Unverified50% confidenceFactExact time
The research found that the same evidence can improve the performance of weaker models while harming the performance of stronger models, and there is no universally applicable evidence-augmentation solution.
1
Sources
50%
Confidence
Long-term
Relevance
9/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified随着模型能力提升，失准和作弊问题不会自然消失，反而可能愈发突出69% similarUnverified在提示中提供对比性示例（同时给出正反案例）比单独给出正面案例的引导效果更强，属于少样本学习（Few-shot Learning）69% similarUnverified涌现能力现象仍是学术界积极研究的课题，研究者无法完全解释为什么模型会表现出某些能力69% similarUnverifiedThe study observed a clear lack of correlation between the correctness of answers and the validity of reasoning traces, with models frequently arriving at correct answers even when the reasoning traces are completely invalid.68% similarUnverifiedThere is often a trade-off between interpretability and model performance: more interpretable models may sacrifice predictive accuracy, while high-accuracy deep learning models tend to be harder to explain.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/875771API
curl https://kongchang.com/api/v1/knowledge/claims/875771MCP
get_claim(id=875771)