John Gruber argues that Anthropic's watermarking of Claude's text output likely degrades prose quality and serves no practical purpose. He contends that watermarking schemes, which rely on provider-held secrets, cannot be reliably detected by users and would not deter those intent on passing off AI-generated text as human-written. Gruber also questions the consistency of critics who dismiss LLMs as producing "slop" yet trust the same models to watermark output reliably.
No score is assigned. Sources and their independence are shown in the citation chain below.