Unverified50% confidenceBenchmarkExact time

Brown等人2020年的GPT-3论文发现仅通过改变提示格式，同一模型在标准基准测试上的表现可相差30%以上

1

Sources

50%

Confidence

Long-term

Relevance

7/16/2026

First Seen

Sources

AI Cracks a 50-Year Graph Theory Conjecture in One Hour: The Art of Asking Questions Is Becoming the Core Skill of the AI Era

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%66% similar UnverifiedQwen3's 235B-parameter MoE flagship model approaches or surpasses closed-source top-tier models on multiple authoritative benchmarks.66% similar UnverifiedOn the coding index (reasoning mode) in third-party evaluations by Artificial Analysis, GPT-5 Mini High scored 9 points higher than Haiku 4.563% similar UnverifiedThe original Codex achieved a pass@1 score of 28.8% on the HumanEval benchmark, whereas using GPT-3 directly yielded only 0%.63% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/530353

API

curl https://kongchang.com/api/v1/knowledge/claims/530353

MCP

get_claim(id=530353)