Is LinuxJournal AI Slop now?
Posted by miversen33@reddit | linux | View on Reddit | 40 comments
Quick intro, this article popped up in my google recommendations this morning
https://www.linuxjournal.com/content/arch-linux-breaks-new-ground-official-rust-init-system-support-arrives
It is a 404 now, but the wayback machine grabbed it before they deleted it
https://web.archive.org/web/20250618001301/https://www.linuxjournal.com/content/arch-linux-breaks-new-ground-official-rust-init-system-support-arrives
Its a complete (and relatively well written) article about a new system init tool called rye-init
(spoiler alert, it doesn't exist). I will not pretend to be the arbiter of AI slop but when I was reading the article, it didn't feel like it was AI generated.
Anyway, the entire premise is bullshit, the project doesn't exist, Arch has announced no such thing, etc etc.
Whoever George Whitaker
is, they are the individual that submitted this article.
So my question, is LinuxJournal AI slop?
AyimaPetalFlower@reddit
obviously it's ai yes
miversen33@reddit (OP)
See that's the problem though. It wasn't obvious. At least not to me. I even started discussing it with a friend and he said the link was a 404. I click it and sure as shit its gone. Then I did a bit more research and turns out the entire thing was fake.
But the article itself didn't feel or read as if it came from AI
AyimaPetalFlower@reddit
trowgundam@reddit
I've taken to calling it the Uncanny Valley in Written Form. It looks real, but something is just "off" about it, and the closer you look the more wrong it looks.
AyimaPetalFlower@reddit
Keep in mind, these current AIs are absolutely not designed to be persuasively human in writing, they're trained to score as high as possible on benchmarks and provide information in the most helpful way possible. If their goal was making writing that was indistinguishable from human writing we'd have that by now already.
trowgundam@reddit
The problem is they are too "perfect." As humans we are, more often than not, defined by our flaws. AI, as it is now, is like a great average of its entire training data. When you average things out a lot of the imperfections become muted or just hard to see, and that is something that will be very difficult to remove from this current batches of LLMs. Its something that arises from the very way they function. They'd have to purposefully introduce those flaws, and doing so in a natural matter will be hard to do convincingly.
AyimaPetalFlower@reddit
They're doing RLHF that biases the ai towards responses that people "like" which means sycophantic responses and flowery language.
They also still have a lot of early chatgpt responses in the training that were RLHF'd by badly paid nigerians and well paid americans who don't give a fuck (me) who poisoned the dataset.
The problem is more that if they tried to make it write more human-like the AI would be worse at doing it's job and the AI companies don't necessarily want to make an AI that would be deceptive. There's thousands of "As an AI..." prompts in the training data.
Each model also has its own easily distinguishable writing style and flaws and are generally predictable
IAmTheOneWhoClicks@reddit
Also the "Why Rust?" and "What's next?". Chat gpt often uses wording like that when it pretends to be human.
CrazyKilla15@reddit
the thing about "pretending to be human" is that its doing things it saw human text do a lot in training. because real actual humans do those things, write that way.
Thats what makes the slop so dangerous. There are no reliable indicators, any that may exist will eventually be fixed.
For example "high-end" image generation hasnt had major issues with hands, faces, or text in ages. The big "tell" recently is the yellow tint on everything, but thats just a temporary issue with one specific model/tool people are commonly using(i assume its among the cheapest right now and thats why everyone is using it), and theres plenty of others without that issue, where its much more difficult to tell at a glance, or may not be possible to tell at a glance.
Already legitimate artists get attacked for "AI use" just because they have the kind of glossy digital style that AI slop often imitates, because thats what it does, imitates things. imitates styles, commonly done by real humans, and now people are conflating the imitation with the real thing and calling it all AI.
TheOneTrueTrench@reddit
What sets it apart from human writing is precisely how it just reads like the average of every human writer. It feels like asking a hundred people for a random numbers, and instead of a modal distribution, where certain numbers show up far more often than others, there's a nice perfect bell curve around 37, because the training data says "37 is the most common random number"
daemonpenguin@reddit
To me it seemed pretty obvious the article was fake. Whether that means all of LinuxJournal is AI-generated slop.... probably not. Lots of bad articles get posted in a rush, whether AI-generated or not. People make mistakes, that doesn't mean it is a trend.
DriNeo@reddit
The last human content on the internet will be Phoronix comments.
jinks@reddit
Which will be used to train early prototypes of AM.
NoTime_SwordIsEnough@reddit
God dammit, you win. I burst out laughing after reading your comment.
(Haven't played or thought about that game in forever, too.)
regreddit@reddit
We're done then.
udsh@reddit
Skimming through the other recent articles on the site shows that they also smell like AI slop. Really unfortunate what happened to LJ, I know they went through a rough patch a few years back and had to lay off a bunch of staff before getting acquired by Slashdot, but I didn't realize it turned into this.
Journalism is already struggling, journalism focusing on free software is an even smaller niche, especially given that the most common methods of monetization like advertising are less effective, or less likely to be considered by an outlet that's trying to look ethical and professional.
LWN still puts out exceptional pieces from time to time. I think they're the last bulwark of free software journalism. Go buy a LWN subscription and help keep them alive.
kombiwombi@reddit
LWN are superb. The series of Grumpy Editor articles where Jon has a problem and goes looking for FOSS software addressing that field are particularly wonderful writing about technology.
They very much could do with your subscription.
TheCrustyCurmudgeon@reddit
They may be superb, but $9/month is a lot for a web-mag sub. Hard pass for me.
is_this_temporary@reddit
Also, even the subscriber only articles still follow Free Software principles.
If you are a subscriber, you can share the article with anyone via subscriber link (and the recipient can as well).
CrazyKilla15@reddit
Arent all the subscriber only articles merely time locked, too? Just wait and you can always read it
unknown_lamer@reddit
If that's the case... it's 100% AI slop.
FYI Slashdot was purchased by borderline spammers about 11 years ago. Entire editorial staff was fired with zero notice except for the chief editor who got a month to train his replacements. Been complete shit since then (not that it was doing great under the previous owners who I'm pretty sure were tricked into buying it with juked numbers and a heavily inflated "goodwill" valuation).
bradmont@reddit
Dang, really? I haven't read slashdot in... probably about 11 years, but I'm super disappointed to hear this.
ang-p@reddit
George has quite a few other AI articles under "his" belt.
berickphilip@reddit
A ton of websites are like that now; internet dark ages are imminent.
syklemil@reddit
Remains to see how long it'll last though. LLMs also need training data, and if you train them on LLM slop they get their own variant of mad cow disease, "model collapse". Pair that with the prohibitive cost of LLM training and no sustainable financing … I guess one possible outcome is that LLM training grinds to a halt, while the current models, obviously good enough for spammers, remain in use somehow (if they can get stable finances for that at least).
__Myrin__@reddit
Welp time to add it to the block list
FryBoyter@reddit
Because a single article was published that was presumably created with a chat bot? Isn't that a bit of an exaggeration?
Irrespective of this, I know of several cases where users have been accused of creating a post with such a tool even though this was not the case.
In short, I think it's stupid to condemn someone completely because of one incident.
TheOneTrueTrench@reddit
Plagiarism consistently causes inaccurate results, and AI slop is the result of plagiarists so lazy they won't even do it themselves or bother to confirm even a single detail... including literally the entire basis of the article.
At least James Somerton took the time to read the shit he stole, this person can't even be bothered to do that.
turdas@reddit
It's never just a single AI slop article.
vetgirig@reddit
You can find "rye init" here: https://rye.astral.sh/guide/commands/init/
But it's something very different ...
Primary_Bad_3778@reddit
dude reads google's recommendations and complains about ai slop, too funny. you're smelling different ends of the same turd.
inbetween-genders@reddit
TIL Linux Journal from my wee lad days is still around 😂
DolitehGreat@reddit
Well, seems like some sort of zombie of it is still around.
ApplicationMaximum84@reddit
I used to buy Linux Format in the UK, they just released their final magazine on their 25th anniversary.
--Apk--@reddit
They deleted the article so clearly not. As you said, it was well written so it obviously went under the radar of their moderation obviously.
-Sa-Kage-@reddit
But as it was apparently the only news site, that reported so, they didn't copy it from somewhere, but someone there had to use AI for news
daemonpenguin@reddit
I saw this posted on here earlier in the day, read the post, determined it was almost certainly fake, and ignored it rather than posting a referencing article/link.
To me the article seemed to be pretty obviously fake, but maybe it's just because I spend more time in/around the Linux community.
liquidpele@reddit
Online “journalism” was already shit, now with AI it’ll be just click spam all day every day. Might as well start adding any news sounding things to your spam filters.
RepentantSororitas@reddit
I find most linux news sources to be pretty "sloppy".
There is only so many times you can compare this ubuntu based distro to base ubuntu
w453y@reddit
Even I fell into the trap and made a post about it, which I later deleted.
https://www.reddit.com/myfakp2