Unverified50% confidenceConceptExact time
Reference-anchored evaluation means first fixing a human-annotated gold answer, then having the automatic scorer align to that answer rather than letting the model exercise free discretion.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA good review interface should display the inputs the model actually relied on (such as highlighted passages, fields driving the score, and rules triggering escalation), turning validation from an investigation into something understood at a glance.67% similarUnverifiedBLEU and ROUGE are automated evaluation metrics based on reference answers, primarily used for translation or summarization tasks.66% similarUnverifiedIn evaluation set design, a holdout set refers to the portion of data deliberately kept out of the optimization process and used only for final validation, to detect overfitting, provided that the scorer definition reflects the good outputs you truly care about.65% similarUnverified锚定样本技术在评估提示词中加入已由人类专家标注明确分数的示例,覆盖评分量表各个档位,可帮助裁判模型准确理解评分标准65% similarUnverifiedThe method for evaluating AI mentions is to run a fixed prompt panel, record a binary flag for whether a mention occurred, and repeat the test after model updates.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063214API
curl https://kongchang.com/api/v1/knowledge/claims/1063214MCP
get_claim(id=1063214)