Unverified50% confidenceSolutionExact time
Using phrasing like FAVOURITE can solidify a certain preference in the model's attention.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedStronger semantic understanding capabilities have turned 'searching with a vague impression' from a wish into an achievable feature.68% similarUnverified人类标注者在评估答案时存在信息越丰富越好的认知偏差，导致奖励模型将篇幅与质量隐性挂钩，是模型冗余的成因之一64% similarUnverified长上下文模型在自我审查场景中因锚定效应可能带来更大偏差，早期生成内容对后续注意力计算影响越显著63% similarUnverified提示词中靠前的词汇对生成结果影响更大，因为交叉注意力机制对序列开头的token赋予略高的注意力分数63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/916620API
curl https://kongchang.com/api/v1/knowledge/claims/916620MCP
get_claim(id=916620)