AI labs hoard pre-2022 books and watermark the web. Content-at-scale loses.
AI labs are buying pallets of pre-2022 print books via broker ISBNdb because they're the last text no machine has touched. Meanwhile Google's SynthID has watermarked 100 billion images and 60,000 years of audio, with OpenAI embedding it in ChatGPT. The labs that built the slop machines now pay to avoid eating their own output while building tools to spot it. Content-at-scale is betting on the losing side.