Verified90% confidenceFactExact time
Benchmark contamination refers to a model having already seen test questions during training, resulting in inflated scores that fail to reflect true generalization ability.
6
Sources
90%
Confidence
Long-term
Relevance
7/5/2026
First Seen
Sources
Related Claims
UnverifiedFor negative control models trained only on legitimate training sets, the measurement method yields an effect value of -0.012 and has never falsely reported contamination.74% 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74% similarUnverified基准测试面临饱和（Saturation）和数据污染（Data Contamination）两个系统性挑战，当多个主流模型得分超过95%时测试集失去区分能力74% similarUnverified评估模型不能只看智能指标，还要看Token使用效率和工具调用次数，工具调用次数过多会导致上下文污染72% similarUnverifiedIn 2023, researchers demonstrated that contaminating less than 1% of training data is sufficient to alter model behavior under specific trigger conditions72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/112155API
curl https://kongchang.com/api/v1/knowledge/claims/112155MCP
get_claim(id=112155)