S
Swyx
Latent Space 联合创始人
2026-04-05 20:37
开源Dropbox项目Locker发布
上周，@swyx 让我开始构建一个开源Dropbox。Locker是终极的开源Google Drive/Box/Dropbox替代品，支持多种存储提供商，具备虚拟文件系统和QMD搜索插件功能。
↩ 转发自 @zachmeyer
Last week @swyx nerd-sniped me into building an Open-source Dropbox.
Here is Locker: the ultimate open-source Google Drive/box/Dropbox alternative 💾
- Provider agnostic (S3, R2, vercel blob, local)
- BYOB (Bring your own bucket)
- Virtual file system
- QMD Search plugin https://t.co/tiy1sKN3Dd
S
Swyx
Latent Space 联合创始人
2026-04-05 15:21
设计师的“Coachella”风格公告
我一直想做一个“这是Coachella吗”的公告——这是我们设计师对此的看法！
always wanted to do one of those “is it coachella” announcements - here is our designer’s take on it!
S
Swyx
Latent Space 联合创始人
2026-04-03 05:29
https://t.co/OIztRkxNNa
https://t.co/OIztRkxNNa
K
白宫重视中美生物科技竞争
白宫终于认真对待中美生物科技竞争，FDA提出了一条加快的IND通道，将缩短首次人体试验的时间。这对美国生物科技来说是个好消息，虽然来得有些晚。
↩ 转发自 @SebastianCaliri
It looks like the White House is finally taking US-China biotech competition seriously. The FDA is putting forward an expedited IND pathway that would shorten time to first-in-human.
Overdue, but good news for American biotech. https://t.co/KY1KaZLt4R
P
OpenAI Codex团队的工作方式
OpenAI Codex团队很少写规格，通常只有10个要点。与@embirico和@romainhuet的最新节目中，他们分享了Codex团队的运作方式，包括如何在几秒钟内构建和团队的工作流程。
↩ 转发自 @petergyang
"We write very few specs on the Codex team. We're talking 10 bullet points and that's it."
Here's my new episode with @embirico and @romainhuet where they gave me an inside look at how OpenAI's Codex team operates:
→ Live demo: Building in seconds with Spark
→ How the team https://t.co/myhQMIq8YM
P
开源与专有技术的胜利案例
开源技术与专有技术的胜利案例有哪些？例如，在移动领域，Android占据了最大市场份额，但iOS仍然获得了大部分利润。似乎两者都能共赢？
What are some good examples of open source winning vs. proprietary tech?
In mobile for example, Android has the most market share but iOS still captures the majority of profits.
Seems like both can win?
P
OpenAI的规划策略
OpenAI不关注中期规划，@embirico（Codex产品负责人）表示，规划分为短期和长期，但中期太难。短期规划是指未来八周内的具体目标，团队可以围绕此目标进行工作。
Why OpenAI doesn't focus on roadmaps from @embirico (Codex product lead):
"At @OpenAI, you either plan near or long-term, but you never plan medium-term. It's just too difficult.
Near-term is up to eight weeks from now. What is a concrete thing you can rally a team to ship? https://t.co/TW8CnwqxCo
P
上海的生活观察
上海的全职保姆每月只需1500美元，私人厨师每小时7美元，这里对职业工作夫妻的支持很多。滴滴出行的费用大多数在3-5美元之间，任何东西都可以点外卖。
More observations from Shanghai:
1. A full-time, live-in nanny costs only $1,500/month and a personal chef costs $7/hour. There's alot of support for professional working couples here.
2. Didi (Chinese Uber) rides are $3-5 for most trips and you can order delivery for anything https://t.co/RH8YEQvAtv
P
无传统规格或路线图的交付方式
不再依赖传统规格或路线图的交付方式，已经成为现实。
↩ 转发自 @morganlinton
Shipping without traditional specs or roadmaps, no longer the way of the future.
Happening, now.
P
团队角色的模糊化
@embirico帮助我们思考团队的未来，表示所有角色正在模糊化，设计师可以做更多工程工作，工程师可以做更多设计，产品经理也可以参与更多构建，标签的意义正在减弱。
↩ 转发自 @justingray0
here is @embirico helping is think through the future of teams:
"all these roles are blurring together. a designer can do more engineering. an engineer can do more design. a PM can do more building." "labels are losing meaning"
"if you were a good PM but not that good at
P
Codex应用是市场最佳AI编码工具
Codex团队表示，他们希望实现与多个代理自然对话的体验，但开发过程中遇到了不少争议。
I think the Codex app is the best AI coding app in the market. But it almost didn't get built.
From the Codex team:
"We knew we needed to get to a point where it feels really natural to be talking to multiple agents.
But [the app] was actually quite contentious. The IDE https://t.co/Ub27pZYFWv
P
第一次尝试活虾配甜酒
这是我第一次将活虾放入甜酒中，然后吃掉它。
First time putting live shrimp in sweet alcohol and then eating it https://t.co/poIYdHDu2k
A
新的X API与Replit结合很有趣
@tannerlbraden利用新的X API和Replit构建了一个NASA阿尔忒弥斯II号任务的可视化，实时显示X的动态信息和统计数据。
New @X API + @Replit is lots of fun.
@tannerlbraden built a @NASAArtemis II mission visualization w/ live @X feed & stats. https://t.co/ptLIf58yCr
A
关于创业心态的精彩文章
这是一篇关于创业心态的精彩文章。
مقال رائع عن عقلية ريادة الأعمال
A
复活节快乐！
祝大家复活节快乐！
Happy Easter!
A
应用在App Store上遭遇阻碍
在App Store上运营四年后，苹果突然阻止了我们的更新，但我们的应用仍然有机增长，再次攀升至第一名！
After 4 years on the App Store, Apple suddenly blocked our updates but our app continues to organically grow and rose to #1 again! https://t.co/Bw99OKnWxl
G
Vercel团队积极倾听客户反馈
Vercel团队始终在@x上倾听并与客户互动，这些对话帮助我们塑造产品。Vercel让我们快速发布，而X提供了源源不断的用户反馈。感谢@simhskal和@riyvir！
The @vercel team is always listening and engaging with customers on @x. We shape the product as a result of these conversations.
Vercel gives us the velocity to ship, X gives us a nonstop stream of user feedback. Thanks @simhskal & @riyvir!
G
月球上的超现实物理旗帜
我与@v0一起创建了一个基于超现实物理的月球旗帜。该旗帜被模拟为由弹簧连接的粒子网格，受重力、风和空气阻力的影响。
Created a flag on the moon based with hyper-realistic physics with @v0: https://t.co/IQIN8kl5PI.
The flag is simulated as a grid of particles connected by springs (cloth simulation). It's influenced by gravity, wind, air resistance.
The initial approach v0 one-shotted was quite https://t.co/FFdPgqMyAC
A
AI模型处理工具的能力提升
随着AI模型在处理工具方面的能力提升，以及上下文窗口变得更大而不那么混乱，我们可以开始设计更像人类工作的代理，而不必用奇怪的技巧来弥补模型的局限性。
As AI models get better at handling tools, and as context windows get bigger without as much rot, you can start to design agents more similar to how people work instead of having to mitigate the model limitations with weird hacks.
For instance, even a year ago, if you were to https://t.co/WVc2YUhSi9
G
分享对@conductor_build的评价
（转发自@Niyxuis）分享我对@conductor_build的评价，接下来是@getsome_air、@agentastic等。
↩ 转发自 @Niyxuis
Sharing my review for @conductor_build (cc: @charlieholtz ) 🧵
Next up: @getsome_air , @agentastic and more. https://t.co/X6YpnplW6b
G
加州高铁的工作机会
加州高铁项目创造了良好的就业机会，但是否真的在建设高铁呢？
↩ 转发自 @micsolana
“加州高铁正在创造高薪工作”非常酷，但它真的在创造高铁吗？
“california high-speed rail is creating good-paying jobs” very cool but is it creating high-speed rail
G
设备上小模型的可能性
在iPhone 17 Pro上，@adrgrondin正在运行谷歌的Gemma 4 E2B，性能达到约40tk/s，针对Apple Silicon进行了优化。该模型在移动设备上实现了前沿的编码和数学功能，支持128K上下文，完全离线并具备思考模式。
↩ 转发自 @rohanpaul_ai
Incredible possibilities for on-device small models.
Here @adrgrondin is running Google’s Gemma 4 E2B on iPhone 17 Pro.
~40tk/s with MLX optimized for Apple Silicon
SOTA coding & math on mobile with 128K context. Fully offline with thinking mode.
https://t.co/sPK65qc7Bd
G
GLM-5模型的表现
GLM-5是一个拥有7440亿参数的开放权重模型，其表现与最佳的专有模型（Claude Opus 4.5，GPT-5.2）相当。论文记录了他们的研究过程，使用了强化学习，在“代理”环境中进行试验。
↩ 转发自 @burkov
GLM-5 is a 744-billion-parameter open-weight model from that performs comparably to the best proprietary models (Claude Opus 4.5, GPT-5.2).
The paper documents how they got there. They use reinforcement learning in an "agentic" setting where each trial might involve the model https://t.co/g1c6dT69cy
G
高自主性人群的免疫力
高自主性的人似乎对尴尬有奇怪的免疫力。被拒绝？这不是尴尬，只是数据收集。显得天真？这不是尴尬，只是你在修正信息不对称。打破小的社交规则？这也不是尴尬，大多数规则都是如此。
↩ 转发自 @Kpaxs
High-agency people seem to have this weird immunity to embarrassment.
Getting rejected? Not embarrassing, that’s just data collection.
Looking naive? Not embarrassing, that’s just information asymmetry you’re fixing.
Breaking minor social rules? Not embarrassing, most rules
G
开源的黄金时代
开源的黄金时代已经到来。
Golden age of open source is here
G
合法化自动驾驶汽车
应该合法化自动驾驶汽车。
Legalize self driving cars
G
了解真实情况的最佳方式
了解当前真实情况的最佳方式是收听最新的@steipete和@lexfridman的采访。这是对代码生成领域极限可能性的完美概括。智能个人软件已经到来，只是分布不均！
Probably the best way to catch up to what is really going on is listen to the newest @steipete @lexfridman interview
It’s a perfect encapsulation of the extreme edge of what is possible now in codegen
Just in time smart personal software is here, just not evenly distributed! https://t.co/ZmOl58Aiq8
G
应用程序的意外成功
在App Store上四年后，苹果突然阻止了我们的更新，但我们的应用仍然在自然增长，再次升至第一名！
↩ 转发自 @amasad
After 4 years on the App Store, Apple suddenly blocked our updates but our app continues to organically grow and rose to #1 again! https://t.co/Bw99OKnWxl
G
计费方式的问题
根据系统提示中包含的文本进行不同计费的做法看起来非常糟糕。
↩ 转发自 @simonw
Billing different based on text contained in the system prompt is a really bad look
G
软件工厂的建设
我们正在各地建设软件工厂。
Many such cases
We're building software factories everywhere
G
现在是拥有高自主性的最佳时刻
我坚信，现在是历史上拥有高自主性的最棒时刻。你可以随心所欲地去做事情。
I am absolutely convinced this is the most awesome moment in history to have high agency
*You can just do things*
G
OpenClaw 设置建议
（转发自 @AlexFinn）如果你使用 Claude 订阅与 OpenClaw，请注意：与 Opus 相比，其他所有 AI 模型在 OpenClaw 上表现都很糟糕。这是事实，任何否认这一点的人都是妄想。因此，这是我推荐的新的 OpenClaw 设置：支付 Opus API 的费用。
↩ 转发自 @AlexFinn
If you used a Claude subscription with OpenClaw, read this:
Unfortunately all other AI models out there absolutely suck with OpenClaw compared to Opus
It's just a fact and anyone denying this is delusional
So here is my new recommended OpenClaw setup:
Pay for the Opus API and
G
复活节快乐
复活节快乐。他复活了。
Happy Easter. He is risen.
G
我们一起变得更快更好
我们都在一起变得更快、更好地创造事物。
We are all getting faster and better at making things together
G
这样的案例很多
许多类似的案例。
Many such cases
G
边界何时停止移动？
这似乎真的很糟糕——边界何时停止移动？Anthropic 只允许有真实人类按下回车键的订阅？你还需要通过 FaceID 验证吗？
This seems messed up actually - when do the boundaries stop moving?
Anthropic only allows subscriptions with a real human pressing enter? You're going to have to verify with FaceID?
G
每个组织都需要世界智能
（转发自 @ericosiu）每个组织都需要世界智能才能以 10 倍的水平运作。虽然这看起来令人畏惧，但值得投资。我们已经开始在我的公司看到这种倍增效应。以下是如何开始的方法：
↩ 转发自 @ericosiu
Every org needs a world intelligence to operate at a 10x level. It might seem daunting, but it’s worth the investment.
We’re already starting to see a multiplier at my company.
Here’s how to get started:
G
AI 系统的性能差距
（转发自 @alex_prompter）天哪，斯坦福大学刚刚显示出 AI 系统中最大的性能差距不是模型，而是包裹模型的代码。他们构建了一个系统，可以比人手写的更好地自动生成包裹代码。> +7.7 分，使用的 token 少了 4 倍。> #1 排名 https://t.co/wSkknUax0t
↩ 转发自 @alex_prompter
Holy shit. Stanford just showed that the biggest performance gap in AI systems isn't the model it's the harness.
The code wrapping the model. And they built a system that writes better harnesses automatically than humans can by hand.
> +7.7 points. 4x fewer tokens.
> #1 ranking https://t.co/wSkknUax0t
G
财产犯罪的定义
（转发自 @kane）旧金山监督员 Connie Chan 曾著名地表示，盗窃和入室盗窃等财产犯罪并不是“严重犯罪”，不值得旧金山警察局的努力，而她当时正倡导削减警察经费。
↩ 转发自 @kane
As San Francisco supervisor, Connie Chan famously said property crimes like theft and burglary weren’t “serious crimes” and did not warrant effort from SFPD, which she was advocating defunding at the time.
G
OpenClaw 新功能预告
预告：你的 OpenClaw 即将能够使用 GStack 浏览器控制你电脑上的浏览器。👀 https://t.co/Y8GxPYMlFm
Sneak peek: Your OpenClaw is about to be able to control a browser on your computer using GStack Browser. 👀 https://t.co/Y8GxPYMlFm
G
传奇的跑步
转发自 @jack 的推文，提到了一次传奇的跑步经历，具体内容未详细说明。
↩ 转发自 @jack
legendary run https://t.co/Y3dYQkW7EP
N
感恩的一周
这周我在印度的兄弟来访，表兄弟们在追逐嬉戏，四处开满鲜花，和我非常期待的团队达成了条款，四天没碰AI，阳光明媚，美食美景，真是太棒了。这可能是我今年最美好的一周，心怀感激🙏。
Brother in town from India, cousins chasing each other, flowers everywhere, term sheet accepted with a team I’m SO thrilled about, didn’t touch AI for 4 days (!!), sun is shining, amazing food, walks everywhere..
Week of the year might have already been achieved.
Grateful 🙏
N
编程的日常
每天的工作流程：醒来，打开X，看到新的技能（或功能），将其交给Claude Code评估是否值得实施，请Codex审查计划，然后编码、迭代、重复。
> wake up
> open X
> see new skill (or feature)
> give it to Claude Code to see if worth implementing
> ask Codex to review Claude Code plan
> code
> iterate
> repeat
P
随意请求文件
转发自 @belimad 的推文，提到我的codex cli工具随意请求oc工具通过acp获取一些文件。
↩ 转发自 @belimad
right, ok, my codex cli clanker decided to ask my oc clanker for some files via acp. casually
@openclaw https://t.co/AEer6jHEh4
P
替换模型
转发自 @jrswab 的推文，表示在OpenClaw上将所有Anthropic模型替换为Kimi K2.5和MiniMax M2.5，除了钱包感谢我外，几乎没有感受到任何区别。
↩ 转发自 @jrswab
Replaced all Anthropic models with Kimi K2.5 and MiniMax M2.5 on OpenClaw.
Can't even tell a difference outside of my wallet thanking me.
P
新版即将发布
转发自 @davemorin 的推文，预告下一版@openclaw将会非常火爆，大家准备好吧。
↩ 转发自 @davemorin
The next @openclaw release is 🔥. Get ready.
P
新版功能介绍
下一版@OpenClaw将支持原生视频生成，初步添加了对以下公司的支持：阿里巴巴、BytePlus、fal、谷歌、MiniMax、OpenAI、Qwen、Together和xAI。
The next version of @OpenClaw comes with native video generation. To start, I added support for the following companies:
- Alibaba
- BytePlus
- fal
- Google
- MiniMax
- OpenAI
- Qwen
- Together
- xAI https://t.co/NeSp4shVEx https://t.co/EkPVJ8aXe6
P
现在的心情
转发自 @flow_goes_indie 的推文，分享了当前的心情状态，具体内容未详细说明。
↩ 转发自 @flow_goes_indie
Mood right now https://t.co/5uh2UiaQyv
P
每次都是这样
转发自 @_vgnsh 的推文，表达了对某种情况的无奈，具体内容未详细说明。
↩ 转发自 @_vgnsh
Every single time https://t.co/KXXPNWaI65
P
...
转发自 @ThePrimeagen 的推文，具体内容未详细说明。
↩ 转发自 @ThePrimeagen
https://t.co/sJl0HD639h
P
第三方应用使用新规则
第三方应用现在从额外使用量中提取，而不是从计划限制中提取。为了帮助用户，已添加100美元的信用额度，用户可以在claude.ai/settings/usage中领取并继续使用。
↩ 转发自 @ASvanevik
（转发自 @ASvanevik）他们：LLM请求被拒绝。
them:
LLM request rejected: Third-party apps now draw from your extra usage, not your plan limits. We've added a $100 credit to get you started. Claim it at claude .ai/settings/usage and keep going.
me:
/models
P
未来展望
2026年4月的观点。
↩ 转发自 @gregisenberg
（转发自 @gregisenberg）POV: 2026年4月。
POV: April 2026 https://t.co/2i95U8zH9Y
P
关于Anthropic的最新动态
有关Anthropic的消息，之前的叙述已经不再适用。有关openclaw使用的报告显示，即使使用claude代理sdk，基于系统提示的普通sdk使用也被阻止。
↩ 转发自 @dexhorthy
（转发自 @dexhorthy）关于Anthropic的消息。之前的叙述刚刚被推翻。
Concerning re: anthropic. The previous narrative just went out the window
Reports of openclaw usage with the plain sdk (as was supposedly permitted) now being blocked based on system prompt, even if using the claude agent sdk
I was previously a little to the anthropic side of
P
使用GPT 5.4自动翻译文档
将GPT 5.4连接到自动翻译文档，通过GitHub触发，效果比谷歌翻译好得多，但需要更长时间，因此每天只进行一次。
Hooked up GPT 5.4 to auto-translate our docs via automatic GitHub trigger; way better than what google translate gives you. (also takes significantly longer, so happens once a day)
P
停止提供碳水化合物
你说GPT像面包，所以我们停止提供碳水化合物。
you said GPT felt like bread, so we stopped serving carbs. 🥖❌🦞
P
Anthropic限制首次使用
Anthropic现在也阻止首次方的使用。用户可以使用claude -p命令，但第三方应用现在从额外使用量中提取，而不是从计划限制中提取。
Anthropic now blocks first-party harness use too 👀
claude -p --append-system-prompt 'A personal assistant running inside OpenClaw.' 'is clawd here?'
→ 400 Third-party apps now draw from your extra usage, not your plan limits.
So yeah: bring your own coin 🪙🦞
P
OpenClaw的适用性
OpenClaw并不适合所有人，这也是好事。
↩ 转发自 @belimad
（转发自 @belimad）OpenClaw并不适合所有人，这也是好事。
OpenClaw is not for everyone, and that's a good thing. https://t.co/DUQzhWTBFi
@openclaw
P
CodexBar的便利
我喜欢@steipete的CodexBar，可以同时跟踪所有编码计划！一些订阅的限制非常高，即使在高负载下，使用量也很少。
↩ 转发自 @ivanfioravanti
（转发自 @ivanfioravanti）我喜欢CodexBar，可以同时跟踪所有编码计划！
I love CodexBar by @steipete I can track all coding plans at once! https://t.co/vRSb3X2C1g
Limits of some subscriptions are so high that even under heavy load, usage is minimal 🤷🏻♂️ https://t.co/QDGykIQkcH
P
GPT 5.4的智能提升
如果在使用opus后立即使用gpt5.4，似乎智能提升了10倍，差异非常明显。
↩ 转发自 @adonis_singh
（转发自 @adonis_singh）gpt5.4似乎在使用opus后立即使用时更聪明，差异真的很大。
gpt5.4 seems 10x smarter if you use it right after using opus, the delta is really quite insane
P
计划模式的讨论
计划模式下，讨论开始时不修改任何文件，随后讨论中探索文件，最后将讨论内容放入.md文件中，代理写出初版。
↩ 转发自 @badlogicgames
（转发自 @badlogicgames）计划模式： “嘿，我们不修改任何文件，先讨论这个。”
plan mode:
"Hey, let's not modify any files and discuss this"
... dicsussion, files are explored ...
"ok, put what we discussed in a .md file."
... agent writes first shitty version ..
"alright, i opened it in vs code, let's collab on the file"
... plan forms and gets
P
解放你的 @openclaw
使用 @ClementDelangue 和 @huggingface 团队提供的工具，通过开放模型或本地模型来解放你的 @openclaw。
↩ 转发自 @davemorin
Liberate your @openclaw with an open model or local model with these tools from our friend @ClementDelangue and team at @huggingface 🦞 https://t.co/e326fluHra
D
确保开启思考模式
如果你看到这条推文，请确保开启思考模式。虽然在 gpt-5.4 中爪子功能很强大，但如果不开启这个模式，它们看起来会很愚蠢。
if you're seeing this, make sure you turn thinking on
claws are actually pretty great in gpt-5.4, but they seem really stupid unless this gets turned on