Cui Bo
3.6K posts

Cui Bo
@zcsoft
Indie AI Builder · 独立AI开发者 🧑💻 Shipping while traveling · 边旅行边发布产品 ⛵️
เข้าร่วม Aralık 2009
549 กำลังติดตาม434 ผู้ติดตาม
Cui Bo รีทวีตแล้ว

Google DeepMind just published the most important AI paper of 2026.
It's called "From AGI to ASI." Marcus Hutter is on it. Iason Gabriel is on it. 14 of the most senior people at Google DeepMind are on it. They published it 3 days ago and almost no one is talking about it.
We have spent the last three years obsessing over when AGI will arrive.
DeepMind is already past that. They are modeling the transition to Artificial Superintelligence (ASI).
And their definition of ASI isn't just an AI that is smarter than a human.
It is a system that is more intelligent and cognitively capable than entire, large-scale human organizations. Not a single genius. An institutional-level intellect.
The report is 57 pages long. It outlines four exact pathways we will take to get there:
- Scaling AGI: Pumping incomprehensible amounts of compute and data into current models.
- Paradigm Shifts: New algorithmic breakthroughs that replace the architectures we use today.
- Recursive Self-Improvement: The AI learns to write better code and upgrade its own intelligence without human help.
- Multi-Agent Collectives: Millions of AI agents coordinating together to form an emergent hive-mind Superintelligence.
But here is the detail that should make you pause.
DeepMind points out that digital intelligence has massive asymmetric advantages over human biology. It can be perfectly copied. It doesn't sleep. Its processing speed isn't constrained by biological neurons.
Because of this, they warn that the arrival of AGI won't be a single, stable milestone where we can catch our breath.
It will trigger a rapid, cascading series of transformative breakthroughs as the AI accelerates itself toward Superintelligence.
The timeline has shifted.
The most advanced AI lab in the world is no longer just researching how to build AGI.
They are mapping out what AGI will build next.

English

@TheoWang1005 谁说 muti agent 是用来对话的?可以用 sub agent 和 main agent 做上下文隔离。比如独立cr,测试,审稿等。
中文

@jinchenma_ai 尝试过各种skills和muti agent等等,说实话我感觉不如只输入提示词,然后回复时候刷手机,刷完回来输下一条提示词。一次搞太多基本意味着要花N倍时间返工。
中文



















