What's going on here?
-
Aaand the full comments thread from the original story: https://arstechnica.com/civis/threads/after-a-routine-code-rejection-an-ai-agent-published-a-hit-piece-on-someone-by-name.1511649/
These were pulled too, but thank you again Wayback:
-
@art_codesmith @tankgrrl @mttaggart they have enough information already to justify immediately yanking the article, so "we'll tell you next week" scans to me as "we need to figure out the PR angle on this" more than "we need to find out what happened".
Maybe their explanation will be a good one, but I'm not holding my breath.
@SnoopJ @art_codesmith @tankgrrl @mttaggart I'm waiting to see what happens in a few days to judge. It's clear the quotes are fake and they acknowledged that, but I can see it taking a few days to identify *how* this happened, and how it made it through editorial. I'm worried though, and I don't know if their answer next week is going to satisfy me.
-
@SnoopJ @art_codesmith @tankgrrl @mttaggart I'm waiting to see what happens in a few days to judge. It's clear the quotes are fake and they acknowledged that, but I can see it taking a few days to identify *how* this happened, and how it made it through editorial. I'm worried though, and I don't know if their answer next week is going to satisfy me.
@misty @art_codesmith @tankgrrl @mttaggart yea, agreed.
-
@RealGene @mttaggart okay fine, if you successfully create massive fashist infrastructure, then yes, you can erase written works on a whim.
It's still a hell of a lot harder than taking down a webpage.
-
@mttaggart this is the weirdest story. Here is a link to SCOTT SHAMBAUGH’s blog explaining the whole thing with an update about the additional AI generated reporting. https://web.archive.org/web/20260214062635/https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/
@IcooIey @mttaggart wild thing indeed. Gatekeeping is in fact not a bad thing at all, and it worked long before AI. Open source communities have their right to place guardrails and policies, and they are not obliged to accept any PR. If they say "place a comment every second line" you should comply. If they say "that is good entry level issue, don't fix it with automated tools" - don't fix it, and don't complain if you do and they reject you, AI or not.
-
Putting this here so all can see it. Ars forum thread where the pull and investigation are mentioned: https://arstechnica.com/civis/threads/journalistic-standards.1511650/
@mttaggart same Ars that let this article hit the front page years back?
-
It should be cut & dry.
Restate your policy on AI generated content.
State you are doing an investigation.
Then move on.This particular wording leaves room for excuses for the continued use of AI summarizers/ writing assistants.
I'm not saying to actually do anything over the weekend.
I'm aware of Condé Nas's internal policies when an article gets pulled from Ars, there's a formal investigation to avoid slandering the writer & chain of trust their work passed through.
@rusty__shackleford @Gaelan sure they could answer the second of your questions right away. It read to me like you were saying they should answer all three right away, which I think isn't realistic. If that's not what you were getting at, fair enough - I just misread you.
-
These were pulled too, but thank you again Wayback:
-
@Gaelan
Strategic ambiguity is what this *appears* to be, it's the calculated vague speak that allows for plausible deniability that gets me.Also, news cycles: Friday news dumps allow stories to die over the weekend. Pushing the response back isn't just about the holiday, it’s about waiting for the news cycle. They're betting that by Tuesday, the "outrage" will have lost its momentum, making vague statements easier to swallow.
I know they have internal processes for this, but not a good look.
@rusty__shackleford to be fair, this is a piece with a dual byline. Unless either Benj or Kyle fesses up directly, it really will require some serious investigation to even try to figure out which one did it.
Then the one that DIDN'T do it, but also didn't catch it, gets to explain why that shit went out with their name on it.
As much as I want to hear that this was resolved firmly, decisively, and without waffling, a couple of business days really is not entirely unreasonable here.
-
UPDATE: They pulled the story, but I had it up and had SingleFile in my browser, so: https://mttaggart.neocities.org/ars-whoopsie
@mttaggart oh man, i wish i could see the comments
-
@mttaggart oh man, i wish i could see the comments
-
What's going on here? The matplotlib maintainer this story is about correctly notes that all the quotes from his post in the article are made up.
UPDATE: Link was pulled; see below.
@mttaggart is an AI agent responsible for the one down vote in that screenshot?
-
@mttaggart @theorangetheme I'm genuinely confused about how this was allowed to happen. I tend to assume Ars has better editorial processes than some of the places I've worked, and both writers have long-term specialisations. My most charitable explanation is that someone created a version that they though would be funny and that was accidentally published. Very curious to see what their investigation yields.
Seems like it very much was the consequence of writers using AI ..!
Edit: or potentially an editor, would be good if they specified which — and either way, it slipped through the editorial process.
-
Seems like it very much was the consequence of writers using AI ..!
Edit: or potentially an editor, would be good if they specified which — and either way, it slipped through the editorial process.
@aliide @mttaggart looks like an adequate response by the editor
-
These were pulled too, but thank you again Wayback:
The final chapter? The statement from Ars:
On Friday afternoon, Ars Technica published an article containing fabricated quotations generated by an AI tool and attributed to a source who did not say them. That is a serious failure of our standards. Direct quotations must always reflect what a source actually said.
-
The final chapter? The statement from Ars:
On Friday afternoon, Ars Technica published an article containing fabricated quotations generated by an AI tool and attributed to a source who did not say them. That is a serious failure of our standards. Direct quotations must always reflect what a source actually said.
Good. No quibbling, just taking responsibility with transparency.
-
The final chapter? The statement from Ars:
On Friday afternoon, Ars Technica published an article containing fabricated quotations generated by an AI tool and attributed to a source who did not say them. That is a serious failure of our standards. Direct quotations must always reflect what a source actually said.
@mttaggart Was the article about how good AI is?
-
The final chapter? The statement from Ars:
On Friday afternoon, Ars Technica published an article containing fabricated quotations generated by an AI tool and attributed to a source who did not say them. That is a serious failure of our standards. Direct quotations must always reflect what a source actually said.
@mttaggart Not "We are sorry for publishing AI slop", just "the quotes should have been verified"? (Edit: it was pointed out to me that if I read the article, the appology was actually for an AI article, not just the quotations. Thanks @mttaggart )
-
@mttaggart Not "We are sorry for publishing AI slop", just "the quotes should have been verified"? (Edit: it was pointed out to me that if I read the article, the appology was actually for an AI article, not just the quotations. Thanks @mttaggart )
@Retreival9096 There's an apology in the linked post.
-
The final chapter? The statement from Ars:
On Friday afternoon, Ars Technica published an article containing fabricated quotations generated by an AI tool and attributed to a source who did not say them. That is a serious failure of our standards. Direct quotations must always reflect what a source actually said.
Not quite the final chapter! Benj Edwards has taken responsiblity in this Bluesky post:
https://bsky.app/profile/benjedwards.com/post/3mewgow6ch22p
For those who won't head over there, a summary:
First, this happened while sick with COVID. Second, Edwards claims this was a new experiment using Claude Code to extract source material. Claude refused to process the blog post (because Shambaugh mentions harassment). Edwards then took the blog post text and pasted it into ChatGPT, which evidently is the source of the fictitious quotes. Edwards takes full responsibility and apologizes, recognizing the irony of an AI reporter falling prey to this kind of mistake.
