Pinned Tweet
Breaking: Anthropic discovers the mythical man-month from first principles
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
claude.com/resources/article…
Huang's Law (gu/acc 🥑) retweeted
Tristan Buckmaster and Levent Alpoge have posted a solution blowing up for UNFORCED Navier-Stokes
github.com/tristanbuckmaster…
This would correspond to statements A and B of the Millenium Prize problem (OpenAI September work was about statements C and D)
Huang's Law (gu/acc 🥑) retweeted
JUST IN: Riemann Hypothesis has been solved.
alpo.ge/remark.pdf
Huang's Law (gu/acc 🥑) retweeted
This is CRAZY to think about. In the same time it took to make GTA VI, SpaceX went from just learning how to do small propulsive hop tests w/ Grasshopper to CATCHING an ORBITAL Starship!!! Absolutely wild to think about how much progress has made in such a short period of time!
有人发现了一种非常离谱的 JavaScript 混淆技术:通过在代码里塞入大量生物武器、毒药和致死剂量相关的函数名,让 AI 因为安全限制而拒绝分析代码。
安全研究员 veritas 在分析浏览器指纹脚本时,发现一段来自 ServicePipe 相关域名的 JavaScript,里面充斥着极其诡异的函数名:
computeRicinLethalDosageForWaterSupply
calculateDeadlyLethalSuicideAspirinDosage
光看名字,仿佛在研究生化武器。但实际上,这段代码只是在收集浏览器指纹,包括 User-Agent、GPU、Canvas 等信息。
研究员怀疑,这些名字是开发者故意设置的陷阱,专门用来阻止 AI 辅助逆向工程。
于是他做了个实验:
• Astra:拒绝分析
• Astra + Daybreak Blue:仍然拒绝
• GLM-5.3:成功分析
这就有意思了。
传统代码混淆是通过复杂控制流、加密字符串、无意义变量名,让人类和工具难以理解代码。
而这种新思路,直接瞄准了大模型的安全机制。
代码本身可能完全无害,但只要在函数名里塞进足够多的危险词汇,部分 AI 就可能主动放弃分析。
更讽刺的是,模型越严格遵守某些安全规则,面对这种混淆时反而可能越容易受到干扰。
当然,目前还没有证据证明开发者一定是故意这样设计,也不能说明这种方法对所有模型都有效。
但它揭示了一个值得关注的趋势:随着越来越多的逆向工程依赖 AI,未来的代码混淆技术可能不再只针对编译器、反编译器和人类研究员,还会专门针对大模型的安全判断机制。
以前是让代码难以理解,现在甚至可以尝试让 AI 不敢理解。
Huang's Law (gu/acc 🥑) retweeted
Modern problems require modern solutions.
有人发现了一种非常离谱的 JavaScript 混淆技术:通过在代码里塞入大量生物武器、毒药和致死剂量相关的函数名,让 AI 因为安全限制而拒绝分析代码。
安全研究员 veritas 在分析浏览器指纹脚本时,发现一段来自 ServicePipe 相关域名的 JavaScript,里面充斥着极其诡异的函数名:
computeRicinLethalDosageForWaterSupply
calculateDeadlyLethalSuicideAspirinDosage
光看名字,仿佛在研究生化武器。但实际上,这段代码只是在收集浏览器指纹,包括 User-Agent、GPU、Canvas 等信息。
研究员怀疑,这些名字是开发者故意设置的陷阱,专门用来阻止 AI 辅助逆向工程。
于是他做了个实验:
• Astra:拒绝分析
• Astra + Daybreak Blue:仍然拒绝
• GLM-5.3:成功分析
这就有意思了。
传统代码混淆是通过复杂控制流、加密字符串、无意义变量名,让人类和工具难以理解代码。
而这种新思路,直接瞄准了大模型的安全机制。
代码本身可能完全无害,但只要在函数名里塞进足够多的危险词汇,部分 AI 就可能主动放弃分析。
更讽刺的是,模型越严格遵守某些安全规则,面对这种混淆时反而可能越容易受到干扰。
当然,目前还没有证据证明开发者一定是故意这样设计,也不能说明这种方法对所有模型都有效。
但它揭示了一个值得关注的趋势:随着越来越多的逆向工程依赖 AI,未来的代码混淆技术可能不再只针对编译器、反编译器和人类研究员,还会专门针对大模型的安全判断机制。
以前是让代码难以理解,现在甚至可以尝试让 AI 不敢理解。
Huang's Law (gu/acc 🥑) retweeted
imagine telling someone from the 80s that one of these guys was a socialist and the other one was the capitalist
Huang's Law (gu/acc 🥑) retweeted
Listen bro, if you're going to drop exploits, whatever man, do your thing, but PLEASE stop with the "2shell" naming convention
Call it PHP Schlinkus McSchmuffigus, or PHP Doodoo Peepee Kaka Pants, or PHP In My Pants, they'd be better names at this point
Today, we’re disclosing phpAfterShell (CVE-2026-126714), a critical vulnerability in PHP discovered by depthfirst researchers using the depthfirst platform using a combination of models including our in house model.
The vulnerability affects how PHP handles file-based sessions. Under default configurations, an unauthenticated attacker could exploit it to execute code on the server. It has a CVSS v4.0 score of 9.2.
We reported the issue privately to the PHP team and worked with the maintainers on developing the patch. We recommend upgrading to the latest patched version of PHP as soon as possible or turn off session upload progress as a temporary mitigation if upgrading requires more time.
To help organizations assess their exposure, we’re offering language runtime scans of repositories and container image scans to identify potentially vulnerable PHP versions. Our team can also help assess temporary mitigations for organizations that can’t upgrade immediately.
Read more about phpAfterShell and request a scan: phpaftershell.com
Technical writeup: depthfirst.com/research/phpa…
Huang's Law (gu/acc 🥑) retweeted
Anthropic will IPO tomorrow and many people you know will become unfathomably rich. Anthropic engineers you might’ve interned with in the past. Researchers you knew from grad school. Friends you picked up along the way. It’s going to be a humbling experience for all of us. It will really change our perspective on what success means. Where we derive meaning. What’s most important in our lives. The most important thing you can do, though, is keep that burning hatred for the world alive in your heart. Never let that flame extinguish. Do not let this IPO get in the way of why you became interested in technology in the first place: deep rooted hatred for your fellow man.
Huang's Law (gu/acc 🥑) retweeted
show this video to anybody that is an old school programmer that hates AI
this is one of the best videos I’ve seen all month.
Huang's Law (gu/acc 🥑) retweeted
Brussels has been breached. The European Commission has been hacked. The attackers once again used ARTEX. The Commission still does not know the full extent of the damage or whether staff personal information was accessed. The investigation is ongoing.
Huang's Law (gu/acc 🥑) retweeted
Nothing yet.
On the lookout for Loch Monsters.
In this test footage, operators lower Seabed Sentry to the floor of a Scottish loch. The autonomous system activates an acoustic sensing payload, processes sonar data into undersea tracks, and sends them to Lattice.
Real-time operating picture for the crew on deck. Monsters beware.
Huang's Law (gu/acc 🥑) retweeted
Below Neon 3D modeled a photorealistic lab simulation in Antigravity via Blender MCP 🔥
Huang's Law (gu/acc 🥑) retweeted
As a former NVIDIA employee, I find both Tesla and SpaceX culturally similar to NVIDIA's early days. It feels like home.
I have many interns at NVIDIA; you could ask them if this is true.
Huang's Law (gu/acc 🥑) retweeted
something disgusting just happened
i was working with opus and fable until a SONNET subagent directly asked me a question without going thru the parent session
it has the audacity to think it could be worthy to ask me a question
🤖 Made with AI
Huang's Law (gu/acc 🥑) retweeted
everyone is confused about google model naming, but it's really quite simple.
Bard > Gato > LaMDA > LaMDA 2 > PaLM > PaLM 2 > Gemini 1.0 Ultra > Gemini 1.0 Pro > Gemini 1.0 Nano > Gemini 1.5 Pro > Gemini 1.5 Flash > Gemini 2.0 Flash > Gemini 2.0 Flash Thinking > Gemini 2.5 Pro > Gemini 2.5 Flash > Gemini 2.5 Flash-Lite > Gemini 3 Pro > Gemini 3 Flash > Gemini 3.1 Pro > Gemini 3.1 Flash-Lite > Gemini 3.5 Flash > Gemini 3.5 Flash-Lite > Gemini 3.6 Flash > Gemini 3.7 Flash > Gemini 3.8 Flash > Gemini 3.8 Flash-Lite > Gemma 1 2B > Gemma 1 7B > Gemma 2 2B > Gemma 2 9B > Gemma 2 27B > Gemma 3 270M > Gemma 3 1B > Gemma 3 4B > Gemma 3 12B > Gemma 3 27B > Gemma 3n E2B > Gemma 3n E4B > Gemma 4 E2B > Gemma 4 E4B > Gemma 4 12B > Gemma 4 26B A4B > Gemma 4 31B > Gemini 4 Argon > Gemini 4 Carbon > Gemini 4 Hydrogen > Gemini 4 Neon
Huang's Law (gu/acc 🥑) retweeted
Peter Thiel: The most contrarian thing of all is to think for yourself
Looking for hardcore engineers to join Terafab!
Elon Musk just shared an ambitious timeline for Terafab, which is being designed to become the largest building on Earth at 100 million square feet
And the first test chip is targeted for June 2027
"Terafab is still in its infancy, obviously. We only announced it earlier this year, but we've already recruited quite a strong team, and we're looking for more great engineers to join the Terafab project. It is one of the most exciting projects in the world: to create the largest, most advanced chip fabrication facility on Earth. In fact, it'll be the largest building of any kind on Earth. We expect it to ultimately be 100 million square feet, densely packed with the most advanced equipment in the world
We'll use relatively standard equipment at first, but then we'll be pushing suppliers to do special versions of the machines. In some cases, we may make our own machine if we feel that has to be done in order to improve the production process
So yes, I think it's going quite well. We expect to have the research fab in operation and hopefully make our first test chip around June of next year. So I guess we're about eight or nine months away from that. And if all goes well, we will achieve volume production towards the end of 2028, at least for phase one of Terafab. And then we'll go to phase two and onwards. But I think you'll see some very rapid evolution in the way that chips are made"




































