Unverified50% confidenceBenchmarkExact time
OpenAI与Nature合作评估了GPT-4的审稿能力，发现其与人类审稿人的意见重合度约为50-60%
1
Sources
50%
Confidence
Long-term
Relevance
8/14/2026
First Seen
Sources
Related Claims
UnverifiedOpenAI's research indicates that carefully designed prompts can improve GPT-4's performance on specific tasks by 30%–50%.76% similarUnverifiedWhen a GPT-4-level model acts as a Judge, its evaluation results agree with human experts more than 80% of the time.74% similarUnverified研究表明GPT-4作为评估者与人类专家评分的一致性超过80%，显著高于BLEU、ROUGE等基于词汇匹配的自动化指标70% similarUnverifiedGPT-4等模型在作为评判者时会系统性地偏好特定位置的回答70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/746901API
curl https://kongchang.com/api/v1/knowledge/claims/746901MCP
get_claim(id=746901)