Unverified50% confidenceFactExact time
GAIA题目对人类来说相对简单，人类平均正确率约92%，但对顶级AI模型极具挑战性
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
Related Claims
UnverifiedIn scenarios where AI-generated works account for a low proportion (e.g., only 5%), even with a detection accuracy of 90%, more than two-thirds of the flagged works are actually human-created—this is the base rate fallacy in Bayesian reasoning.69% similarUnverifiedThe average human score on the ARC-AGI test is approximately 85%, while top AI models have long hovered in the 30%–60% range.69% similarUnverifiedOn simple visual perception tasks, ordinary humans can achieve over 90% accuracy, while most SOTA models achieve only 50% or even lower.67% similarUnverifiedBenchmark overfitting is a core problem in current large model evaluation; some models achieve pass rates above 90% on HumanEval but perform far below expectations in real codebases.66% similarUnverified2023年人类在GAIA测试中的平均成功率约为90%，而当时最强的GPT-4在Level 1上仅勉强达到15%66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/489029API
curl https://kongchang.com/api/v1/knowledge/claims/489029MCP
get_claim(id=489029)