📰 AI 博客每日精选 — 2026-06-16
来自 Karpathy 推荐的 92 个顶级技术博客，AI 精选 Top 21
📝 今日看点
今日技术圈聚焦两大主线：AI监管博弈持续升温，Anthropic因与白宫人格冲突导致模型下线，欧盟亦裁定谷歌将Gemini集成Android违反《数字市场法案》，科技巨头与政策制定者的角力进入深水区；与此同时，AI工具正加速向实用化演进，Mux推出视频智能工作流、WorkOS发布AI代理注册协议，开发者生态迎来更高效的数据处理与自动化能力。此外，GIF格式诞生38周年与x86模拟器团队修复“烂代码”的趣闻，也为技术社区增添了历史与人文的回响。
🏆 今日必读
🥇 Mux — 面向开发者的视频服务
‘Anthropic’s Safety Superpower’ — daringfireball.net · 06-15 17:18 · 🤖 AI / ML
视频文件包含大量数据，传统技术栈难以读取和利用其中的音频、对象和场景信息。Mux Robots 可将视频数据转化为智能，只需配置一次工作流，即可在新上传时自动执行问答、摘要、关键片段查找等任务，无需资产 webhook 或自托管胶水代码。Mux 是受 Synthesia、Shopify 和美国足协信赖的开发者视频基础设施。
💡 为什么值得读: 如果你正在寻找一种无需自建复杂系统就能快速实现视频智能分析的方法，Mux 提供了一站式解决方案，能显著降低开发成本。
🏷️ Anthropic, AI safety, LLM, competition
🥈 测试五边形数
Cloudflare CAPTCHA on at least one ampersand — simonwillison.net · 06-16 00:21 · 🤖 AI / ML
五边形数 Pn 由公式 Pn = (3n² − n)/2 定义，其中 n 为正整数，对应 n 个同心五边形的点数。若 n 为非正整数，则得到广义五边形数。给定 n 可轻松计算 Pn，但反过来判断一个给定数是否为五边形数则需要求解二次方程。文章探讨了如何高效测试一个数是否为五边形数。
💡 为什么值得读: 对于数学爱好者和算法设计者，本文提供了判断五边形数的简洁方法，有助于理解数论中的整数判定技巧。
🏷️ OpenAI, losses, AI industry, spending
🥉 GIF 于 1987 年 6 月首次亮相
The Washington Post on the EU’s DMA Folly — daringfireball.net · 06-16 02:20 · 💡 观点 / 杂谈
1987 年 6 月 16 日，GIF 文件格式正式诞生，其历史早于现代互联网。20 世纪 90 年代初，首批网页浏览器就已支持 GIF，因为它在当时已非常流行且兼容性良好。GIF 的普及为早期网络图像传播奠定了基础。
💡 为什么值得读: 了解 GIF 的起源有助于理解互联网图像格式的演进脉络，对技术史感兴趣者不容错过。
🏷️ DMA, Apple, EU regulation, AI
🤖 AI / ML
1. Mux — 面向开发者的视频服务
‘Anthropic’s Safety Superpower’ — daringfireball.net · 06-15 17:18 · ⭐ 26/30
视频文件包含大量数据，传统技术栈难以读取和利用其中的音频、对象和场景信息。Mux Robots 可将视频数据转化为智能，只需配置一次工作流，即可在新上传时自动执行问答、摘要、关键片段查找等任务，无需资产 webhook 或自托管胶水代码。Mux 是受 Synthesia、Shopify 和美国足协信赖的开发者视频基础设施。
🏷️ Anthropic, AI safety, LLM, competition
2. 测试五边形数
Cloudflare CAPTCHA on at least one ampersand — simonwillison.net · 06-16 00:21 · ⭐ 25/30
五边形数 Pn 由公式 Pn = (3n² − n)/2 定义，其中 n 为正整数，对应 n 个同心五边形的点数。若 n 为非正整数，则得到广义五边形数。给定 n 可轻松计算 Pn，但反过来判断一个给定数是否为五边形数则需要求解二次方程。文章探讨了如何高效测试一个数是否为五边形数。
🏷️ OpenAI, losses, AI industry, spending
3. Writing Prolog with ChatGPT
Writing Prolog with ChatGPT — johndcook.com · 06-15 17:22 · ⭐ 23/30
A few days ago I wrote about using Claude to solve a chess puzzle by writing Prolog code. This morning I tried a similar chess puzzle with ChatGPT. The task is to place a queen, king, rook, bishop, an
🏷️ Prolog, ChatGPT, chess puzzle, logic programming
4. AI's Brokenomics
AI's Brokenomics — wheresyoured.at · 06-15 19:28 · ⭐ 23/30
If you liked this piece, you should subscribe to my premium newsletter. It’s $70 a year, or $7 a month, and in return you get a weekly newsletter that’s usually anywhere from 5,000 to 18
🏷️ AI, economics, NVIDIA, Anthropic
5. “他们坑了我们”：人格冲突导致Anthropic模型下线
"They screwed us": Personality clashes sent Anthropic's models offline — simonwillison.net · 06-15 14:57 · ⭐ 22/30
Anthropic与美国政府之间的人格冲突直接导致其AI模型被强制下线。Axios报道援引多位知情人士透露，白宫与Anthropic在出口管制政策上存在严重分歧，双方互相指责。文章汇集了大量幕后消息，揭示了Mythos项目背后的权力博弈和个人恩怨。结论是，这次事件凸显了AI公司与政府监管之间的脆弱关系，以及高层个性对技术部署的致命影响。
🏷️ Anthropic, AI, policy, conflict
6. 四元数旋转、Claude与Lean
Quaternion Rotations, Claude, and Lean — johndcook.com · 06-15 19:31 · ⭐ 22/30
作者收到读者邮件指出一年前博客文章中关于四元数与旋转矩阵转换的笔误，但决定测试Claude Sonnet 4.6 Medium能否自行发现。他给出了包含具体代码和数学公式的提示，要求Claude找出错误。Claude成功定位了笔误位置并给出了修正建议。结论是，AI可以高效辅助技术文档校对，尤其适合数学和编程领域的错误检测。
🏷️ quaternions, rotations, Claude, Lean
7. The Fable 5 Export Controls Harm US Cyber Defense
The Fable 5 Export Controls Harm US Cyber Defense — simonwillison.net · 06-16 05:20 · ⭐ 20/30
The Fable 5 Export Controls Harm US Cyber Defense
🏷️ JAX, GPU, memory, async
📝 其他
8. JAX: commitment issues
JAX: commitment issues — gilesthomas.com · 06-15 21:30 · ⭐ 15/30
Imagine you have JAX code like this, and run it on a machine with CUDA set up:
key =
9. Lean, not backpressure
Lean, not backpressure — entropicthoughts.com · 06-15 22:00 · ⭐ 15/30
Lucas Costa has written a good article on how to build systems that can handle
code-generating robots. Unfortunately, when calling it backpressure, he used
the wrong metaphor.
Backpressure is abo
10. Exclusive: OpenAI Losses Increased Nearly 8X in 2025, With Spending Hitting $34 Billion
Exclusive: OpenAI Losses Increased Nearly 8X in 2025, With Spending Hitting $34 Billion — wheresyoured.at · 06-16 03:58 · ⭐ 15/30
Soundtrack: In Flames - ColonyTo further support my independent journalism, please subscribe to my premium newsletter. It’s $7 a month or $70 a year. If you’re subscribed to the free new
11. Testing pentagonal numbers
Testing pentagonal numbers — johndcook.com · 06-15 23:57 · ⭐ 11/30
The nth pentagonal number Pn is the number of dots in diagrams like those below with n concentric pentagons. We have the formula Pn = (3n² − n)/2 where n is a positive integer. If n is an integer but
🏷️ pentagonal numbers, mathematics, number theory
12. GIF’s June 1987 debut
GIF’s June 1987 debut — dfarq.homeip.net · 06-15 11:00 · ⭐ 10/30
On June 16, 1987, the GIF file format made its debut. That’s right, GIF predates the modern Internet. The first web browsers supported GIF because it was already very popular in the early 1990s and we
🏷️ GIF, file format, history
💡 观点 / 杂谈
13. GIF 于 1987 年 6 月首次亮相
The Washington Post on the EU’s DMA Folly — daringfireball.net · 06-16 02:20 · ⭐ 24/30
1987 年 6 月 16 日，GIF 文件格式正式诞生，其历史早于现代互联网。20 世纪 90 年代初，首批网页浏览器就已支持 GIF，因为它在当时已非常流行且兼容性良好。GIF 的普及为早期网络图像传播奠定了基础。
🏷️ DMA, Apple, EU regulation, AI
14. 欧盟委员会数月前裁定谷歌在Android中集成Gemini违反《数字市场法案》
The European Commission Ruled Months Ago That Google’s Integration of Gemini in Android Violates the DMA — daringfireball.net · 06-15 18:58 · ⭐ 22/30
欧盟委员会裁定谷歌将Gemini深度集成到Android系统违反了《数字市场法案》（DMA）。监管机构提议多项改革，包括允许第三方AI助手通过热词或按钮在系统级调用，允许AI工具在用户授权下读取屏幕上下文，以及允许替代AI系统访问本地数据以生成主动建议和摘要。这些措施旨在打破谷歌对AI助手的垄断，迫使Android平台向其他AI助手开放。结论是，谷歌可能面临重大整改，Android生态的AI竞争格局将彻底改变。
🏷️ DMA, Google, Gemini, Android
15. EU & Civil Society need to progress on Digital Autonomy
EU & Civil Society need to progress on Digital Autonomy — berthub.eu · 06-15 13:00 · ⭐ 19/30
By now (happily) everyone wants to talk about digital autonomy, although some parties insist we talk about sovereignty. Fine.
tl;dr: The discussions on digital autonomy are now going round in circles.
🏷️ EU, digital autonomy, civil society
16. Pluralistic: AI and amateurism (15 Jun 2026)
Pluralistic: AI and amateurism (15 Jun 2026) — pluralistic.net · 06-15 15:40 · ⭐ 16/30
Today's links AI and amateurism: When is generative content vernacular? Hey look at this: Delights to delectate. Object permanence: Disney characters x clean underwear; Transparent Pontiac; Makers v d
🏷️ AI, amateurism, generative content, vernacular
🛠 工具 / 开源
17. WorkOS Launches Auth.md — an Open Protocol for Agent Registration
WorkOS Launches Auth.md — an Open Protocol for Agent Registration — daringfireball.net · 06-15 17:53 · ⭐ 24/30
My thanks to WorkOS for sponsoring DF last week to promote Auth.md, their new open protocol for AI agent registration.
Sign-up forms were built for humans in browsers, so how do AI agents programmati
🏷️ Auth.md, AI agent, registration, open protocol
18. datasette-agent 0.3a0
datasette-agent 0.3a0 — simonwillison.net · 06-15 17:19 · ⭐ 17/30
Release: datasette-agent 0.3a0
<blockquote>
- New tool,
execute_write_sql
🏷️ datasette, agent, SQL, permissions
19. [Sponsor] Mux — Video for Developers
[Sponsor] Mux — Video for Developers — daringfireball.net · 06-16 02:33 · ⭐ 12/30
Video is a boatload of data. Every video file in your product contains audio, objects, and scenes that most stacks can’t read or access.
Mux Robots turns that data into video intelligence. Configure
🏷️ video, AI, intelligence, Mux
⚙️ 工程
20. The time the x86 emulator team found code so bad that they fixed it during emulation
The time the x86 emulator team found code so bad that they fixed it during emulation — devblogs.microsoft.com/oldnewthing · 06-15 14:00 · ⭐ 21/30
The time the x86 emulator team found code so bad that they fixed it during emulation
🏷️ x86 emulator, code quality, software engineering, debugging
21. Quoting Matteo Wong, The Atlantic
Quoting Matteo Wong, The Atlantic — simonwillison.net · 06-16 03:07 · ⭐ 20/30
Quoting Matteo Wong, The Atlantic
🏷️ backpressure, systems, design, metaphor
生成于 2026-06-16 11:21 | 扫描 83 源 · 共 2467 篇 · 28h 内新发布 21 篇 · 精选 21 篇
基于 Hacker News Popularity Contest 2025 RSS 源列表，由 Andrej Karpathy 推荐