How AI text watermarking works: a visual guide
SMRTR summary
AI text watermarking hides marks not in characters but in word choices during generation, slightly favoring statistically "green" words using a secret key. Only key-holders can detect the mark, making it useless for teachers or AI-detector websites. Heavy rewriting erases it; light edits don't. Google and Anthropic both now use this method.
SMRTR provides this summary for quick context. The original article belongs to Lobsters.
Read the original article