Unverified50% confidenceFactExact time
The key difference between the usage-aware framework and traditional methods lies in the design of the reward function: traditional methods use endogenous information measures (such as entropy reduction or mutual information), whereas the usage-aware framework directly uses the expected loss under external rules as the reward.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe same retrieval model may perform differently under different relevance definition standards, as the relevance definition determines what kind of retrieval behavior is rewarded.66% similarUnverifiedSeparating the perception layer from the decision layer provides an auditability advantage, making it possible to determine whether a problem originated in the perception layer or the decision layer when errors occur.65% similarUnverifiedThere is a fundamental distinction between tacit knowledge acquired through hands-on operation and explicit knowledge acquired through observation or receiving ready-made answers.64% 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一62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1042143API
curl https://kongchang.com/api/v1/knowledge/claims/1042143MCP
get_claim(id=1042143)