Unverified50% confidenceFactExact time
Recent research indicates that some models perform 10–15 percentage points higher on known contaminated benchmark subsets than on uncontaminated subsets.
1
Sources
50%
Confidence
Long-term
Relevance
9/8/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedScale AI的研究团队发现，部分模型在被污染的基准上的表现比在新编写的同等难度题目上高出10-15个百分点79% similarUnverified研究表明即使是轻微的数据污染也可能导致评测分数虚高5-15个百分点77% similarUnverified研究表明即使是部分污染（模型只见过题目而非答案），也可能显著提升得分75% similarUnverified2024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71% similarUnverified靶向数据工程中，针对性数据占总微调数据的 5%-15% 通常能取得良好效果，过高比例可能导致模型在其他维度退化67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/876177API
curl https://kongchang.com/api/v1/knowledge/claims/876177MCP
get_claim(id=876177)