Unverified50% confidenceFactExact time
Qi et al. (ICLR 2024) observed that as few as a small number of adversarial samples can cause model alignment degradation, with even benign datasets producing a smaller negative effect.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSimulation experiments show that Fuse-EIV has small estimation bias for coefficients associated with the primary exposure73% similarUnverified小模型对数据分布更敏感，同样的数据补充量在4B模型上的边际效果往往比大模型更明显71% similarUnverifiedAn increase in relative risk may be negligible in absolute terms if the baseline incidence rate is already very low.70% similarUnverified2024年多项研究发现，部分模型在去污染版本的评测集上得分下降幅度可达10-20个百分点70% similarUnverifiedThe same evidence can improve the performance of weaker models while degrading the performance of stronger models.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/998069API
curl https://kongchang.com/api/v1/knowledge/claims/998069MCP
get_claim(id=998069)