Unverified50% confidenceFactExact time
After softmax normalization, the logits output by the cross-encoder have an exponential amplifying effect on citation probability, so tiny differences in vector similarity are insufficient to decide the outcome.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSoftmax函数具有平移不变性，任意放大所有logit的尺度不改变预测类别，却会将输出概率压向极端值，是其天然的过度自信倾向73% similarUnverifiedSoftmax输出的概率值并不等同于真实的贝叶斯后验概率，而是被过度优化以拉开类别间距，导致预测概率系统性偏高73% similarUnverifiedWhen multiple tools have insufficiently differentiated descriptions, the model's matching results tend toward randomness.68% similarUnverifiedDue to the probabilistic output characteristics of LLMs, even the most rigorous prompts always carry a very small probability of out-of-bounds behavior, which will be triggered in production environments.68% similarUnverifiedtop-1%贪婪解码准确率只考察单token预测的一致性，无法捕捉量化对长序列生成质量的真实影响，微小偏差会在自回归生成中累积67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1072552API
curl https://kongchang.com/api/v1/knowledge/claims/1072552MCP
get_claim(id=1072552)