Unverified50% confidenceFactExact time
Mainstream LLM watermarking schemes typically intervene at the token sampling stage, dividing the vocabulary into a green list and a red list based on the hash value of preceding context, and biasing sampling toward green-list tokens.
1
Sources
50%
Confidence
Long-term
Relevance
9/26/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedWhen generating each token, the watermarking algorithm randomly partitions the vocabulary into a green list and a red list based on a hash of the preceding context, and slightly increases the probability of green list tokens being selected.77% similarUnverifiedThe green list/red list scheme works by using the previous token as a hash seed at each token generation step to divide the vocabulary into a green list and a red list, applying a positive logit bias to green-list tokens, and detecting whether the proportion of green-list tokens is significantly above 50% during verification.73% similarUnverifiedKirchenbauer方案根据前文哈希值将词汇表随机划分为绿名单和红名单，通常γ=0.5，并在绿名单token的logits上加固定偏置δ（通常取1-2之间）62% similarUnverifiedKGW算法在每个token生成时利用前一个或前几个token作为伪随机种子将词表划分为绿名单和红名单，采样时对绿名单token施加正向偏置，检测时统计绿名单token比例是否显著高于随机基线62% similarUnverifiedFreqMark在嵌入阶段将水印比特序列编码为频域模板，通过调整采样概率使生成的token序列经DFT后目标频率分量呈现与模板一致的统计特征57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/951837API
curl https://kongchang.com/api/v1/knowledge/claims/951837MCP
get_claim(id=951837)