SchmidhuberAI

6 atoms · 跨 6 天 · 首见 2026-04-30 · 最近 2026-06-29

三色: 🟦 fact 3 · 🟥 take 3 · stance ▲0/▼0/◆5

来源: X 6

时态: fresh:6

标签: 好信源:4 · 好观点:2

叙事 (narrative) (2)

• 2026-06-29 · tweeted about 1991 ULTRA in 2022 with quote about linearized self-attention
  30 years ago: Transformers with linearized self-attention in NECO 1992, equivalent to fast weight programmers (apart from normalization), separating storage and control. Key/value was called FROM/TO.
  tags: 好信源 · → daily
• 2026-06-21 · 2017 normalized quadratic Transformer [TR1] is based on the principles of the 1991 unnormalized linear Transformer [ULTRA]
  Google's 2017 normalized quadratic Transformer [TR1] is based on the principles of the 1991 unnormalized linear Transformer [ULTRA].
  tags: 好信源 · → daily

反驳 (refute) (2)

• 2026-06-25 · refutes that ULTRA is an extension of Gauss-Legendre method
  No, it isn't. Here is the Gauss-Legendre page: https://t.co/udI7EOdUcS
  tags: 好观点 · → daily
• 2026-06-24 · refutes that ULTRA has little or nothing to do with recursive least-squares
  No, it has little or nothing to do with that - and here's the true Gauss-Legendre story (1795-1805) you are referring to: https://t.co/Nayn0k9K4r
  tags: 好观点 · → daily

事实 (fact) (2)

• 2026-06-19 · 1991年团队成果
  I am proud of the work my team did in Munich in 1991, when compute was millions of times more expensive. We published the roots of today's trillion-dollar AI boom: ★ 3/1991: the first kind of Transformer (see the T in ChatGPT) - now called the unnormalized linear Transformer: the predecessor of the normalized quadratic Transformer ★ 4/1991: Pre-Training (the P in ChatGPT) & Neural Net Distillation (see DeepSeek and many other LLMs) ★ 6/1991: Deep Residual Learning, basis of LSTM & Highway Net / ResNet (most-cited AIs of their centuries) ★ 8/1991: conference paper on GANs for World Models trained by Artificial Curiosity
  tags: 好信源 · → daily
• 2026-04-30 · 发布 Neural World Model Boom 技术笔记
  JS. The Neural World Model Boom. Technical Note IDSIA-2-26, 2/4/2026, updated 4/29/2026
  tags: 好信源 · → daily
  value: qty=1 · date=2026-04-29

时间轴 (近 20)

• 2026-06-29 · narrative · tweeted about 1991 ULTRA in 2022 with quote about linearized · → daily
• 2026-06-25 · refute · refutes that ULTRA is an extension of Gauss-Legendre method · → daily
• 2026-06-24 · refute · refutes that ULTRA has little or nothing to do with recursiv · → daily
• 2026-06-21 · narrative · 2017 normalized quadratic Transformer TR1] is based on the · [→ daily
• 2026-06-19 · fact · 1991年团队成果 · → daily
• 2026-04-30 · fact · 发布 Neural World Model Boom 技术笔记 · → daily

← 实体目录 · 系统日志