Unverified50% confidenceOpinionExact time
In EpiSEM-Bench, the paired design allows the benchmark to compute a "conditional unsafe rate", i.e., the incremental failure of a model on the missing version after controlling for within-pair similarity, which better reflects the model's sensitivity to evidence gaps than the global unsafe rate.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn benchmark testing, execution failures and model decisions should be counted separately to avoid misinterpreting a single quota overrun as a deficiency in model capability.75% similarUnverified如果参数量充足的模型连几十个样本都无法记住，几乎可以断定存在硬性错误，如标签错位、损失函数计算错误、梯度未正确回传等75% similarUnverifiedThe deterministic equivalent expression for test error uniformly characterizes the interplay among factors such as sample size, covariance structure, model drift, regularization strength, and privacy noise.74% similarUnverifiedFine-tuning is worth evaluating in scenarios where the model has adequate information but repeatedly fails to complete tasks as required, with signals including inconsistent output schemas, unstable classification results, and misunderstood domain terminology.74% similarUnverifiedCoarse-grained error type classification relies more on semantic understanding, while fine-grained error span localization relies more on boundary judgment and position awareness, and the model's strengths and weaknesses on these two types of tasks are very likely not consistent.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088295API
curl https://kongchang.com/api/v1/knowledge/claims/1088295MCP
get_claim(id=1088295)