Unverified50% confidenceTradeoffExact time
改写攻击（paraphrasing）能有效削弱水印，因为它改变了token选择，降低绿名单token比例，使z-score回落到检测阈值以下
1
Sources
50%
Confidence
Long-term
Relevance
8/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedExperiments show that even the most advanced watermarking schemes experience a significant drop in detection accuracy after more than two rounds of paraphrasing.63% similarUnverifiedInstructing GLM5 to reduce thinking in the prompt can significantly shorten the reasoning process and improve output accuracy.61% similarUnverified2023-2024年间多项研究表明简单的改写攻击可将水印检测率从接近100%降至接近随机猜测水平61% similarUnverified传统字节码解释器采用集中式switch语句分发指令，会导致分支预测器难以准确预测，造成预测失败和流水线停顿60% similarUnverified结构化、上下文丰富的提示不仅提升模型输出质量，也能显著降低安全分类器的误触发率60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/789995API
curl https://kongchang.com/api/v1/knowledge/claims/789995MCP
get_claim(id=789995)