ACAman Chaturvediinaidetector.hashnode.dev·Mar 4 · 9 min readI Built a Dedicated Screenshot Classifier Because My AI Detector Had an Identity CrisisIn my last post, I admitted that 99.2% accuracy was a lie. I fixed the social media compression problem, accuracy dropped to 84.8%, and I called that more honest.Then I started testing screenshots.The00
ACAman Chaturvediinaidetector.hashnode.dev·Feb 27 · 6 min readFrom 99.2% to 84.8%: How I made my AI detector actually better by making it worseIn my last post, I talked about building an AI image detector without deep learning — pure signal processing, an SVM, and math borrowed from research papers. I ended that post admitting the tool strug00
ACAman Chaturvediinaidetector.hashnode.dev·Feb 15 · 4 min readCan We Detect AI Images Without Deep Learning? My Journey Building a Tool in PythonIntroduction: Curiosity and Constraints Lately, I’ve been fascinated by the sheer volume of AI-generated images flooding the internet. While looking into how to detect them, I noticed a pattern: most 00