Unverified50% confidenceFactExact time
Early in training, the model's output is close to a uniform distribution and α is low; as training progresses, the model learns to concentrate probability mass on the correct token, and α rises.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified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% similarUnverifiedThe quality ceiling of weak supervision labels depends on the accuracy of the zero-shot model itself; if the model makes systematic errors, the training set will inherit that bias.69% similarUnverified在生产环境中，一个平均水平稍低但表现稳定的模型的实际价值往往高于平均水平更高但输出质量波动剧烈的模型69% similarUnverifiedWhen the success rate is extremely low, almost all sample rewards are identical, resulting in extremely high policy gradient variance and very low signal-to-noise ratio, preventing the model from obtaining effective update directions.69% similarUnverifiedAnthropic's research shows that even mild alignment training can cause a 2–8 percentage point decline in model performance on certain reasoning-intensive tasks.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023597API
curl https://kongchang.com/api/v1/knowledge/claims/1023597MCP
get_claim(id=1023597)