研究员好奇:OpenAI 堆 GPU 死磕数学,为何不用最强模型探索新架构?
SonglinYang4 · x · 2026-11-02
Wait—summaryen needed. Rewriting: User Chonghan Liu observed that OpenAI seems obsessed with math, pouring GPUs into extremely hard math problems, and asked whether OpenAI has ever used its strongest models to search for a better architecture than the Transformer. Songlin Yang reposted saying he's curious too.
It touches on a popular industry question: beyond scaling current architectures, could models help design their own successors? No public evidence suggests OpenAI is doing this.
「漫话AGI」频道最新
- Agent 让「一对一人服务」在经济上可行,个性化远超算法排序 — signulll · 2026-11-02
- NYU 教授 Togelius 谈 AI 与游戏:EA FC 守门员竟是 RL 训练的 — ziv_ravid · 2026-11-02
- 学者批反蒸馏运动双标：自己用全网数据训练却禁别人学习 — StanfordAILab · 2026-11-02
- Stripe 数据揭示 AI 采用与增长的因果难辨：领先企业本就增长更快 — soumitrashukla9 · 2026-11-02
- 前员工回忆录揭早期 Anthropic：联创称战略只能靠市场与公民社会 — ZeroStateReflex · 2026-11-02
- tszzl 力挺《星际迷航》最高指导原则:反对被单一文明抹平 — tszzl · 2026-11-02