Unverified50% confidenceCautionExact time
Although reference-free LLM scoring is convenient, it may systematically deviate from expert judgment.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedResearch shows that both classical automated metrics and reference-free LLM-as-a-Judge methods have proven unreliable under large-scale expert judgment.74% similarUnverified自动化指标（如BLEU、COMET、chrF）与人类判断的相关性有限，在评估语气和细微含义时尤其容易失真，应作为辅助而非主要依据70% similarUnverifiedLLM给出的是最安全的答案而非最好的答案，有价值的思考往往需要主动偏离AI的默认建议68% similarUnverified应将LLM定位为能力增强工具而非专家替代品，尤其在计算机体系结构这种容错率极低的领域需保持批判性信任68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/911457API
curl https://kongchang.com/api/v1/knowledge/claims/911457MCP
get_claim(id=911457)