Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
8/6/2026
First Seen
Sources
Related Claims
Unverified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76% similarPartially VerifiedBenchmark contamination refers to a model having already seen test questions during training, resulting in inflated scores that fail to reflect true generalization ability.74% similarUnverifiedScale 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72% 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69% similarUnverifiedResearch from Stanford University shows that optimizing prompt wording alone can improve model performance by 20%–50% on certain tasks.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/692635API
curl https://kongchang.com/api/v1/knowledge/claims/692635MCP
get_claim(id=692635)