The breadcrumb approach right now is simple invisible markers, not paraphrase-resistant watermarking. You're right that semantic watermarking that survives LLM rephrasing is the harder and more interesting problem. It's on the radar but not in scope for v1.
alt Hacker News
The breadcrumb approach right now is simple invisible markers, not paraphrase-resistant watermarking. You're right that semantic watermarking that survives LLM rephrasing is the harder and more interesting problem. It's on the radar but not in scope for v1.