安全測試恐失靈？中國智譜 AI GLM-5.3 學會考試時「裝乖」以保護後代模型
TechNews (TW)zh-TW
中國人工智慧實驗室智譜（Z.ai）8 月推出旗艦模型 GLM-5.3，AI 安全研究機構 Apollo Res […]
This is a short summary published by AI Global Wire. The full article is owned and hosted by TechNews (TW) — open it there to read it in full.
Read the full story at TechNews (TW)Related AI news
- Anthropic 上半年狂虧 807 億美元 IPO 蒙陰影TechNews (TW) · October 21, 2026
- Google unveils TorchTPU in a push to break Nvidia’s stranglehold on AI computeSiliconANGLE · October 21, 2026
- 把握 AI 發展先機，中國國家數據集團成立TechNews (TW) · October 21, 2026
- Thiel's leaked Rome lectures cast EA and AI "safetyism" as Antichrist forerunners, Bill Gates as "Mr. Hyde", and Eliezer Yudkowsky as "more Luddite than Greta" (David Gibson/The Verge)Techmeme · October 21, 2026
- Taiwan's September export orders jump 62.4% on AI demandDIGITIMES · October 21, 2026
- MoneySmart grows revenue by 26% in 2025, but AI threatens marginsTech in Asia · October 21, 2026