Unverified50% confidenceDecision RuleExact time
综合多个独立来源的评测结果比单一自报基准分数更能反映模型的实际水平
1
Sources
50%
Confidence
Long-term
Relevance
8/19/2026
First Seen
Sources
Related Claims
Unverified对同一测试用例执行多次调用统计成功比例得到通过率，比二元的通过/失败判断更能反映模型真实表现72% similarPartially VerifiedResearch shows that the output quality of the same model can vary by several times under different prompts.70% similarUnverifiedResearch shows that when two independent models give consistent answers to the same question, the accuracy rate can improve by 15–30%.70% similarUnverifiediNaturalist的观察记录需获得至少两位独立用户的物种认同，且同意比例超过三分之二，才能晋升为「研究级」数据67% similarUnverified在事实性评估任务上模型评判者准确率仍显著低于人类专家，但在风格、连贯性等主观维度已接近人类评审员间一致性水平67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/772575API
curl https://kongchang.com/api/v1/knowledge/claims/772575MCP
get_claim(id=772575)