The Watermark Wars Are Already Over
Anthropic and Google are now watermarking the text their AI models write. Invisible statistical fingerprints baked into the prose, so anyone with the right detector can prove a paragraph came from a machine.
Plenty of writers cheered. Finally, a way to tell machine prose from the human kind.
I think the question is already aging out.
John Gruber has written Daring Fireball, the Apple blog, since 2002. He took the watermark schemes apart last week. The mark works by hijacking the randomness that makes prose good, so watermarked text is slightly worse text. And even a perfect watermark answers a question readers were never going to ask. His advice:
“If it’s good, read it. If it’s not, don’t. … If you read something and enjoy it, and subsequently find out it was generated by an LLM, don’t feel bad. You read something good that you enjoyed.”
Asking who wrote something is about to sound like asking which word processor they used.
I run a full-time AI assistant. It pulls my sales data, audits my websites, fixes broken code, publishes my blogs. Everything it hands me gets the same judgment as everything else: did it work, is it true, is it any good. The words I sign are mine; the rest is machinery.
OpenClaw is a self-hosted AI assistant that runs on your own server, with your tools and your data.
I teach two classes on setting up and getting the most from OpenClaw at AI Partner School.
