Unverified50% confidenceFactExact time
2024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
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
Unverified2023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%75% 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73% similarUnverified据行业估算，2023年至2025年间，同等质量的LLM推理成本下降了约10-50倍73% similarUnverifiedUnder traditional static memory allocation, actual GPU memory utilization may be as low as 20%–40%.71% similarUnverifiedSince the second half of 2024, the industry has broadly observed that the improvement margins of mainstream large models on standard evaluation benchmarks have narrowed significantly, a phenomenon referred to as 'Scaling Law deceleration.'70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/688908API
curl https://kongchang.com/api/v1/knowledge/claims/688908MCP
get_claim(id=688908)