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

Suriraj built an open-source AI 'slop detector' Chrome extension using Juny AI to identify low-information-density content on the internet, defining slop as text lacking verifiable claims and high in predictability.

• An AI-powered 'Slop Shield' Chrome extension was created to detect low-information-density content.

• Slop is defined by a lack of verifiable claims and high predictability, quantified by a 'slop score'.

• The detector uses a local, value-driven AI (Quen) for robust, private content evaluation against internet 'slop'.

A Chrome extension called 'Slop Shield' was developed to detect 'slop,' defined as low-information-density content characterized by a lack of verifiable claims and high predictability, rather than conveying meaning. This tool provides a 'slop score' to assess content trustworthiness, serving as a personal defense against the internet's increasing disloification. The project highlights the necessity for individuals to build and use their own AIs to effectively parse reality and verify claims online.

Defining 'Slop'

• 00:00:56 Slop is defined as low information density, measured by the ratio of verifiable claims to content length. It is not merely poor writing but a quantifiable absence of signal, designed to occupy space instead of conveying meaning. Slop is characterized by high predictability and a deficit of verifiable claims, necessitating a system to convert vague prose into structured, trust-scored assertions.

Building the First Detector

• 00:02:09 The initial 'Slop Shield' Chrome extension was built in approximately 4 minutes and 30 seconds using Juny AI, an autonomous agent in JetBrains IDE. This first version employed heuristics to detect slop, identifying features like few concrete entities, low lexical variety, repeated phrases, and flat sentence cadence. However, it proved insufficient for accurately detecting slop in certain research papers, indicating a need for improvement beyond rule-based detection.

Improving with Local AI Values

• 00:04:06 To enhance accuracy, the detector was improved by integrating a local AI, a 7-billion parameter model named Quen, which runs privately on about 4 GB of VRAM using Olama. This AI incorporates 'constitutional' values such as falsifiability—the ability for claims to be checked against external reality—and epistemic modesty, acknowledging uncertainty rather than hallucinating confidence. Encoding these values through a system prompt allows the AI to use reasoning and judgment instead of rigid heuristics, making the detection more robust.

Vision for Personal AI Defense

• 00:06:17 The 'Slop Shield' is presented as a crucial tool for personal defense against the increasing 'disloification' of the internet, with its open-source nature encouraging community involvement. The developer advocates for individuals to build and utilize their own AIs as a primary defense to filter reality, parse claims, and understand web content. This approach empowers users to navigate the internet more effectively and critically evaluate information.