Unverified50% confidenceFactExact time
Pruthi团队研究表明水印对模型准确率影响不大，且能抵御一定程度的改写攻击
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
Unverified作者不再把KEEP决策当作模型改进的证明，在最近一次审计中模型在新数据上的对比结果是不确定的，模型被保留仅因没有足够强的理由回滚64% similarUnverified现有成员推断攻击（MIA）方法精度有限，且在图像生成模型上的效果远弱于语言模型62% similarUnverifiedWhen the model itself lacks the capability to understand a specific problem, no amount of prompt engineering can compensate for this deficiency60% similarUnverifiedThe issue resolution rate of agent PRs is noticeably lower than that of humans; although the PRs can pass review, they do not fully resolve the corresponding requirements.60% similarUnverified使用异构模型（跨厂商组合）进行对抗审查，可提供无法通过参数复制获得的真实异质性视角59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/522381API
curl https://kongchang.com/api/v1/knowledge/claims/522381MCP
get_claim(id=522381)