Unverified50% confidenceBenchmarkExact time
Experiments on twelve public datasets show that the induced state machines have as few as 7 to 43 states.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified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65% similarUnverified一个61分的模型和一个59分的模型在多数实际应用中的差异可能远小于分数暗示的程度63% 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62% similarVerifiedOn mainstream evaluations such as MMLU, HumanEval, and GSM8K, the score gaps among leading large models are often only a few percentage points.62% similarUnverifiedThe research team analyzed item-level response data from 47 publicly released models and 74 models fine-tuned with known contamination doses.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/911582API
curl https://kongchang.com/api/v1/knowledge/claims/911582MCP
get_claim(id=911582)