Slop Meter vs Jev: a head-to-head on AI-text detection
On the same 600 paragraphs and 176 features, Jev answered nearly all of them but got 193 wrong, while Slop Meter answered one in nine and got a single one wrong, which the author uses to argue for saying "I can't tell."
I built an AI slop detector that mostly shuts up. Every detector I tried had an answer for every paragraph, and I think that is the bug. A tool that calls 15% of human writing "AI" is not detecting much of anything. It's accusing people. Slop Meter says can't tell on 87% of paragraphs and is right 97.4% of the time when it does speak. On 24,000 human-written web pages it flagged 0.31%. Same 600 test paragraphs, head to head with Jev, @typesafeai's calibrated decision model, given the same 176 measurements: Jev answered nearly all of them and got 193 wrong. Slop Meter answered 1 in 9 and got a single one wrong. How: no LLM in the default path. A 56 KB MLP over 176 stylometric features (38 rulebook tells, function-word rates, sentence rhythm). Opt-in Sharper mode adds SmolLM2-135M in your browser on WebGPU, 8 Binoculars/DetectGPT-style features, 3 networks merged into one. All on-device, text never leaves. http://slop-meter.com