The Internet Is 50% Fake. I Built a Detector.

Why It Matters

A freely available, locally‑run AI slop detector empowers individuals and organizations to filter low‑value content, strengthening information hygiene and reducing the spread of misinformation.

Key Takeaways

• AI slop detector flags half of web content as low-value.
• Built with Juny AI in IntelliJ, deployed as Chrome extension.
• Uses density formula: verifiable claims per length to score slop.
• Transitioned from heuristic rules to local 7B model for accuracy.
• Open‑source, free tool encourages personal AI defenses against misinformation.

Summary

Suriraj demonstrates an AI‑driven "slop" detector that labels roughly 50% of the Internet as low‑information content, showcasing a Chrome extension that warns users in real time. He defines slop as low information density—verifiable claims divided by text length—and frames the tool as a "slop shield" that quantifies trustworthiness on a 0‑100 scale. The video walks through building the detector with Juny AI inside JetBrains IntelliJ, using voice prompts to generate a Chrome extension in under five minutes. Initial versions relied on heuristic cues such as lexical variety and repeated n‑grams, but performance lagged on research papers, prompting a shift to a local 7‑billion‑parameter model (Quen) run via the Ollama inference engine. The system prompt encodes values like falsifiability and epistemic modesty, mirroring Anthropic’s recent constitutional AI approach. Suriraj highlights concrete examples: the detector flags an entire Wikipedia page, a YouTube video, and a research paper as slop, and provides detailed breakdowns of why each source scores low. He credits Juny’s planning capabilities for the rapid development cycle and emphasizes the open‑source nature of the code, inviting others to customize and improve the model. The broader implication is a democratized defense against information overload and misinformation. By making a free, locally‑run AI tool that evaluates claim density, users can reclaim agency over the web’s signal‑to‑noise ratio, potentially reshaping content curation across industries that rely on high‑quality data.

Original Description

The internet is drowning in low-information "slop." I built an AI-powered Chrome Extension to fight back.

🛡️ Get the Code (free): https://jb.gg/JunieAICoding

🔧 Built with Junie AI: https://github.com/llSourcell/SlopShield

"Slop" isn't just bad writing, it's a measurable lack of signal. Density = Verifiable Claims / Length. In this video, I build the Slop Shield: a Chrome Extension that uses a local LLM (Qwen 7B via Ollama) to score any webpage's truthfulness. No data leaves your machine.

The Stack:

• Agent: Junie AI (IntelliJ IDE)

• Backend: Python + Flask

• AI Engine: Ollama (Qwen 7B, local)

• Frontend: Chrome Extension (Manifest V3)

Timestamps:

0:00 - Slop Demo

0:54 - Defining Information Density

2:12 - Building with Junie AI

3:33 - V1: Heuristic Failure

4:03 - V2: Local AI Judge

5:09 - Constitutional System Prompt

6:19 - Why You Need Personal AI Defense

📬 Business inquiries: [e-mail]

📲 Follow

X: https://x.com/sirajraval

Instagram: https://instagram.com/sirajraval

LinkedIn: https://linkedin.com/in/sirajraval

#AI #Coding #JunieAI #DeadInternetTheory #LocalLLM #Ollama #ChromeExtension #aislopdiy