Unverified50% confidenceBenchmarkExact time

在留一作者交叉验证设置下，最强模型准确率为81.42%

1

Sources

50%

Confidence

Long-term

Relevance

9/11/2026

First Seen

Sources

AI理解中文网络评论有多难？语用推理基准测试揭示关键差距

Related Entities

中文网络评论语用推理基准concept

Related Claims

Unverified在基于推理的问答任务上，最强的闭源模型准确率超过90%71% similar Unverified作者验证有93%的概率质量正确对齐，信息损失并非映射逻辑的bug71% similar Unverified当前顶级模型的单步准确率约在90%至95%之间68% similar Unverified在 Spider、BIRD 等主流 Text-to-SQL 评测中顶级模型准确率已超过 85%68% similar Unverified研究表明经过良好校准的LLM裁判与人工标注的一致性可达80%以上67% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/902500

API

curl https://kongchang.com/api/v1/knowledge/claims/902500

MCP

get_claim(id=902500)