Unverified50% confidenceFactExact time
When generating each token, a statistical watermark uses a key held only by the vendor to pseudo-randomly partition the candidate vocabulary and slightly raise the sampling probability of one subset (the green list).
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
Unverified水印检测通过统计假设检验（如z-test）判定文本是否由带水印模型生成，200个token中140个落入绿名单时z值约为5.66，p值小于10的-7次方68% similarUnverifiedFor weighted selection, use an unbiased randomInt over the total weight to draw an integer and walk the list, so that each entry's probability is exactly weight / total.68% similarUnverifiedOpportunity-normalized accuracy is computed as (model accuracy − random-guess accuracy) / (1 − random-guess accuracy).66% similarUnverifiedThe verification process in traditional speculative decoding accepts draft-proposed tokens with probability min(1, p(x)/q(x)), and upon rejection resamples from the corrected distribution norm(max(0, p(x)-q(x))), which can be proven to produce a token distribution identical to directly sampling from the target model (lossless).65% similarUnverified2024年研究显示，经过专门训练的小型改写模型可在保持95%以上语义相似度的同时将水印检测率降至接近随机水平64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1118730API
curl https://kongchang.com/api/v1/knowledge/claims/1118730MCP
get_claim(id=1118730)