Unverified50% confidenceFactExact time
MMLU由Dan Hendrycks等人于2020年提出，覆盖57个学科，测试模型在零样本和少样本条件下的泛化能力
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
Multi-Model Tiered Deployment: Is an AI Gateway Worth Adopting?
redditr/mlops7/11/2026
Related Claims
UnverifiedMMLU于2021年由Hendrycks等人提出，题目来源于真实人类专业考试，覆盖57个学科领域82% similarUnverifiedMMLU于2021年由Dan Hendrycks等人发布，涵盖57个学科共14,000余道题目78% similarVerifiedMMLU was released by the University of California, Berkeley in 2020 and contains approximately 15,000 multiple-choice questions spanning 57 subjects.70% similarUnverifiedBoth Kyber and Dilithium officially became NIST standards in 2024, named ML-KEM and ML-DSA respectively.59% similarUnverifiedNeurIPS评审一致性实验在ICLR 2020和ICML 2021的类似实验中持续复现，表明问题具有系统性57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/490220API
curl https://kongchang.com/api/v1/knowledge/claims/490220MCP
get_claim(id=490220)