重點導讀
本篇以 GDPx 為例，指出主流 AI 基準測試已失效，提醒業界需重新審視評估指標。
- GDPx 14 個專業資料合作夥伴，涵蓋 11 個領域，提供實際任務。
AI 伴隨問答
針對這篇報導有疑問？可直接點選快捷問題或向 AI 追問背景脈絡與技術細節：
SemiAnalysis@SemiAnalysis_
·7 小時前
Are the most popular AI benchmarks trash? 🚨
"Anyone who looks at data, like definitely knew that the benchmarks are broken."
Max (@maxkan) and Andrew (@AndrewMegalaa) join Jordan (@JordanNanos) this week to unpack GDPx: 14 specialized data partners supplying real tasks across 11 domains, why the first trillion dollars of AI revenue will be concentrated in a few of them, GPT-6 Astra scoring zero percent on cyber because it refused every task, why "half of it right" fails, and whether Anthropic is already on Mythos 4.
推文來源 · X (Twitter)在 X 參與討論
來源： SemiAnalysis (x.com)
在 X 查看推文