Hello everyone, given the recent 3 articles (1, 2, 3), which feature a bunch of random emojis in the headers, wording that hints clearly at the usage of LLMs for its writing and even header images that are all a bit off (squished text, old logos, …), probably due to being generated by some from of generative AI, I would like to start a discussion whether this is the content we want for Fedora Magazine.
I know, that I am not interested in AI generated articles. It certainly helps with the publishing frequency, but I rather had no article that an article that is just put together by an LLM.
So I would like to start the discussion about what is considered good usage of LLMs in the context for Fedora Magazine.
Some question that I would love to see some opinions on:
Do we want LLM generated articles?
Do we want LLM assisted articles?
Should articles disclose if they used AI in the process of their generation?
How can we provide authors with the right resources to write better articles, without these articles feeling cheap?
Disclaimer: I notice that all articles are written by the same author, I want to make clear, that this is not about the author but about the quality of articles we want to produce in general. It’s completely fine, if an author writes an article and translate it with an LLM or just uses an LLM to generate the tedious parts of an article. However, I would expect that the editorial part of the magazine-team than works on refining this article with them before it is published.
No, I’d personally not like to see articles that are generated by AI/LLM with no or little human intervention.
I’m okay with folks using AI to help/assist" with their text as long as it’s limited to stuff like sentence/paragraph structuring and doesn’t overly change the context of what the human has written.
I am not in favour of them. Specifically, I have always felt any articles published to the magazine should be a showpiece of the Fedora community. Also, as someone who has written for the magazine, I researched the topics and tested everything before submitting for publishing.
I think in limited use cases, such as helping with readability, topic continuity, etc …
Definitely without exception.
This is a topic that should be considered, I’m not sure of the current state of the Wordpress tooling we use
The author said that AI was not used. The images are distorted because they were submitted in the wrong size and the editor resized them without taking care to preserve the aspect ratio. You might be seeing AI where none exists.
Those emojis are not “random”. Each one has at least some, at least passing, relevance to the title and topic of the section. In addition, while editing, there were human generated artifacts that were corrected and would have been caught by an AI process.
A novel use of presentation is not evidence of AI influences.
Yet, I’m not the only person thinking it is. This is not only due to the usage of emojis, but also due to the usage of certain structures:
dracut is more than a utility—it’s your boot-time engineer. From creating lightweight images to resolving boot failures, it offers unparalleled flexibility.
This is a very common pattern in LLM generated texts: “Not only emdash but also” followed by a list of 3. This video goes over some of the more common patterns:
There is more, but I also don’t want to go ahead and dissect 3 articles. I’m happy that the question was raised pre-publishing. Maybe it’s one of these cases were people start to write more like LLMs (which is apparently a phenomenon these days).
I think my point stands in general, that it might be good to have enough editorial adjustments that texts don’t even look like generated by AI after they are published. Whether they are or not.
There is an obvious error scattered all throughout that article though – Dracut is the proper name of the software package and it should be capitalized in many places throughout the article (except where it is referring to the command rather than the software package) but it is not. It seems unlikely to me that AI would make such an obvious mistake.
Listing three things is in no way evidence of AI. I myself used a three-element list in a Dracut article I wrote quite some time ago – InitRAMFS, Dracut, and the Dracut Emergency Shell – but I certainly wasn’t using any form of AI.[1]
I was, however, drawing inspiration from C. S. Lewis’ title “The Lion, the Witch and the Wardrobe” ↩︎
I believe you might be enlightened by reading many of the comments on that Youtube video. Items pointed out as indicators are simply good form and actually taught.
I use semi-colons like the thumbnail often; I take minor offense of (likely monetized) YouTube videos probably made with AI assistance trying to call out behaviors used before AI was a thing
I glanced at the 3 articles and they didn’t give AI vibes.
For anyone interested, go to ChatGPT and ask the following (or similar):
Please write an article around the subject of sosreport and data pack layout in Fedora & RHEL. Please also generate a title and use icons in front of the title and the headings.
Of course, one might say that AI had already scraped the Fedora Magazine article, hence the resemblance.
Going further though, for 2 of the articles there are comments about deprecated information. This is again an issue often seen in AI generated text.
They shouldn’t have to because there was no use of AI.
By vetting people more thoroughly so they don’t just sign up for an account and decide to throw an article together using AI. The person in question joined on the 21st and was proposing articles on the 24th. (Edit: Just checked, he joined on the 21st and was proposing articles on the 22nd… who does that? It’s like every time there’s a coding workshop some GitHub repo gets mauled by random individuals from all over the globe pushing fluff to buff their history).
When I signed up, I went through the various Fedora sites to see how things worked and was surprised to find I had access to the Blog AND Magazine site to create an article. And I’ll be honest, my first thought was “WTF” and “have I discovered a bug, should I be here?”.
This example output is definitely AI-generated by approximating what /usr/lib/dracut/modules.d/ might contain, rather than reflecting an actual system. It could have been easily verified, but appears not to have been.
Access to the “Blog” and the Magazine site is part of the process. Why would they be restricted? If you signed the FAS agreement some amount of trust is assumed.
Your first post on discuss is from May 2019. Your first post about even being interested in being a Fedora editor is from Nov 2020. Over a year later. A month later you pushed your first article. You clearly took the time to be interested in and involved in the community before creating a community article for a community blog and/or community magazine.
What the OP articles linked to effectively did was a drive by articling. It’s not bad enough that it’s AI, but it’s most likely been done to buff a CV or college project. Which isn’t bad in and of itself. If the articles worked.
The comments on two of the articles show they didn’t work. There was no way in hell I was going to test an article about dracut from the same author. But thankfully @yuntaz took the risk and we find out that doesn’t work either. So that’s three out of three articles that the original author didn’t check, because he didn’t write the articles, and three out of three articles that were OK’d by editors that didn’t check an authors work.
If this was programming, pushing three AI generated commits without testing would get my ass banned at least temporarily (if not permanently) from pushing to that repository, and those reviewing the commits reprimanded.
Every article that was posted contained errors and every article was allowed through because it wasn’t checked.
Newbies and experts alike browse Fedora magazine. There are going to be people that randomly copy and paste article commands into their terminal. At least in this case nothing catastrophic happened. But I’d prefer AI generated slop not be a risk of messing up people’s installs because the multi-billion dollar companies behind them haven’t pointed them at the latest manpage.
Someone that has the time to at least run the commands the article is proposing in a VM to see if the original author haven’t missed something. If I was submitting an article my work would be checked thrice by myself and I still wouldn’t be offended if an editor went over it once more to do a check if what I was submitting even worked.
Maybe too much trust is assumed? It’s been nearly 2 weeks since our good friend has interacted with the community and there isn’t a single interaction that isn’t about article proposals when he was interacting with the community.
You have no idea about my motivations or anyone elses. I find your fixation on this, especially data mining for details about individuals to be creepy, if not downright invasive. I suggest you find a new hobby.
This conversation should come to and end at this point.
It’s worth mentioning that anyone’s contributions to the project bring their value, regardless of whether the contributors are old or new in the Fedora space, whether they manage a certain area of the project or they just start bringing their input, as long as they all act in Fedora’s interest.
AI generated content brings a new paradigm into every corner of the internet, and it takes time for stakeholders to identify them easily, to evaluate what’s of value and what is not, and to create policies and rules accordingly (see, enumeration of three !).
But at the end of the day, we have three consecutively posted AI-generated articles, that are damaging Fedora‘s image IMO. So I suggest really considering taking them down.
The milder route of giving the author a chance of correcting them themselves could be tried, but I doubt it would help, given that the author might not have the proper knowledge to do that, and given that if the articles were not rewritten from scratch, they would still be identified as AI-generated by the wider audience.
Regardless of AI generation or not. I reckon articles/How-To guides should at the very least be sense checked and tested before being published/made public.
Often people will lurk for some time without joining, then when they do join, come prepared with stuff for the community. In this case it could be something like that. I know that I stewed over writing an article for Fedora Magazine for about a year before I made the plunge. And, point of fact, I lurked on the mailing lists and comm blog long before I even became a Fedora Community Member.
!).