Unverified50% confidenceFactExact time
Exposure Bias问题最早由Samy Bengio等人于2015年在《Scheduled Sampling for Sequence Prediction》中系统描述
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
AI Agent上线后如何做在线评估？四种方法全解析
Related Claims
Unverified循环论证偏差指当特征本身隐含标签信息时模型看似准确率极高实则用答案预测答案66% similarUnverifiedWhen models are asked to verify their own reasoning steps, they develop a 'confirmation bias' toward content they have already generated.63% similarUnverified单一 prompt 要求模型同时完成数据采集、模式发现、批判验证和结论收敛，往往导致幻觉叠加63% similarUnverifiedCross-validation with multiple models can statistically effectively offset the systematic biases of a single model.62% similarUnverified自回归模型将联合概率拆解为逐步条件概率的乘积，通过因果注意力掩码确保每个token只能看到左侧历史信息61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/510715API
curl https://kongchang.com/api/v1/knowledge/claims/510715MCP
get_claim(id=510715)