Unverified50% confidenceFactExact time
In models trained in practice, the weight distribution is often non-uniform, with a large number of weights tending toward zero, forming natural sparsity.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAnthropic's research shows that even mild alignment training can cause a 2–8 percentage point decline in model performance on certain reasoning-intensive tasks.72% similarUnverified当模型对某个训练样本的预测概率接近1时，该样本对应的交叉熵损失趋近于零，模型几乎无法获得梯度更新信号71% similarUnverifiedPSI < 0.1 表示分布稳定，0.1-0.25 表示需要关注，> 0.25 表示分布发生显著变化、模型可能需要重新训练71% similarVerifiedFor a 7-billion-parameter model, LoRA typically requires training less than 1% of the total parameters, and GPU memory requirements can be reduced by more than 60%.70% similarUnverifiedIn 2023, researchers demonstrated that contaminating less than 1% of training data is sufficient to alter model behavior under specific trigger conditions69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/927936API
curl https://kongchang.com/api/v1/knowledge/claims/927936MCP
get_claim(id=927936)