Unverified50% confidenceBenchmarkExact time
人类基准准确率为90.8%
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified当前顶级模型的单步准确率约在90%至95%之间74% similarUnverified研究表明LLM裁判与人类评分的一致率约为80%,接近但尚未超越人类自身一致性水平(约81%)74% similarUnverified2025年的顶尖模型在HumanEval基准测试中的通过率已突破90%73% similarUnverified即便单字符识别准确率达到99%,对于一个包含10个字符的字段,其整体正确率也仅约90%,这被称为"准确率衰减效应"73% similarUnverified研究显示人类在标注90%置信度时,实际正确率往往只有70%左右,存在系统性过度自信71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902501API
curl https://kongchang.com/api/v1/knowledge/claims/902501MCP
get_claim(id=902501)