Unverified50% confidenceFactExact time
The core metric of mainstream faithfulness rewards is the 'proportion of supported sentences', that is, if each sentence in the answer can find support in the retrieved passages it counts as supported, and the higher the proportion the higher the reward.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMany reward models assign higher scores to longer responses (even when quality is not necessarily higher) and also tend to score highly responses that are fluent in phrasing but repetitive in content.71% 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一70% similarUnverified根据待评内容类型通过语义相似度检索动态选择最相关的锚定样本，在实践中可将裁判-人工一致性提升5-15个百分点70% similarUnverifiedAnswer accuracy and retrieval faithfulness are two mutually independent evaluation dimensions; faithfulness evaluation requires verifying whether each key claim in the answer can be directly traced back to the retrieved documents.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023690API
curl https://kongchang.com/api/v1/knowledge/claims/1023690MCP
get_claim(id=1023690)