I work on JetBrains .NET IDEs (Rider & ReSharper) · Mostly about IDE tooling · Drones · 3D printing · Opinions are mine
Amsterdam, Netherlands
Joined May 2009
- Tweets56.7K
- Following878
- Followers2.8K
- Likes6.3K
Aleksandr Shvedov retweeted
I need to hire this genius.
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。
If you’ve ever worked at a company while it was going through a major crisis or falling apart completely… what observations come to mind? What stood out to you the most? How did employees and management behave?
Aleksandr Shvedov retweeted
AI bros discover co-routines 40 years later 😭
AWS just open-sourced a runtime that can fit 75,000 concurrent AI agents in 8GB of RAM.
the trick is that agents don’t sit around consuming resources while waiting for tools to finish. the runtime saves their state, suspends them, and resumes them when something happens.
75,000 agents. raspberry pi 5 memory.
that number is kinda ridiculous.
Absolutely true stories that definitely happened. At this point, tech is becoming indistinguishable from fake guru grifting
There is an absolutely WILD story in this @GergelyOrosz podcast episode
1. An unnamed CTO of a cash-flow-negative company was debating doing layoffs. AI
2. He says "I don't want to hurt morale like that, I won't do it"
3. A few months later, he says to @GergelyOrosz that he's cash flow positive. WTF. How?
4. The company had been licensing a component from a licensing company that was critical for their business.
5. One day, a dev thinks "what if we got an agent to build our own?"
6. The agent does so well that they produce a BETTER, FASTER version of the component they were licensing.
7. They debate open-sourcing it, and they ask the licensing company about it
8. The licensing company panics, and says "we will pay you NOT to open-source this"
9. They also say "we'll also pay you extra so we can re-license it"
10. So a big outlay became a recurring windfall, thanks to the ingenuity of a dev asking the right question.
Individual contributors have never been more powerful. Think bigger.
youtu.be/mQu0zN4IvSI?si=P_v9…
Aleksandr Shvedov retweeted
chaser: do you know what other company takes 30% while doing way less than what Steam does and is aggressively protecting their monopoly? I'll give you three guesses.
woah, this is crazy. Types are just the partial functions in Verse, so `(x:T) => e` basically just desugars into `x => { x := T(x); e }`. I never saw anything even remotely similar to this design...
🤡🤡🤡 What is the problem? Why not be proud of a great products shipped with AI label? If AI is so good, why shouldn’t we disclose its usage?
Aleksandr Shvedov retweeted
This is not confusing at all
Copilot Code is a new AI-powered coding experience in Microsoft Copilot that helps people turn ideas into working software. Learn more: msft.it/6011aaNIB
damn... in my archives I've found a .dat file from the year 2008, no visible textual data inside... almost removed it... Codex: "this is the Miranda IM database file, 90k ICQ messages from three years"🤯🤯🤯
TL;DR if you haven't watched it yet
The End of Programming? Video of my talk yesterday at #scaladays. youtube.com/watch?v=jlJyuJhD…
Aleksandr Shvedov retweeted
Anthropic is offering a 145k salary for a personal assistant in London.
Responsibilities include:
"Scheduling and managing calendars for leader(s) and their organization"
"Building and maintaining organizational systems for tracking and managing inbound requests"
"Managing the leader's inbox and communications with precise attention to detail"
"Handling travel logistics"
Somebody should really build some sort of tech tool that can handle this stuff autonomously.
this week in AI native software engineering: let's kill all or "intermediary" abstractions. "every layer is just one more place for the agent to guess wrong".
my new AI native software engineering principle is to delete as many layers of abstraction as possible
at this point, all my webapps are written in rust and compiled to wasm, UI is rendered directly in html and/or raw webgpu calls
no react/threejs/nextjs or any other intermediary abstractions
all my effort goes into writing custom validation harnesses for each app that fully quantifies perf / usability / what "good" means for the coding agent so it can hillclimb it
every agent sessions are planned via my explore unknowns skill (github.com/dzhng/skills), so all major decisions are frontloaded, then I just let the agent rip for 12+ hrs
at this point I'm solo shipping products that I wouldn't even dare to dream of a year ago
Aleksandr Shvedov retweeted
$300 for a bug that gives free access to OpenAI’s paid models with no API key or account.
Zero reason to report anything else I find to them.
