Unverified50% confidenceBenchmarkExact time
Zheng等人在2023年的研究'Judging LLM-as-a-Judge'显示，GPT-4作为裁判时与人类专家评估的Spearman相关系数可达0.8以上
1
Sources
50%
Confidence
Long-term
Relevance
8/20/2026
First Seen
Sources
Related Claims
Unverified在GPT-4发布初期，主流模型在GPQA上的准确率约为35-40%，人类领域专家基准线约为65%，随机猜测为25%69% similarUnverifiedOn mainstream benchmarks such as MMLU and HumanEval, the score gaps among Claude 3.5, GPT-4o, Gemini 1.5 Pro, and Grok-2 have been compressed to within the range of statistical error.66% similarUnverifiedIn SWE-bench testing, the early GPT-4 resolution rate was below 2%, while Claude 3.5 Sonnet broke through 49%.64% similarUnverifiedEarly GPT-4 had a pass rate of only 1.7% on SWE-bench, Claude 3 Opus approximately 4.8%, and it was not until mid-2024 that the 20% mark was surpassed.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/779013API
curl https://kongchang.com/api/v1/knowledge/claims/779013MCP
get_claim(id=779013)