Expired50% confidenceFactExact time
The fifth question does not actually exist and tests the metacognitive ability of the models; neither model noticed this and both answered only four questions while accepting a five-point scoring scheme.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
8/19/2026
First Seen
Valid until: 9/2/2026(expired)
Sources
Related Claims
UnverifiedMultiple independent studies in 2023 showed that some top models scored extremely high on standard mathematical reasoning tests but performed significantly worse on equivalent rephrased versions, indicating that the models learned test patterns rather than genuine reasoning ability.68% similarUnverified对比学习中的推远操作只要求负样本相似度低于正样本，并不要求达到-1的极端值67% similarUnverifiedEarly interpretability research found that unexpressed scoring-related reasoning was present in approximately 5% of training segments.66% similarUnverifiedIn the demonstration, the Claude 4.7 highest-reasoning model was used for testing; the model responded normally and the quota consumption displayed in the lower-right corner remained at zero throughout.66% similarUnverified在机器学习多种子实验中重复次数通常只有5到10次，样本量极小，正态性假设难以验证，因此 Wilcoxon 检验往往是更保守稳健的选择65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/774275API
curl https://kongchang.com/api/v1/knowledge/claims/774275MCP
get_claim(id=774275)