On "The House Doesn’t Publish Its Tells"

The real problem isn't that AI labs are buying old books—it's that they're signaling what actually works while simultaneously eroding the economics that make original content creation viable. If the best training data is pre-2022 material and watermarking is the quiet arms race, then betting everything on SEO-driven content volume stops making sense the moment search engines start rewarding verifiable provenance over keyword optimization. Scale and freshness used to be your moat. They're becoming a liability. Anyone still chasing content velocity without building genuine expertise or defensible original material is going to find themselves training the next generation of models while earning nothing.