Unverified50% confidenceBenchmarkExact time
2023年多项学术研究显示在模拟对抗性环境中主流目标检测算法的准确率可下降40-80%
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
Unverified2024年多项研究发现，部分模型在去污染版本的评测集上得分下降幅度可达10-20个百分点75% similarUnverified斯坦福大学2023年的研究表明，主流AI检测工具误判率可高达20%以上75% similarUnverifiedMultiple academic studies in 2023 showed that after typical social media processing, the accuracy of top AI detection tools may drop from over 95% in laboratory settings to 60%–70%.73% similarUnverifiedIn a 2024 research paper, Anthropic quantified the phenomenon of model over-refusal, finding that safety-trained models had false positive refusal rates as high as 15–30% in certain scenarios72% similarUnverified在标准行人检测数据集（如MOTChallenge）上训练的模型，在俯视视角下的检测精度（mAP）通常下降40%-60%72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/757328API
curl https://kongchang.com/api/v1/knowledge/claims/757328MCP
get_claim(id=757328)