对Codex的期待与担忧
我很高兴大家都在转向Codex，但Tibo，如果你开始限制我的使用频率或让我使用更差的模型……
I am happy everyone is switching to Codex, but Tibo if you start rate limiting me or making me use worse models...
AI 领袖动态
2026-04-17 · 41 条
Meta与AI的转折
在Meta与AI的宏大叙事中，我们经历了Llama 4的失败，现在我们看到转机：自2025年的“汤战争”以来，招聘人数增加；Zuck实际上与Alexandr和Nat同住，并再次开始编码；终于推出了Opus级别的模型（没有API，不开放，但链接在此）。
in the grand narrative of Meta x AI, we saw the flop (Llama 4 hurhurhur), and now we’re seeing the turn:
- *more* hiring since the soup wars of 2025
- Zuck literally moved in with Alexandr and Nat and is koding again
- finally GA’ed Opus-ish level model (no api, not open, but https://t.co/4KHWPuTAQ9
AI工具将颠覆设计行业
Anthropic或其他AI原生工具可能会颠覆设计工具以及Google Workspace/Microsoft Office。这些工具大多是为人类逐个手工制作幻灯片、文档和电子表格而设计的。现在可以通过理解你的代理来生成这些内容。
Anthropic or some other AI native tool will likely disrupt design tools and Google Workspace / Microsoft Office.
Most of these tools were designed for humans to hand craft slides, docs, and spreadsheets one by one.
Getting an agent who understands you to generate these
更新集中管理
我很高兴能将我们的更新集中在一个地方！
Excited to centralize our updates in one place!
项目进展加速
50%折扣——特别适合运行并行代理，加快项目进展！
50% off -- especially useful to run parallel agents and make faster progress on your project!
部署到欧盟
部署到欧盟！
Deploy to EU!
未来的生活
很容易忘记自己生活在未来，但偶尔会看到这样的事情……通过手机回应客户的投诉，进行重大产品更改（本地化），同时在自动驾驶汽车的后座与软件代理交谈。
It's easy to forget that you're living in the future. But every now and then you see something like this...
Responding to a client's complaint by making a major product change (localization) from your phone, talking to your software agent in the back of a self-driving car
代理与后端的挑战
代理和后端最难的事情是耐用性。@workflowsdk解决了这个问题。你调用的那个LLM肯定会宕机；那个服务肯定会限制你的使用频率；那个数据库肯定会意外变慢；你肯定会收到警报💀。我已经寻找独角兽十年了。
The hardest thing about agents and backends is durability. @workflowsdk fixes this.
That LLM you're calling *will* go down. That service *will* rate limit you. That database *will* unexpectedly slow down. You *will* get paged 💀
I've been looking for a unicorn for a decade. I https://t.co/vkvgBMpk1W
AI进步的速度
AI的进步速度令人不懈。你可以通过@aisdk和@vercel AI Gateway捕捉其波动带来的好处。祝贺@anthropicai又一成功发布，但@xai、@openai和@googleai也在赶来。今年将会很有趣。
The rate of progress in AI is relentless. You can capture the upside of its volatility with @aisdk and @vercel AI Gateway.
Congrats to @anthropicai on another banger ship, but @xai, @openai, and @googleai are coming. Gonna be a fun year.
Opus 4.7的亮点
Opus 4.7中我最喜欢的一些功能：- 非常擅长异步工作和遵循指令；- 令牌控制的努力水平更加可预测（+ 新的xhigh级别）；- 不再缩小高分辨率图像；- 在用户界面、幻灯片和文档方面明显更有品位。
Some of my favorite things in Opus 4.7:
- Very good at async work and following instructions
- Effort levels are far more predictable for token control (+ new xhigh level)
- No more downscaling of high-res images
- Noticeably more taste in UIs, slides, docs
新的Codex将改变知识工作者的工作方式
新的Codex将使代理能够编写代码、使用工具和计算机，从而在各个工作领域后台执行长期任务，如撰写报告和设置数据室。
The new Codex is another jump in what agents will look like for knowledge workers. Agents that can code, work with tools, and use computers, can begin to execute long running tasks in the background for all areas of work.
This can mean drafting reports, setting up data rooms https://t.co/H4gaOU47yy
AI将创造更多工作机会
AI将加速某一领域的产出，但最终在流程中会出现新的瓶颈，仍需人类介入。FT的一个例子清楚地说明了这一点。
Why will AI create more jobs in plenty of industries? It’s because we’re going to use AI to accelerate output in one area, and then eventually you run into a new bottleneck somewhere else in the process that still requires humans.
This example from the FT is an obvious one. More https://t.co/rIx2b4EN6M
开源项目GBrain即将发布
我可能会将我的项目GBrain作为开源软件发布。
Hi, I might give it away as open source (it's called GBrain)
https://t.co/yFpFU4pn5b
左翼对暴力事件的轻描淡写
旧金山的极左派将暴力事件淡化为简单的破坏行为，如果极端的切萨·布丁仍然是地区检察官，可能连严重指控都不会被追究。
I will say the hard left contingent of SF is going to downplay violence as simple vandalism and if extremist Chesa Boudin were still DA there likely wouldn’t even be serious charges being pursued
语音助手的未来已来
我在台北尝试智能调度的语音助手时，STARLUX的空乘人员立即询问这是什么，是否可以使用。她是电影《她》的粉丝，未来的语音技术让我们离电影中的场景越来越近。
I was trying out my voice agent re smart scheduling and the STARLUX flight attendant in Taipei immediately asked what is that and can I have it yet?
She is a fan of the movie Her.
The future becomes obvious with voice. We are so close to what we saw in the movies.
资产扣押税的误导性言论
资产扣押税的起草者发表了许多基本上是无知的谎言，这些学者缺乏商业现实经验。愚蠢的记者继续引用他们，这是一种新闻失职。
The asset seizure tax drafters say a lot of things but they’re all basically uninformed lies by academics who have no experience in business reality
That they keep being quoted by gullible reporters is a miscarriage of journalism
开源软件在新神话时代更安全
在新的神话时代，开源软件的安全性将远超闭源软件。
Open source software will be many times more secure than closed source software in the new Mythos era
前往新加坡的旅程
我正在前往新加坡的路上，STARLUX在台北的WiFi非常好，因此预计在接下来的24小时内会有大量GStack和GBrain的bug修复和新功能发布。
I'm en route to Singapore and there's a ton of good wifi on Starlux via Taipei, so expect a lot of GStack and GBrain bug fixes and features dropping the next 24 hours https://t.co/qp2SG7BGCE
支持Phil Kim就是支持常识
Phil Kim竞选旧金山学校委员会，他的对手曾公开我的家庭住址，导致我家不得不搬家，只因她想要让我沉默。为了什么？她的虚伪信号议程正在破坏旧金山每个公立学校孩子的教育。投票支持Phil Kim。
Phil Kim for SF School Board is a vote for common sense
His opponent literally doxxed my home address and had to move my family because she wanted to silence me
For what? Her virtue signal agenda to destroy the educations of every public school kid in SF
Vote for Phil Kim
令人印象深刻的软件工厂
这是一个令人印象深刻的软件工厂。
This is an impressive software factory actually
他们还不知道...
但他们会知道的！
They don't know yet...
But they will know!
Karpathy的混淆协议已加入GStack
Karpathy指出，#1人工智能编码失败模式是在模糊决策点上，代理自信地选择错误的路径。这会导致10分钟的工作白费，必须重新开始。现在，gstack在每个https://t.co/6HkrDXXt1T中内置了一个模糊性门。
Karpathy's Confusion Protocol is now in GStack
Karpathy called it: the #1 AI coding failure mode is the agent confidently picking the wrong path at an ambiguous decision point. You lose 10 minutes of work and have to start over.
gstack now has an ambiguity gate built into every https://t.co/6HkrDXXt1T
反对人工智能的力量真实且有组织
即使我们讨论的是有帮助和实用的工具，反对人工智能的力量依然存在并且组织严密。
Anti-AI forces are real and organized, even when we are talking about helpful and useful tools
...
...
https://t.co/0HZS9Yyhla
Acts 17将于4月30日来到纽约
将与@kareemszaki和@saquon进行关于真正重要的事情的对话。不要错过！https://t.co/n0vj8bt36J https://t.co/BReTFUvH9R
Acts 17 is coming to NYC April 30 for a convo with @kareemszaki and @saquon on what truly matters
Don’t miss it
https://t.co/n0vj8bt36J https://t.co/BReTFUvH9R
Claude Code CLI表现出色
但网页和桌面视图就有点不尽如人意。
The Claude Code CLI is 🔥 but the web and desktop views are uhhh
2008年的人无法想象Markdown的重要性
如果你告诉2008年的人Markdown在2026年将会有多重要，他们是不会相信的。
If you told someone from 2008 how important Markdown would be in 2026 they wouldn’t believe you
Gemini Voice真是超凡脱俗
它令人难以置信。即将登陆GBrain和你的OpenClaw/Hermes。
Gemini Voice really is a cut above. It’s incredible.
Coming to GBrain and your OpenClaw/Hermes shortly
我奶奶今天去世了
https://t.co/SE5gcLGun9
My grandma passed away today. https://t.co/SE5gcLGun9
GBrain与多个代理同时工作良好
由于使用git+postgres，GBrain在同时与多个代理协作时表现出色。
GBrain thanks to being git+postgres works wonderfully with multiple agents simultaneously
找到真正关心你的人
生活太短暂，不值得与那些不在乎你的人共事。寻找一个真正关心你的人，比如导师、创始人、风险投资者、顾问或客户，这可能会成为你最大的护城河。
Find someone who genuinely cares.
A mentor. A founder. A VC. An advisor. A customer.
Life’s too short to work with people who don’t give a sh!t.
It might end up being the greatest moat of them all.
ClawCon 密歇根州的盛况
在密歇根州的 ClawCon 看到近 2000 人真是太棒了！这是一个奇怪的龙虾文化聚会。
Amazing to see almost 2000 people at ClawCon Michigan!!
Weird lobster cult 🦞🥳
OpenClaw 的不安全性
有人说 OpenClaw 很不安全，看看这些 GHSAs！但实际上，我们只是即将到来的风暴的一个指示器。
they: OpenClaw is so insecure look at all these GHSAs!
reality: we are just an indicator of the coming storm
我们进行了哲学家选拔
如果每个模型实验室可以选择历史上的一位哲学家，他们会选择谁，为什么？
we ran a philosopher draft
which philosopher from history would each model lab hire if they could, and why? https://t.co/KDryCy15V1
神话公告的战略意义
如果神话公告是一个四维棋局的举动，让特朗普继续与 Anthropic 签订合同，那将是不可思议的。对此我持乐观态度。
it would be incredible if the mythos announcement was a 4d chess move to make Trump keep the anthropic contracts
bullish
Anthropic 的 Alex Albert 加入直播
Anthropic 的 @alexalbert__ 刚刚加入我们的直播 Opus 氛围检查，快来参与吧！
and @alexalbert__ from anthropic just joined our LIVE opus vibe check
get in here
https://t.co/rRXbhWoTUN
Opus 4.7 正式发布
Opus 4.7 刚刚发布，我们正在进行实时氛围检查！
Opus 4.7 just dropped and we're LIVE VIBE CHECKING it right now
https://t.co/kSWFPzLXeY
实时氛围检查：Opus 4.7 发布
实时氛围检查：Opus 4.7 正式发布！
LIVE VIBE CHECK: OPUS 4.7 DROPS https://t.co/rRXbhWoTUN
需要重写所有代码吗？
天哪，我现在需要用 4.7 重写我所有的代码吗？？
Oh god, am I going to have to rewrite all my code with 4.7 now????
LLM 怀疑论者的演变
LLM 怀疑论者的演变过程：1. 他们只是自动纠错；2. 他们只是下一个词预测器；3. 他们会产生幻觉；4. 他们只能对训练数据有用；5. 他们只适合编码；6. 他们只适合可验证的事情。承认吧！LLM 是...
The arc of a LLM skeptic:
1. They are just auto-correct
2. They are just next-token predictors
3. They hallucinate
4. They can only be useful for the data they are trained on
5. They are only good for coding
6. They are only good for verifiable things
Just admit it! LLMs are
LLMs的潜力仍在探索中
每天我都会发现有人能让大型语言模型（LLMs）做一些奇怪而有趣的事情。我必须不断提醒自己，我们正处于探索这种神奇技术可以构建什么的早期阶段。
Every day I find someone who manages to get LLMs to do something weird and interesting.
I have to keep on reminding myself that we are in the super early innings of figuring out what we can build with this magical technology.