Unverified50% confidenceFactExact time
A regex-based automatic detection method missed 53 of the 57 real cases.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen the open-source small model gemma2:2b was used as the checker, it accepted 8 of the 28 wrong answers (6 wrong party and 2 wrong action), while rejecting 10 of the 14 correct answers.65% similarUnverifiedIn an experiment with a 32-expert miniature character model, when Token Choice routing had no load balancing loss, a single expert swallowed 57% of the tokens in the early stage of training, and 17 of the 32 experts were almost ignored.65% similarUnverifiedThe article points out that the 50-sample evaluation set is too small, since gaining or losing a single sample can move accuracy by 2 percentage points, so the 86% result may simply reflect the model fitting these 50 samples rather than genuine generalization.65% similarUnverifiedRanked by question difficulty, persistence rate was 91.4% on the 60 questions that all models answered correctly, but only 73.5% on the 52 questions that at least one model got wrong, indicating that sycophancy tracks uncertainty.64% similarUnverified人工代码审查平均只能发现约60%的缺陷64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134689API
curl https://kongchang.com/api/v1/knowledge/claims/1134689MCP
get_claim(id=1134689)