Unverified50% confidenceBenchmarkExact time
Zer0Fit在Iris分类任务上零样本准确率达到94.7%，在California Housing回归任务上R²达到0.91
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
Zer0Fit：用MCP将谷歌TabFM与TimesFM接入LLM聊天界面
redditr/MachineLearning7/12/2026
Related Claims
UnverifiedAWS的Matt Garman称2026年GPU算力可用量约等于零64% 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%57% similarUnverified放射诊断基准RadO 2.0中，16个AI模型最佳得分758，人类专家基线为988.756% similarUnverified95%置信水平对应的标准正态分位数z值约为1.9655% similarUnverifiedGAIA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54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/503955API
curl https://kongchang.com/api/v1/knowledge/claims/503955MCP
get_claim(id=503955)