The Fall of Slop (Even Microsoft Admits There's a Problem)
Slopwatch is no more because the issue has been contained somewhat.
I am still keeping abreast of about half a dozen fake "Linux" sites (with "linux" in their domain name) which spew out LLM slop. Among them: FOSSLinux, FOSSPost, Linuxiac, LinuxSecurity, LinuxJournal, and LinuxTeck.
There are a few more, but they are less actively updated. Recently, UbuntuPit had second thoughts and apparently deleted every single 'article' that it had published using LLMs. That's good. At least there's some late realisation, maybe remorse.
LLMs won't replace journalism. LLMs will discredit sites/outlets. That's all they can do. Once caught slopping, there will be scandals. When there are scandals there will be disputes and a downfall. One former news site, which did fine investigative journalism, decided to misuse the names of people who used to write there to make LLM slop seem like real articles. They are striking back. This is a catastrophe.
"Just fifth of journalists now say they don't use [slop]," the Press Gazette says this week, "down from a third a year ago".
Correction.
If they use slop, then they are not journalists, they're cheaters who pretend to be journalists. They're 'hacks'.
Utterly terrible if true, but we should recognise it for what it is. It's not about journalists but about sites that went rogue and are probably forced to fake 'journalism' - using slop imposed by bosses who are not journalists.
As an associate put it last night, slop means "factual errors / disinformation" (not journalism).
LLMs are not for news; they were not meant for news. They can fake the capability, but they cannot do the bare basics right and they cannot really study stories, they just parrot strings of words and often get the overall message wrong (or mix together totally unrelated things, even from decades ago/apart). For instance, a story about Microsoft layoffs in 2024 might fuse together information from 2014 (about Microsoft destroying Nokia and then having ~14,000 layoffs). The LLMs lack a grasp of time and context.
As an associate put it, "news requires accuracy and any connection between facts and LLM output is purely coincidental. It is the very nature of the technology and no amount of money or electricity will improve the accuracy. The companies selling RAM, SSDs, CPUs, GPUs, and network cards have convinced chumps that adding more of the aforementioned will magically make LLMs useful."
It's not about how much energy is spent on inherently flawed processes, as we saw before. LLMs won't improve and even Microsoft, its biggest and loudest pusher, now admits they're for "entertainment purposes only" and are generally too expensive. To quote a new report:
Major tech companies are considering refining their approaches to [Slop], as rising token costs and increased token demand from [Slop] agents make the costs harder to justify, with limited return on the investment.
This specifically names Microsoft. If Microsoft admits that slop is too expensive and is for "entertainment purposes" because it cannot be relied upon, why would anyone other than the pushers and profiteers still insist that slop bears potential? █