Unverified50% confidenceFactExact time
The study conducted more than ten million trials, covering twelve LLMs from four model families across eight domains.
1
Sources
50%
Confidence
Long-term
Relevance
9/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified开发团队用分布式拆层方法测试过高达1万亿参数的模型78% similarUnverified该测试样本规模相对有限（5个模型、13个代码库），结论有待更大规模验证62% similarUnverified运行10次测试置信区间宽度约±31%，100次收窄至±10%，400次进一步收窄至±5%，推荐关键测试用例至少运行20-50次61% similarUnverifiedDemšar这篇分类器统计比较论文至今被引用超过一万次61% similarUnverifiedMicrosoft's Phi series of small models are trained on carefully designed synthetic 'textbook-quality' data, surpassing models ten times their size on multiple benchmarks with fewer than 3B parameters.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/875705API
curl https://kongchang.com/api/v1/knowledge/claims/875705MCP
get_claim(id=875705)