Decision-making AI model Jev has found another job: helping a Chrome extension identify and flag low-value writing on LinkedIn.
AI enthusiast, author, and business consultant Tom Frazier released Slop Mop as a Chrome extension on Tuesday, as well as publishing its code on GitHub for those who want to tinker with it on their own. Frazier said it’s completely free, doesn’t require any signup, and doesn’t just filter out AI slop, either - it catches garbage written by humans too, actually.
The extension relies on nine signs drawn from research by SEO firm Graphite Growth into patterns that appear disproportionately often in AI-generated writing – a particularly relevant issue on LinkedIn. AI-detection outfit Pangram classified 41 percent of the LinkedIn posts longer than 250 words in its sample as fully AI-generated, which doesn't include posts human authors use AI to improve. The problem has become so bad that LinkedIn recently added a button to the platform that allows users to report posts as "Seems Like AI Slop."
Slop Mop distills Graphite's much broader research into nine tells: Posts stop users from scrolling by including one line of text and a gap that requires viewers to extend the post, use lots of hype words, contain no-catch promises, are written with flowery language, contain a suspiciously neat lesson, are engagement bait, rely on empty praise, define terms by contrast, and are worded stiffly.
Those nine tells are what Slop Mop uses to identify potentially low-value writing, and two more questions are posed to prevent quality content from getting caught up in a Jevian : Whether what's written sounds like a particular person and whether it contains anything useful to readers. Both of those questions lower the score, Frazier explains on Slop Mop's website. A 12th question asks how likely it is that a model drafted the post, but that answer can only lower the slop score for writing that otherwise reads as human-written and cannot cause a post to be flagged on its own.
None of those factors alone will distinguish AI from human writing, we note - as does Frazier. He describes Slop Mop less as an AI detector and more as a filter for eliminating internet garbage, no matter who or what writes it.
“Slop Mop flags poor human content equally to poor AI content,” Frazier told . “I feel that is the right decision because AI content can also be good and human content can also be bad.”
The key question behind Slop Mop, Frazier writes in an FAQ on the extension’s website, is whether a post is worth your attention.
Slop Mop will either flag or hide posts that meet its threshold for slop, whether AI- or human-written, meaning that they won’t actually be removed from a LinkedIn feed and can still be reported using the “Seems Like AI Slop” button.
By Jev, he might be on to something!
Rather than being frustrated by LinkedIn slop and looking for a way to stop seeing it, Frazier told us his primary motivation in building Slop Mop was to experiment with Jev, the new probabilistic decision AI model from TypeSafe that was released last week.
Jev isn’t a chatbot and can’t be interacted with like one. Instead, it is designed to make probabilistic decisions for other software systems and is accessed via an API. It’s quickly become a hit on the internet, with developers finding lots of new applications for the cheap, fast model.
Slop Mop, in other words, is another attempt by another developer to run Jev through its paces, and it's been open-sourced so others can do the same if they wish, and they might want to: The entire thing is running on Frazier’s Jev account, and thus is running requests through his API key, so if it proves to be a hit, he might be looking at some sizeable bills.
Frazier has implemented daily post caps for users to prevent them from running up his bill. Up to 250 posts can be analyzed per installation per day, which he said well exceeds how most use LinkedIn.
“From my early install base I am seeing a cost per user of ~$0.005 per day,” Frazier told us. “If this project goes viral, I might re-evaluate the free service portion of it but for now I am happy to eat those costs to understand tuning parameters and build the dataset.”
If enough people find it useful and it doesn’t break the bank, Frazier also said he may extend its capabilities beyond LinkedIn to other platforms, but for now he’s treating it purely as a Jev experiment. Speaking of which, he’s pleased with the early results, saying that the project shows off Jev’s capabilities as “a new class of AI infrastructure that is here to stay.”
“The context aware decision making with a confidence score will do more for enterprise workflows than LLMs have to date,” Frazier predicts.
You can grab your own copy of Slop Mop on GitHub, or install it from the Chrome Web Store. ®