团队的努力令人欣喜
我喜欢团队做这样的事情。
i love that the team does stuff like this
AI 领袖动态
2026-04-30 · 32 条
时间旅行的想法
我想象自己在说：“互联网被AI垃圾污染，我们需要低背景的代币。”然后我想，如果我们能时间旅行，看看100年前的祖先会对我们说什么，那该多好。现有的复古模型都小于4B，我们需要一个调优过的13B复古模型。
> be me
> "the internet is polluted by ai slop, we need low-background tokens"
> "wouldnt it be cool if we could time travel and see what our ancestors 100 years ago would say to us"
> all the existing vintage models are like <4B
> we need a chat tuned 13B vintage model
>
Gemini的新功能
Gemini新增功能：生成文件并导出。只需告诉Gemini您想创建的内容和格式，它现在可以为您完成这项工作。现在支持Google Docs、Word (.docx)、PDF、Google Sheets、Excel (.xlsx)、CSV、Google Slides、Markdown、LaTeX、TXT和RTF。
New in Gemini: Generate files and export them
Tell Gemini what you want to create and the format, and it now does the work for you.
Now supporting:
📄 Google Docs, Word (.docx) & PDFs
📊 Google Sheets, Excel (.xlsx) & CSV
🖥️ Google Slides
🛠️ Markdown, LaTeX, TXT, RTF
Available https://t.co/4WKIiMJwKb
放射科医生的真正职责
放射科医生的工作并不是解读X光片，而是治愈病人。如果AI能加速理解X光片的过程，放射科医生就能看到更多病人并治愈他们。
It turns out, the job of a radiologist isn't to read x-rays.
It's to cure people.
And if AI can help speed the process of understanding x-rays, radiologists can see and cure more people.
关于@openclaw的最新动态
我没有跟上@openclaw的最新更新——现在是否可以进行低延迟的实时通话？
I haven't kept up to date with the latest @openclaw updates - is live low-latency calling with your claw now possible?
@claudeai的响应延迟
不知道@claudeai发生了什么，但我发现现在使用基本的网页界面时，常常需要随机等待3-5秒才能得到响应。
Don't know what's going on with @claudeai but I find that I often now have to wait 3-5 seconds randomly for it to respond just using the basic web UI.
个人助手的聪明之处
这对个人助手来说实际上很聪明，@Shpigford——我现在要把这个想法用在我的OpenClaw上。
This is actually brilliant for personal agents @Shpigford - stealing this for my OpenClaw now. https://t.co/gghJwyIA1W
对独立AI开发者的尊重
我非常尊重独立的AI开发者。现在世界是他们的游乐场。
I have alot of respect for solo AI builders.
The world is their playground now.
晚餐时被通知网站故障
现在每个人都可以体验到在晚餐时因为网站故障而被通知的感觉😅。
Now everyone can experience getting paged at dinner because the site is down 😅 https://t.co/BOp9OmnhQb
教师和学生的优惠
教师免费账户，学生享受50%的折扣！
Free accounts for teachers, 50% off for students!
Replit Slides 制作
使用 Replit Slides 制作的内容。
Made with Replit Slides.
维护应用程序的问题
“是的，代理可以帮助你构建应用程序，但谁来维护它们？”同时，代理们也在继续工作。
“Yeah agents can help you build apps, but who maintains them?”
Also agents:
软件工作不会消失
我会一直强调，软件工作不会消失。代理是历史上技术人员最大的杠杆工具。现在是技术人员能够独立、团队或公司完成工作的最佳时机。
Will keep saying this, but software jobs aren’t going away. Agents are the single biggest form of leverage for anyone technical in history. Probably has never been a better time to be technical in terms of being able to accomplish something solo, in a team, or company.
We think
加入 YC 论文俱乐部
加入 YC 论文俱乐部。
Join the YC Paper Club
旧金山教育局的审查问题
旧金山统一学区为其民族研究课程支付了 14.7 万美元进行“独立”审查，但审查中充满了内部人士，没有通过标准，并且压制了批评声音。尽管如此，董事会还是投票批准了这一审查。即将面临诉讼。
SFUSD paid $147K for an "independent" review of its ethnic studies curriculum that was stacked with insiders, had no passing threshold, and silenced critics. The Board voted to approve it anyway.
Litigation incoming.
https://t.co/W8pnmgsm1v
教师裁员与无用的 PowerPoint
旧金山统一学区正在裁掉真正的教师，以优先支付六位数的费用购买无用的 PowerPoint，这些内容对旧金山的孩子们没有帮助，公众应该知道这一点并据此投票。
San Francisco Unified School District is laying off real teachers to prioritize paying 6 figure sums to buy worthless PowerPoints that don't help the kids of San Francisco, and people should know this and vote accordingly
大脑的样本效率
大脑的样本效率并非来自架构，而是来自损失函数。皮层下结构编码了丰富的评估信号，使学习比在权重上进行原始梯度下降更有效。AI 代理也可能如此，正在进行实验。
The brain's sample efficiency doesn't come from architecture. It comes from loss functions.
Subcortical structures encode rich evaluation signals that make learning more efficient than raw gradient descent on weights.
Same thing could be true for AI agents. Experimenting now
在其他领域的实用性
很高兴看到它在其他领域也有用。
Cool to see it is useful in other areas
代理的未来
如果你厌倦了听到代理的消息，2026 年将会是你永恒的九月。
If you’re sick of hearing about agents 2026 is going to be the eternal September for you
这个真的很搞笑
这种方式真的很搞笑。
kind of hilarious this works https://t.co/z1JYNjn65m
风投晚宴的标志
你知道自己在风投晚宴上时的感觉。
You know you’re at a VC dinner when https://t.co/OF6mSVNdY6
小“代理”帮我接触自然
我的小“代理”在为我接触大自然。🌻
Having my little “agent” touch grass for me 🌻 https://t.co/AR1uLiQRvf
30岁时的追求
30岁时的目标是拥有一个浆果农场和一个HVAC公司。
30s are all about trying to own a berry farm and an HVAC rollup
你听我说了吗？
你听我说了吗，匿名者？
Did you listen to me anon? https://t.co/NcxfP9hB2t
代币的真实价值
人们认为代币很贵……这是@useblacksmith（他们赞助OpenClaw，🫶🦞）。
And people think tokens are expensive... this is @useblacksmith (they sponsor OpenClaw, 🫶🦞) https://t.co/pCcG5uzbhd
Codex的自动修复机制
Codex现在在每次提交时运行，进行审查。如果发现错误，会启动一个新的Codex并（如果仍然相关）提交修复请求。然后启动一个审查代理，如果发现问题，另一个代理会修复这些问题（最多循环5次）。
codex now runs on each commit we land, reviews it - and if a booboo is found, a new codex spins up and (if still relevant) makes a PR for the fix.
Then a review agent spins up. If an issue is found, another agent will fix the issues. (up to 5 loops) https://t.co/k6o7tNcDyf
关于代理的一些观察
从大量时间花在安装、使用、调试和构建OpenClaw、Hermes等产品中，我有一些观察。大多数对OpenClaw、HermesAgent等赞不绝口的人实际上并没有使用过它们。事实是，代理现在是开发者产品。
Some observations on agents from a lot of time spent installing, using, debugging and building things with OpenClaw, Hermes, etc.
- Most people who rave about OpenClaw, HermesAgent etc. have not actually used it. Facts.
- Agents are developer products today. They are so so so
无法形容的精彩
我无法形容这有多棒。
I cannot explain how awesome this is.
Bevel 3.0的目标
Bevel 3.0的目标从来不是成为一家追踪和分析公司。每次发布都让我们更接近长期目标：成为最智能的健康和生物伴侣，真正理解你并帮助你过上更健康、更快乐的生活。
Bevel 3.0
The point was never to be a tracking and analytics company.
Each release of @bevel brings us closer to our long-term goal: To be the most intelligent health and biology companion that truly understands you and can help you lead a healthier and happier life.
It was https://t.co/kpufUIdS0A
人类进步的新里程碑
人类进步不仅仅是萨维打破两小时的记录，而是两个人做到了人类从未做到的事情。
Human progress is not just Sawe breaking the 2-hour mark but that TWO people did something that no human had ever done.
迈向后提示时代
我们正接近一个后提示的世界。
We're approaching a post-prompting world.
如何理解旧金山的现状？
我们正经历过去20年来最疯狂的计算革命，而在旧金山4英里半径内创造的价值占95%（感谢@eladgil）。这让人难以理解，为什么会有如此高的集中度？这背后有什么原因？
How does one make sense of SF?
We are living through the craziest computing revolution of the last 2 decades....and 95% of the value being created is within a 4-mile radius in SF (h/t @eladgil )
It's just hard to understand....why is the concentration so high?
What is it about