A person named hereticpleb has spent an hour and a half building a machine that does one thing: it finds LinkedIn posts about useless computer vision projects. The result, he says, is a detector that proves how easy it is to make something look impressive on the platform. The joke, he admits, is that the detector itself is the punchline.
The project is called “LinkedIn Larpmaxxing,” and his argument is simple: the platform rewards noise over skill, and the people who survive there are not the ones who get better. They are the ones who look interesting.
The Feed Problem
Hereticpleb opens by describing what he calls “the most unbearable stuff you would see.” His LinkedIn feed is full of posts from people whose relatives have been hit by trucks, and who use those tragedies as lessons about business management. The posts carry a tone of being “visionary” and “forward-thinking” and “professional,” he writes, adding that it is “SO tiring.”
Then there are the projects. Every time he opens the app, he gets hit by videos of some guy waving his hand around while a camera follows him. These are Computer Vision Projects, he says, and they are everywhere. “It is just every other post some guy waving his hand around doing NONSENSE,” he writes.
The Pothole Detector
One example stands out. He shows an image of a pothole-detection system, a project that promises to monitor roads for damage. His response is blunt: “What is the point?” He asks. “You know what else could detect potholes? Eyes. They are very, very good at detecting those.”
He pushes further. The system is not connected to anything. It does not send data anywhere. If it were, that would be cute. But it is not. “BUT IT’S NOT,” he writes.
The bigger idea behind the project, he acknowledges, is that AI can be used for smart infrastructure monitoring. Road conditions could be assessed more efficiently, and maintenance teams could make better data-driven decisions. That is the pitch. The execution, he says, is trivial.
Training the Model
To prove his point, hereticpleb decided to build a detector for the exact kind of posts he hates. He calls it the “🚀 Excited to announce I made a YOLO project” detector. The name is a reference to the YOLO object detection framework, which appears in the code he shares.
He learned everything from scratch. He has never done this before. He compares the experience to cooking ramen after never having cooked before. “Can’t critique cooking without ever cooking ramen,” he writes.
The process started with data. He needed to train a model on images of LinkedIn posts that bragged about YOLO projects. He wrote a script to collect the images automatically. It scrolls through his feed, takes screenshots, and saves them to a folder.
The Annotation Step
Next, he created a Roboflow account, uploaded his screenshots, and annotated them. He drew boxes around the parts of the posts he wanted the model to recognize. It was mechanical, he says, and it took about 20 minutes.
Once the images were labeled, he exported them in the YOLOv8 format and downloaded the dataset. Then he trained the model.
The Code Behind It
The training code is short. He uses the Ultralytics library, which provides a simple interface for working with YOLO models. He loads a pre-trained model, configures the training parameters, and runs the experiment.
Here is the code he shared:
python
from ultralytics import YOLO
model = YOLO('yolov8n.pt')
results = model.train(
data='dataset/data.yaml',
epochs=50,
imgsz=320,
name='slop_detector'
)
He adds that the training took 30 minutes on his potato PC. Once the model was ready, he wrote a Python script to run it. The script loads the trained weights, processes an image, and saves the results.
The Detector Is Complete
The whole thing took an hour and a half. Hereticpleb celebrates the achievement with a single exclamation: “YAYYY I got my very own slop detector!!!”
He jokes that he can now add computer vision expert, Python savant, and AI and ML thought leader to his resume. The irony is not lost on him.
The Real Problem
The conclusion is harsh. LinkedIn rewards people for looking interesting, not for getting better. “Nothing on that site rewards you for getting better,” he writes. “It’s a platform built around selling yourself for a job, so what survives isn’t skill; it’s looking interesting.”
He points to profiles where the same guy detects potholes over and over, with zero signs of improvement. The pattern repeats. The skill does not grow.
A Working Machine
Hereticpleb’s detector is a clever piece of satire. It is also a working machine learning project, complete with code and a step-by-step description. The difference between the two is the point.
The projects on LinkedIn are trivial. They are marketing dressed up as productivity. The detector finds posts about YOLO projects, and it does so quickly.
| Comparison | Pothole Detection Project | LinkedIn Slop Detector |
|---|---|---|
| Time to build | Not stated | About an hour and a half |
| Purpose | Detect road damage | Detect LinkedIn posts about useless projects |
| Actual value | Trivial, per hereticpleb | Proves the platform rewards noise |
| Skill required | None shown | Minimal, since he built it himself |
| Result | Repeatedly mocked | Completed successfully |
The comparison is worth holding. The pothole projects he mocked are trivial. The YOLO detector he built took an hour and a half. The skill gap is not between the two projects. It is between the people who show up on LinkedIn and the people who laugh at them.
Hereticpleb’s detector is funny because it is accurate. It is accurate because it is trivial. And it is trivial because LinkedIn rewards the wrong thing.
The platform is built around selling yourself for a job. Skill is not the currency. Looking interesting is. The detector proves it.
See the video the story is built around at vercel.app.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.