Skip to content
AGI Dawn
Research
Explore
Overview
All research
Deep reads
The canon
Series: From Attention to Agents
Glossary
Frontier updates
What we publish
RSS
Product
Explore
Overview
Claude Code
Models
Top up
Teams & enterprise
Announcements
Console →
Developers
Explore
Developer home
Docs
Claude Code guide
Pricing
Company
Explore
About
Contact
Terms of service
Privacy policy
中文
Log in
Get API key
Open console
Research
All research
Deep read
All
Deep read
Pick
Series
Review
Foundations
Oct 16, 2026
22 min read
注意力就是全部:Transformer 如何改写了序列建模
S
Chinese
它把「按顺序读」换成「同时看全句、按相关性取信息」,让训练可以完全并行;此后所有大模型都长在这副骨架上。
Deep read