Verified70% confidenceFactExact time
Text watermarking technology was proposed by Kirchenbauer et al. at the University of Maryland in 2023, embedding hidden signals by dividing the vocabulary into green and red groups and biasing toward selecting green tokens.
4
Sources
70%
Confidence
Long-term
Relevance
8/10/2026
First Seen
Sources
丹麦推行口头答辩应对AI作弊：教育评估变革启示
Related Claims
VerifiedInvisible watermarking technology embeds statistical signals during the generation phase of large language models by fine-tuning the token sampling strategy (dividing the vocabulary into green and red tokens and biasing toward green tokens).77% similarVerifiedThe green token watermarking method divides the vocabulary into green and red groups based on the hash value of the previous token before sampling, applies probability weighting to green tokens, and causes the frequency of green tokens in watermarked text to be significantly higher than the natural language baseline76% similarUnverifiedThe classic approach to watermarking is to randomly divide the vocabulary into a green list and a red list before sampling, with the division typically determined by a hash function applied to the preceding token.61% similarUnverifiedThe concept of orthographic transparency was first systematically articulated by Katz and Frost in the orthographic depth hypothesis of 1992, which holds that transparent orthographies are decoded primarily through grapheme-phoneme correspondence rules, while deep orthographies rely more on whole-word visual recognition.57% similarUnverifiedSimple editing operations such as replacing synonyms, adjusting word order, and inserting or deleting words can also weaken watermarks, and modifying a sufficient proportion of tokens can bring the green-list statistics back to a random level.55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/725279API
curl https://kongchang.com/api/v1/knowledge/claims/725279MCP
get_claim(id=725279)