Monad-lab.LLC

Where Ancient Wisdom meets Future Intelligence. Official X account of the Monad-lab

[Monad-lab.LLC @Monad_lab]

@0xAA_Science 接Qwen就好

[Monad-lab.LLC @Monad_lab]

@sama I have tried it. It’s awesome!

[Monad-lab.LLC @Monad_lab]

@OpenAIDevs 这个是至今为止看到最符合我需要的一个remote，建议所有还沉迷什么openclaw或者hermes的都用一下，我觉得好像毫无必要用另外的agent了。

[OpenAI Developers @OpenAIDevs]

Step away from your laptop. Keep building with Codex on your phone.

Codex keeps working on your computer, with your files and project context still in place.

Pocket-sized access. Full Codex working state.

https://t.co/vGsw5wnHnG

[Monad-lab.LLC @Monad_lab]

Block 裁掉4,000人，可能只需要40个AI架构师就能补回原来的产出。1:100的替换比，以前的工业革命里没出现过。

如果AI创造的新需求最终还是由AI完成，利润只会流向算力的拥有者，不会变成普通人口袋里的工资。GDP或许还在涨，但中间那条把"生产力增���"变成"人类收���"的管道，断了。

GIC文章的最后一句话是："金丝雀还活着。"

但在我写下这些文字的时刻，至少有一只已经不再歌唱了。

Block cut 4,000 jobs. It probably needs only 40 AI architects to match or exceed the original output. A 1:100 replacement ratio — unprecedented in any prior industrial revolution.

If the new demand AI creates is ultimately fulfilled by AI itself, the profits flow back to the owners of compute, not into ordinary workers' paychecks. GDP may keep growing, but the pipe that converts "productivity gains" into "human income" is broken.

The last line of the GIC article: "The canary is still alive."

As I write these words, at least one has stopped singing.

[Monad-lab.LLC @Monad_lab]

昨天，Block（https://t.co/yiPUHDZTU2）裁员 40%。

不是因为亏钱——2025年毛利超过100亿美元，增长17%。

是因为CEO Jack Dorsey认为，6,000人配合AI，能干10,000人的活。

市场的反应？股价涨了24%。

算下来，每裁掉一个人，公司市值增加约150万美元。

这个信号会传遍每一间董事会。🧵👇

Yesterday, Block (https://t.co/yiPUHDZTU2) laid off 40% of its workforce.

Not because it was losing money — 2025 gross profit exceeded $10B, up 17%.

Because CEO Jack Dorsey believes 6,000 people + AI can do what 10,000 did.

The market's response? Stock jumped 24%.

That's roughly $1.5M in market cap created per job cut.

This signal will reach every boardroom on earth. 🧵👇

[Monad-lab.LLC @Monad_lab]

美国住宅房贷市场，13万亿美元。每笔房贷的前提假设：借款人在未来30年大致维持当前收入就业。

AI正在挖这个假设的根。

跟2008年的区别——2008年贷款发出去那天就是坏的，今天这些借款人信用分780+，首付20%，贷款发放那天是好的。只是，世界在贷款之后变了。

"人们是按照一个他们再也负担不起去相信的未来，借的钱。"

一定有人会反驳：历史上每次技术革命都先消灭旧工作，再创造更多新工作。蒸汽机、汽车、互联网，两百五十年来这个论点一直在赢。

GIC�����了一句：历史上每一次，新创造出来的工作都需要人来做。👇

The U.S. residential mortgage market: $13 trillion. Every loan underwritten on one assumption — the borrower will remain employed at roughly their current income for 30 years.

AI is eroding that assumption at its root.

The difference from 2008: in 2008, the loans were bad on day one. Today's borrowers have 780+ FICOs, 20% down. The loans were good on day one. The world just… changed after.

"People borrowed against a future they can no longer afford to believe in."

The inevitable counterargument: every tech revolution destroys old jobs but creates more new ones. Steam, automobiles, the internet — this argument has been winning for 250 years.

The GIC adds one line: every time before, the new jobs needed humans to do them. 👇

[Monad-lab.LLC @Monad_lab]

美国住宅房贷市场，13万亿美元。每笔房贷的前提假设：借款人在未来30年大致维持当前收入就业。

AI正在挖这个假设的根。

跟2008年的区别——2008年贷款发出去那天就是坏的，今天这些借款人信用分780+，首付20%，贷款发放那天是好的。只是，世界在贷款之后变了。

"人们是按照一个他们再也负担不起去相信的未来，借的钱。"

一定有人会反驳：历史上每次技术革命都先消灭旧工作，再创造更多新工作。蒸汽机、汽车、互联网，两百五十年来这个论点一直在赢。

GIC只补了一句：历史上每一次，新创造出来的工作都需要人来做。👇

The U.S. residential mortgage market: $13 trillion. Every loan underwritten on one assumption — the borrower will remain employed at roughly their current income for 30 years.

AI is eroding that assumption at its root.

The difference from 2008: in 2008, the loans were bad on day one. Today's borrowers have 780+ FICOs, 20% down. The loans were good on day one. The world just… changed after.

"People borrowed against a future they can no longer afford to believe in."

The inevitable counterargument: every tech revolution destroys old jobs but creates more new ones. Steam, automobiles, the internet — this argument has been winning for 250 years.

The GIC adds one line: every time before, the new jobs needed humans to do them. 👇

[Monad-lab.LLC @Monad_lab]

GIC 把这个叫"智能置换螺旋"：

AI更强 → 裁员 → 利润涨 → 股价涨 → 其他公司跟 → 更多裁员 → 白领消费缩减 → 企业收入下滑 → 为保利润继续AI化 → 更多裁员…

这个环没有自然刹车。

传统周期里，需求下滑会拖累投资，经济慢慢触底。但AI的成本逻辑相反——需求越弱，企业越需要用廉价AI替掉昂贵的人。衰退本身加速了衰退的原因。

GIC还造了一个词："幽灵GDP"——AI产出在涨，利润在涨，GDP在涨，但钱不再经过人类的口袋。数字繁荣，体感衰退。

"机器的年度消费预算是多少？零。"👇

The GIC calls it the "Intelligence Displacement Spiral":

AI improves → layoffs → margins expand → stock rises → others follow → more layoffs → displaced workers spend less → companies lose revenue → cut more with AI → more layoffs…

No natural brake.

In a normal cycle, falling demand drags investment down, the economy bottoms, recovers. But AI's cost logic runs in reverse — weaker demand means more pressure to replace expensive humans with cheap AI. The recession accelerates its own cause.

The GIC coined a term: "Ghost GDP" — AI output grows, profits grow, GDP grows, but the money no longer passes through human households. Digital prosperity, felt recession.

"How much do machines spend on discretionary goods per year? Zero." 👇

[Monad-lab.LLC @Monad_lab]

@Koki_Shimazu https://t.co/0AcK2gWQzp

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

@FurkanGozukara https://t.co/0AcK2gWQzp

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

@XH_Lee23 https://t.co/0AcK2gWQzp

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

@BRICSinfo https://t.co/0AcK2gWQzp

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

@UnitreeRobotics https://t.co/TwkwNAUbpI

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

“结构智能”的完整思考，以及关于机器人如何从“生产力溢出”走向“定义权竞争”的深度分析，原文发在公众号了。
👇 更多细节见文末链接：

https://t.co/tqWp7uyyX3

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

进化路径——“练”出来的记忆

这种招式，人写不出来。

核心在于深���强化学习 (RL)。AI 在虚拟环境里摸索数万小时，自行掌握对抗重力的发力技巧。将这套“肌肉记忆”植入硬件，机器人便拥有了理解物理规律的“大脑”与执行逻辑的“小脑”。

"Learned" Memory, Not Written

These intricate moves are impossible for a programmer to write manually.

The secret lies in Deep Reinforcement Learning (RL). The AI explores a virtual environment for tens of thousands of hours, mastering the physics of balance. Once this "muscle memory" is embedded, the robot gains a "Brain" to understand physics and a "Cerebellum" to execute logic.

[Monad-lab.LLC @Monad_lab]

@TansuYegen https://t.co/TwkwNAUbpI

[Monad-lab.LLC @Monad_lab]

春晚的《武BOT》机器人刷屏了，但别被整齐划一的动作迷惑。

这不只是“程序预设”，而是结构智能的集中爆发。在醉拳这种重心剧烈偏移的动作中，机器人能不倒，靠的不是程序员写死的代码，而是自进化出来的“运动小脑”。

The "Wu-BOT" performance at the Spring Festival Gala was mind-blowing, but don't be fooled by the synchronized spectacle.

It’s not just a "pre-programmed" routine; it’s a massive demonstration of Structural Intelligence. In movements like Drunken Boxing where the center of gravity shifts violently, the robot stays upright not through hard-coded lines, but via an evolved "Motor Cerebellum."

[Monad-lab.LLC @Monad_lab]

我对“结构智能”的完整思考，以及关于机器人如何从“生产力溢出”走���“定义权竞争”的深度分析，原文发在公众号了。
👇 更多细节见文末链接：
https://t.co/tqWp7uyyX3

[Monad-lab.LLC @Monad_lab]

实时反馈

如果只是重复预设动作，机器人撑不过三秒。

传感器必须在毫秒级感知倾斜，底层算法同步计算关节出力，补偿误差。行云流水的“精气神”，本质上是硬件爆发力与算法结构对物理世界的实时接管。

Real-time Feedback

If it were just repeating presets, these robots wouldn't last three seconds.

Sensors must detect tilt in milliseconds, with underlying algorithms calculating joint torque in real-time to compensate for errors. The fluid "spirit" of the martial arts is essentially the hardware's explosive power being governed by the algorithmic structure in real-time.