remove COVID from the equation, is slightly less troubling than what I had assumed.

I'm not sure what the best way to move forward is, personally I would like to continue reading reporting by Benj Edwards here on ars.

Yeah, this is reasonable and something I will spend some time thinking on. But actual, legit infection(or poison or whatever)-induced-literal-brain-malfunction, causing someone to do something insane in the literal sense, would be a genuinely ameliorating factor if the organization then acts in a systemic way to prevent that being a SPOF again. I can see that being a blind spot for an org that starts fairly young and in tech and it's something I've contemplated on as well as I age and watch my parents (and before they passed recently grandparents) age. We are all running on actual physical wetware at the end of the day, and while normally that's reliable enough we can act as full abstraction, the risk of malfunction and degradation is real, even for the best of us. As in the airline industry and so on it's up to the organization to help deal with the reality of messy humans in the loop. Maybe here there needs to be a much stronger policy around "when you're at all sick, all deadlines are suspended" or something.

Though that wouldn't excuse how the response played out, which as I said I feel was kinda fumbled and is a separate area for improvement from the authorship itself. An official retraction policy document that the community could see and would get followed each time might be helpful. Checklists made and thought through well ahead of time are most valuable precisely when everyone is most stressed, emotions are running highest, and things feel crunched as it seems may have happened here.

 

[Rajjn]

Wow, the convenience of using AI to scrape Shambaugh’s blog for quotes versus just visiting the blog and doing it themselves seems negligible. If the authors were that lazy, their use of AI certainly didn’t stop there. I appreciate Ars’ acknowledgement of the issue - but the standard was already in place and was still disregarded by the writers. That’s a bigger issue that Ars must get a handle on, and quickly.

 

[MilanKraft]

...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.

[snip]
[and then] It's often still incorrect.

Not trying to be a jerk but this kind of overly-lenient attitude among those experimenting with LLMs needs to stop. The bolded parts are literally contradictory.

An LLM can't be "shockingly...good" AND (very) "often wrong" and then when corrected "still wrong", at the same time. The often wrong and still wrong parts, mean these are flat-out crappy products propelled by a lot of false advertising and spin. Full stop.

If people actually hold LLMs to the same standards they'd hold a family member or co-worker (when relying on them for something even somewhat important), this sentiment would be more accurate:

"Well, sometimes Jim does a good job and when that happens it's pretty impressive. But he makes so many mistakes the rest of the time, even making additional mistakes when called out instead of correcting the problems rapidly, that he's completely untrustworthy. Unless the thing you're asking him to do is so trivial that it literally doesn't matter if he gets it wrong, don't rely on Jim."

 

[sjl]

I appreciate Benj's clarification, but -- aren't both Benj and Kyle now taking to Bluesky to comment on a situation on which they were specifically requested not to comment?

They were specifically requested not to comment until Ars management had said their piece in public. Presumably, once the retraction note was posted, they were clear to make their own statements public.

I don't think it's unreasonable for Benj to make a public post taking responsibility for the screwup, especially if it's after management has said, "right, we've said our piece, go nuts". It's the responsible thing to do, especially since there are two potential culprits.

Now it's for Ars management to look at what Benj (in particular) has to say - both in public and in private - and decide what to do, and how much to make public about the process. My view is that more transparency is better, with the caveat that there are good reasons to keep some of the details behind closed doors, sometimes.

 

[User_E]

I don't know what that policy is, but as someone who spent many years working remotely, you just don't want to waste your PTO on being sick. You probably are going to sit in front of the computer anyway because what else are you going to do, you aren't going to infect anyone, and usually you think it's better to just muddle through rather than spending a limited resource or let down your co-workers.

I don't know that this is necessarily always the right attitude, but I think it's very common.

That just means the PTO is inadequate. I know that's very common in the US, but that doesn't mean it's right.

Really, sick leave should be unlimited and coordinated with the company disability plans (so disability can take over if the employee is sick enough to be out for an extended time). It's absurd that we accept having a limited number of days to be sick. I'm fine with having reasonable guardrails to prevent abuse (requiring a doctor's note for extended absences).

 

[choco bo]

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.

This has been the case with "AI" for a long time.

I can not count the number of instances where "AI" gave me very, very confident instructions on how to configure certain piece of software, or how to generate the configuration files or which options to use.

The only problem was - instructions were completely, and I mean completely wrong.

Not to mention that it would try convincing me that software used SQLite DB for configuration, while it was nothing like it. And so on, and so on.

I use "AI" as a very, very good search engine. It excels in this task.

Anything else - yeah, no. Not yet, at least.

 

[CatNamedHugs]

This is great to hear -- positive reinforcement for handling a mistake promptly and responsibly. Thank you for rewarding good behavior!

It will be interesting to see what follow up comes from the editors. I agree with others that this statement is a good start, but I want to see more. Why did this happen? Were other articles affected? Perhaps even a word from the people involved, explaining why they did this.

Is this AI?

 

[matchstick_1]

I think this indicates that editorial policy has to be altered to ban any kind of LLM tools, including for research and background tasks

Any class of tool where the failure modes includes the potential to fabricate compete bullshit have no place in journalism.

 

[multimediavt]

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.

1. That's not how an article "retraction" works for most credible media outlets; that's how you cover a mistake by deleting it (trying to, thank you Internet Archive) and making some PR statement as a band aid on a bullet wound
2. "Probably"? Unless they have some really unprofessional people working for them they damn well better know as a lawsuit from Mr. Shambaugh I'm sure was on the table, if not still is.
3. The statement "says", it's not an article. And the statement is lacking from a professional journalistic point of view, and the actions so far are nothing short of dubious.


I cancelled my subscription some time ago after learning Ars' parent company (Conde Nast) still advertises with X/Twitter, and when Ars still embedded tweets rather than quoting and linking them thereby not allowing X cookies and trackers to enter readers' computers.

Some of the contributors I still like to read here should consider their own avenues of expression. This one has lost its integrity and then some. Good luck in your continued enshitification endeavors Ars. Me thinks you may be going the way of /.

 

[SuperDave]

That's it, I'm done then, I'm cancelling my multi year subscription. It was fun guys, but if your staff are going to use LLMs to write your articles without disclosing it (or even at all), I'm out. I'll go back to being a freebie reader with Adblock enabled.

In other words, you're going to ragequit because of what you read, and then keep on reading.

 

[xoa]

I think I'm even more disappointed in the way the retraction was handled than in the original error.

I’m more open to AI than many Ars commenters, I think, and using an LLM workflow to pull quotes and messing it up is a human mistake I can imagine making. That wouldn’t justify it (especially not checking the accuracy yourself), but it’s imaginable. I may be a bit biased as a fan of Benj’s video game history work long before he started writing for Ars.

But the weasel retraction is really poor form. Not linking the original article, obscuring the original article, hiding the names of the authors, and deleting the comment thread? Transparency this ain’t.

Errors happen and employees screw up. But hiding those mistakes, especially in journalism, is really bad for trust.

Indeed, the old saying "it's not the crime, it's the coverup" has stuck around for good reason. In this case, putting the notice of retraction, link to retraction article(s), a {strike} around the whole original article and delisting it, but leaving it up with the comments thread up as well (but locked from further replies) seems like a better response. Nothing to hide, no links broken, people can see what happened, while also having it be very clear that it's retracted.

That just means the PTO is inadequate. I know that's very common in the US, but that doesn't mean it's right.

Really, sick leave should be unlimited and coordinated with the company disability plans (so disability can take over if the employee is sick enough to be out for an extended time). It's absurd that we accept having a limited number of days to be sick. I'm fine with having reasonable guardrails to prevent abuse (requiring a doctor's note for extended absences).

Maybe with the ever rising tide of AI, journalism organizations will start having to treat people more like aircraft pilots or the like, where if someone is sick enough it's effectively a "safety risk" and they should just be outright forbidden from any further work (enforce it technically too! disable the VPN etc) until they're recovered. Or at least any public facing work, maybe doing some equivalent of desktop clean up is ok, but nothing in the hot seat. Both for them and for the org. It's true that a lot of people might just try to work remotely normally but perhaps enforcing stronger work/life separation (you will take some time and relax and you will like it ) in this age of blurring lines.

 

[bushrat011899]

This is also concerning insomuch as it suggests Ars staff may not have adequate sick leave, or may have a culture that discourages its use.

I wouldn't attribute that to Ars. It's extremely common in white-collar work to just soldier through sickness, and that's only been made worse by remote work (ironic, given its origins in COVID). I know I've worked from home plenty of days when I probably should've just taken advantage of a sick day.

 

[sarusa]

Alright, thank you Benj for acknowledging this. Personally, what I'd like going forward is a promise that such "tools" won't be used again, and I don't think that's an unreasonable request.

The policy already in place is that no AI stuff gets used without full disclosure, which was violated. So they can 'promise' again, but it's really all up to the authors to say they won't use AI and actually comply with that. The editors can't fact check every single thing in an article - if they could, they wouldn't need authors, could just write it themselves. So a reasonable sounding quote from someone goes right by, because why would an author just make that shit up when they're supposedly not using chatbots?

If the last sentence was just meant for Benj, then sure, but I think it's too late for his career now. He'll have to go work for places like WaPo, Fox, or Kotaku where they just don't care if anything's real. Which is a real tragedy, your entire journalistic trust gone up in smoke with one article - but then he was the one regularly using AI chatbots to write articles. The 'oops' here is that he copypasta-ed the 'wrong thing' while sick. But using them in the first place is a big NO, especially if you're getting things that really need to be correct, like quotes from the person involved.

 

[loneatom]

I’ve been a reader and subscriber of Ars Technica for a long time in large part for what I’ve perceived as irreproachable journalistic integrity.

It’s of the utmost importance that Ars Technica show complete transparency and conduct/publish a thorough analysis about what went wrong, and which contributors made this situation possible. That’s the only way to preserve trust.

 

[hillspuck]

2. "Probably"? Unless they have some really unprofessional people working for them they damn well better know as a lawsuit from Mr. Shambaugh I'm sure was on the table, if not still is.

There is an almost zero chance of a lawsuit. While false, the quotes did not in any way harm Shambaugh's reputation. Plus it would be incredibly hard they were made intentionally rather than through stupid choices. The article overall was very supportive of Shambaugh.

Defamation law doesn't allow you to sue (well, successfully) someone for claiming you said things that were plausible for you to say.

(=, I agree with the rest of your post and just needed to make this particular point)

Edit: I will say they are incredibly "lucky" that if they had to have something like this happen, it didn't make up something that was defamatory. Though they'd still probably get off the hook with the explanation that there was no intent to harm and it was just stupidly using the AI. Which just makes me shake my head at how this will become a go-to excuse and might actually work under defamation law. And I can't think of any good way to patch it that doesn't empower SLAPPers.

 

[TylerH]

Reading Benj’s explanation, I was going to say something to the effect that the only way to avoid this would be to have two computers, one where you can use AI for research, and another for the article, and you can never cut and paste between the two… but even that is not enough. You could just look at the fabricated quote and type it out with your hands.

It’s only going to get worse as AI is inserted everywhere. It’s in freaking notepad.

The only way to avoid this is to just not use AI, which is pretty easy; we all managed to do it before a couple of years ago. Many of us still manage to do it today.

 

[chickenboo]

In other words, you're going to ragequit because of what you read, and then keep on reading.

That's my prerogative yes.

 

[hob_g]

I come here for real journalism not AI slop. I specifically come because Ars seems to put journalistic standards and integrity forward. Has the author's previous work been reviewed? If there were no editorial safeguards in place to catch this, and it took the misquoted person coming to the comments, is there any way to know if it's an isolated incident? What kind of a journalist uses AI to write an article about the abuses of AI and then doesn't even proofread it?

 

[coonwhiz]

Has the author's previous work been reviewed? If there were no editorial safeguards in place to catch this, and it took the misquoted person coming to the comments, is there any way to know if it's an isolated incident?

Did you even read the statement? It literally says "We have reviewed recent work and have not identified additional issues. At this time, this appears to be an isolated incident."

 

[PhysicsGuy]

I am looking forward to the cloudflare-like post mortem.

I am not threatening to not resubscribe, but a full “we ate shit, here’s how” accounting will make me resubscribe enthusiastically.

 

[Resistance]

They were specifically requested not to comment until Ars management had said their piece in public. Presumably, once the retraction note was posted, they were clear to make their own statements public.

I don't think it's unreasonable for Benj to make a public post taking responsibility for the screwup, especially if it's after management has said, "right, we've said our piece, go nuts". It's the responsible thing to do, especially since there are two potential culprits.

Now it's for Ars management to look at what Benj (in particular) has to say - both in public and in private - and decide what to do, and how much to make public about the process. My view is that more transparency is better, with the caveat that there are good reasons to keep some of the details behind closed doors, sometimes.

We only know that Kyle was asked not to comment.

 

[The Ambient One]

I became aware of this incident yesterday, when the retracted piece appeared in the RSS feed, but the link to it 404d. I spent a little while pursuing the matter and found Mr. Shambaugh's blog posts. This afternoon, a 404 Media piece alerted me to this retraction notice.

Like many commenting here, I found this incident disturbing. Ars Technica has been valuable to me and many other people for many years, due to the consistently high quality of its research, writing, and editing. I would hate to see it go the way of so many other publications ruined by greed and stupidity (Deadspin, The Washington Post, etc.).

Unfortunately, I also don't find this incident very surprising. More and more foolish people have drunk the "AI" Kool-Aid. Apparently, at least one of the authors of the retracted piece is among them. Like Kool-Aid, what currently passes for AI is sweet, in a cheap, artificially-flavored way, but like the Kool-Aid served at Jonestown, it's poisonous.

I appreciate Mr. Fisher's stand here, and I hope he'll maintain it. However, I expect he'll be under increasing pressure not to. His corporate masters at Condé Nast already decided it's fine for "AI" to ingest Ars Technica, including these comments (which is why I rarely comment here anymore); see:

https://meincmagazine.com/information-technology/2024/08/openai-signs-ai-deal-with-conde-nast/

Granted, facilitating the production of "AI" slop isn't the same as publishing "AI" slop themselves. However, I think it's reasonable to construe it as a step down a slippery slope. (Sloppery slope?)

Regarding this particular story, I hope Mr. Fisher et al. will spend the time to find out the full truth. The 404 Media piece says, "Like many of the stories coming out of the current frenzy around AI agents, it sounded extraordinary, but given the information that was available online, there’s no way of knowing if MJ Rathbun is actually an AI agent acting autonomously, if it actually wrote a 'hit piece,' or if it's just a human pretending to be an AI." To a skilled investigative journalist, that might sound like a challenge worth meeting.

 

[hob_g]

Did you even read the statement? It literally says "We have reviewed recent work and have not identified additional issues. At this time, this appears to be an isolated incident."

I mean with other authors, not just this one. Failure of the editorial process.

 

[josephhansen]

Considering that the authors reads the comments it's disappointing that they haven't taken the message conveyed to heart.

The author lied to his editor and to us. He broke the rules of his workplace and damaged the public image and credibility of his employee in the process. I hope he reads these comments and recognizes how serious this is. I hope he sees that many of us are considering unsubscribing because of his dishonesty. I hope the management also read these comments and use them in the sit-down chat with Benji as evidence of why his employment has been ended- costing the company money through fraud and dishonesty. These are grounds for termination at any company

 

[thunderbird32]

That's my prerogative yes.

Right, Ars are not obligated to our subscriber money. I'm not going to quit reading, but I've also set my subscription to non-renew.

 

[Resistance]

I am starting to be less pissed off at Ben for putting fabricated quotes into his article and more pissed off at Ars for making him choose between working while having COVID and having less PTO for recreation. (I know that Ars is not the only company that does this, but it is a stupid policy that results in incidents like this.)

A policy like this incentivizes employees to work sick, which means poor quality work is incentivized. Poor quality work should not be incentivized.

It is that fucking simple.