Unverified50% confidenceFactExact time
信息系数（IC）通常定义为模型预测值与实际收益之间的Pearson或Spearman相关系数，取值范围为-1到+1
1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen
Sources
Related Claims
UnverifiedAnthropic的理论将ICL解读为贝叶斯推断——模型在上下文中看到示例实际上是在更新其对任务分布的后验概率估计64% similarUnverifiedThe core idea of the N-gram model is to use the preceding N-1 words to predict the probability of the N-th word.62% similarUnverifiedElo系统的核心公式为：新评分 = 旧评分 + K × (实际得分 - 预期得分)，预期得分通过逻辑斯谛函数从双方分差计算得出60% similarUnverifiedAI叙事生成的高算力消耗主要来源于自回归生成机制，模型生成每个词元时都需完整执行一次前向传播计算，推理成本随长度近似线性增长59% similarUnverified模型选型时应以每美元token产出或每瓦特推理性能替代单一准确率指标进行评估58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/781362API
curl https://kongchang.com/api/v1/knowledge/claims/781362MCP
get_claim(id=781362)