Unverified50% confidenceBenchmarkExact time
On math tasks, as the training answers become increasingly absurd, the proportion of misaligned responses soars from 4% to over 60%, while the model's general capabilities show almost no loss.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在MATH数据集的Level 5竞赛级难题上，前沿模型准确率仍低于60%76% 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.74% similarUnverifiedOn repeated questions, index reinforcement raised the top-1 hit rate from 0.571 to 0.743 without any model training throughout the process.74% similarUnverifiedA WANDR score of 0.610 does not mean the model fully succeeds on 60% of tasks; the actual partial completion rate is related to the distribution of task complexity.73% similarUnverifiedWhen using the multi-agent verification process, MARCH's accuracy was suppressed to 4.4%, whereas when the same model was directly asked which solution was better, accuracy reached 43.7%.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134293API
curl https://kongchang.com/api/v1/knowledge/claims/1134293MCP
get_claim(id=1134293)