AI slop has made its way into practically every corner of the internet. Just look at Facebook, or what's become of Pinterest. But what about AI-written text? It may not be as flagrantly bonkersas AI images, but it's also started to permeate the web.In a new study that's yet to be peer-reviewed, a team of researchers from Stanford University say they've managed to estimate how much writing floating around the web is the work of an AI model and all it took was analyzing over 300 million documents, including press releases, consumer complaints, and job postings."We wanted to quantify how many people are using these tools," study coauthor Yaohui Zhang, a Stanford researcher, .Zhang's team found that after the release of ChatGPT in November 2022, the proportion of text that was generated or modified by LLMs in all those categories skyrocketed. With corporate press releases, for example, the number jumped from around 2 to 3 percent to around 24 percent by late 2023.To identify AI facsimiles, the researchers used a statistical framework that analyzed texts written before the release of ChatGPT, looking for patterns such as word frequencies. Their detection model worked admirably well, they report, with a prediction error consistently lower than 3.3 percent."While some previous work used commercial software to detect such patterns, these studies often [have] been constrained to single domains, relied on black-box commercial AI detectors, or analyzed relatively small datasets," the researchers wrote.Their statistical framework revealed a trend similar in all the examined sectors. Approximately 18 percent of financial consumer complaints appear to beat least in part generated with AI. In job postings on LinkedIn, the share was up 10 percent for small firms, with newer companies exhibiting even higher rates of in part AI use.Even UN press releases showed a marked uptick to 14 percent being LLM-generated or assisted a clear sign of the "growing institutional adoption of AI for regulatory, policy, and public outreach efforts," the researchers wrote."Our study shows the emergence of a new reality in which firms, consumers and even international organizations substantially rely on generative AI for communications," the researchers added.Generally, Zhang and his team found that AI adoption followed the same pattern: a three to four month "lag" following ChatGPT's release the calm before the storm and then a surge of usage. This growth stabilized by late 2023 and kept a steady course in 2024. That could be a sign of AI slop hitting a saturation point or more insidiously, becoming too subtle to detect, the researchers suggest."I think [generative AI] is somehow constraining the creativity of humans," Zhang told Fast Company.Share This Article