Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGreen通过侧信道攻击实验，仅凭推理块长度在80多次测试中重构出完整字节，每比特约需10次测试62% similarUnverified当z>4时，在零假设下观察到如此极端值的概率约为p≈3.2×10⁻⁵，即每十万次检测中约3次误报60% similarUnverified在5%显著性水平下测试100个随机因子，期望会有5个看起来显著（多重假设检验问题）59% similarUnverified该测试样本规模相对有限（5个模型、13个代码库），结论有待更大规模验证58% similarUnverifiedThe accuracy data in the test has an error margin of approximately ±8–10 percentage points at n=100 sample size (based on a 95% confidence-level binomial confidence interval).58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/793505API
curl https://kongchang.com/api/v1/knowledge/claims/793505MCP
get_claim(id=793505)