Using AI here is akin to mishandling the chain of evidence. Even on the AI beat for a tech outlet, using these tools at any point in the process is a serious breach of journalistic integrity. As someone who has been reading and following you for years, I'm disappointed.

This seems like a totally plausible sequence of events, and I was glad to hear Ars did a review to determine this was a one-off. It is refreshing to see someone taking full responsibility for their error; it's become rare these days. Feel better soon!

Thanks I deeply appreciate it

I think you need to further define what generated means. You state none of the articles are AI generated…but you used AI to generate a list of quotes (whether they were real or not is immaterial).

wouldn't you have needed to track down and double check the "verbatim" things Claude pulled out of the source? did you think that would be faster? which of these tools has been a routine part of your workflow? are they routine parts of coworker's workflows? is the risk of errors like this worth it?

Get well, dude. And please for the sake of whatever might be left of your reputation, absolutely swear off the AI "assist". It's no better than a toy, and is immensely destructive in the professional world, as you have just experienced. Honor is earned by learning from error. Good fortunes.

Hopefully you learn and grow from this experience and come to understand that there is no magic inside the stochastic parrot.

Also, everything generative AI does is generative - it's in the name. It cannot 'extract' content. It can only repeat it again, which is a generative task that it can 'hallucinate' on. Nothing just passes straight through, every token from the model comes from the model, only steered by the prompt.

yeah definitely dont believe this was isolated. you finally got caught because scott's blog had anti-crawler systems that stopped chatgpt from grabbing content.

I am sorry you've been sick. No one died from this error and I look forward to your future articles, just as I've enjoyed your insightful and always highly topically educated pieces for many years. Thanks for explaining.

Thank you for being clear about what happened, Benj. Owning the error is the right way to approach it. A great lesson on the risks of using AI.

Feel better.

> “During the process, I decided to try an experimental Claude Code-based Al tool to help me extract relevant verbatim source material.” Sorry but can you explain why you needed an agent tool to extract material from a blog post that you ostensibly read yourself?

Sorry that this happened to you Benj, I hope you feel better. It's hard to keep the plates spinning all the time especially when sick.

Damn, man. Rough week. I'd stop checking your email and socials for awhile until the crazies move on to the next thing. Hope you and your family get well soon.

Thanks for being honest and explaining, Benj. I have tried AI a few times recently to see if it could help find quotes in large corpuses for my Break Into Chat research. And it fabricated quotes multiple times. So infuriating.

Sorry to hear you're sick man. Rest up and get well soon! Sorry to hear about the mess too. None of us are perfect and owning what happened really shows the strength of your character. You've always been good people Benj; don't let this get you down. I will still always read a new Benj piece.

If Ars policy is that you can't publish work created by AI unless noted, and you're using AI in any way in your work process and not disclosing that in each article, then you're in violation of the policy whether the material is correct or not. This is not ok.

does Ars Technica not employ fact checkers? How does an article go out without someone else verifying the quote?

Accidents happen, you're alright in my book. For whatever that's worth. I'll keep reading. 🫡

I feel bad that you're going to have to deal with the witch hunters now.

We all sick my dude, the flu is brutal this year! Hope you feel better soon!

As I’ve observed as a 20+ yr Ars subscriber you are a great journo and an even better guy. That a mindful practitioner of his craft is so vulnerable to the frailties of AI is a lesson for us all to heed. Get well soon, looking forward to seeing you back on the beat at Ars.

In the name of transparency and also demonstrating a good faith effort in reestablishing trust, I think it's incumbent on you to actually name the tool and not just obfuscate behind an "experimental Claude-based tool". I mean this isn't a confidential informant, so why the need for secrecy?

Really appreciate the apology and explanation, Benj. Hope your health recovers completely and quickly.

Wow. Just Wow. The "Senior AI Report" can't take some time off when he is "sick in bed with a high fever". What a shitty place to work.

That’s working in the US, yes.

This is a design failure, the fact that user is expected to apologize for a bad faith output is wild. Blame the people saying AI is trustworthy and stop letting them evade critical analysis.

We love you, Benj! Shit happens. It’ll be in the rearview soon enough!