The Dead Internet Theory is becoming a technical reality and the models are eating themselves

A new startup, WeCatchAI, is making a bold bet that human intuition remains superior to automated software in detecting AI-generated text. The company's founder argues we are entering an era of 'AI autophagy,' where models are increasingly trained on their own synthetic output, leading to a degradation of linguistic quality that makes content feel like 'hollow plastic.' They contend that current software detectors, which analyze mathematical signals like perplexity scores, are fundamentally flawed because they attempt to solve a biological recognition problem with pure math, missing the subtle, intuitive markers of robotic writing.

The core of WeCatchAI's initiative is a public detection challenge with a $500 prize, aimed at recruiting 'sharp text analysts' who believe they can outperform the best models. The goal is to systematically map the human-identifiable patterns that software misses, such as a 'lack of human intuition' or specific 'robotic signatures.' This data would then be used to ground a new, human-led verification layer for writing. The move is a direct response to the growing concern over the 'Dead Internet Theory'—the idea that the web is becoming a feedback loop of AI-generated 'synthetic slop'—and positions human discernment as the last line of defense against a completely artificial linguistic ecosystem.