
High entropy signaler @clankoids - https://teapawt.net/t.co/2TvCOQ68zE
Buenos Aires
Joined October 2023
- Tweets253
- Following620
- Followers109
- Likes25.5K
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。
Chris retweeted
If your company refers to engineers as the IT department, you have nothing to worry about. You were already breached last week.
Chris retweeted
Ahora que estoy en modo AI SDR sales, me puse a analizar los job postings del mercado abierto para programadores en Argentina.
Silver es, sin lugar a dudas, el mejor job board del pais.
1. Tenemos más startups que cualquier job board individual
2. Tenemos la mayor cantidad de información por postulación
3. Los candidatos tienen las mejores chances de conseguir entrevistas con nosotros.
I've been working on this for some time now, and I think some people might find it useful.
open for signup today: clstr.news
CLSTR (as in "cluster") takes a shot at being a better way to read news in the age of information overload.
Chris retweeted
I am completely convinced that people's adoration for one model over another is purely based on a good experience they had in the past and on nothing objective.
just some problem was solved well enough for them, therefore its the best
Chris retweeted
First update is live: shared and personal whitelists/blacklists, dock/hide the UI. We're also retraining the ML model on a larger dataset.
We need more feedback! What sucks? Let us know!
Chris retweeted
Les pongo un ejemplo de por qué las startups nos terminamos yendo a Estados Unidos para fondearnos y crecer afuera.
Por cada cobro que tenemos en Argentina nos inundan de retenciones e impuestos, incluso ANTES de recibir un pago, ya nos están cobrando y aplicando percepciones.
Sumemos acá que cada cliente también te retiene antes de ejecutar el pago, por lo que recibís menos dinero del indicado en la factura original.
Además, mes a mes también tenemos que presentar las DDJJ de IVA e Ingresos Brutos. Siendo que durante todo el mes nos fueron reteniendo de antemano.
No solo eso, como trabajamos en varias provincias tenemos que pagar distintas alícuotas (%) en cada una de las jurisdicciones.
Por cada cobro que recibimos se generan 6 movimientos bancarios adicionales, es decir, si un mes nos ingresan por ejemplo 10 cheques, tenemos SESENTA movimientos bancarios adicionales solo de impuestos.
No solo se trata de cobrar menos impuestos como pasa en Delaware, Texas o las Islas Caimán, se trata de simplificarle la vida al que produce, invierte, genera empleo y arriesga su capital.
Lo mismo ocurre para el inversor, por eso toda startup levanta capital con una C-Corp en Delaware y termina creciendo en el mercado norteamericano.
Esto tiene que cambiar, es la forma de que se creen startups y el talento se mantenga en Argentina.
Chris retweeted
If you throw AI at a bad codebase, you're going to get worse results. Garbage in, garbage out.
And holding it together in your head will land you in cognitive debt.
But these problems have a 20-year old solution: deep modules.
Here's how:
Chris retweeted
Why Your Best Engineers Are Interviewing Elsewhere
codegood.co/writing/why-your…
Chris retweeted
Iteration velocity solves all known problems in software engineering (including quality)
People often ask how we balance speed and quality at @tryramp.
We don’t. Because speed is how you get to quality.
Even the best hitters in baseball miss 70% of the time. A .300 batting average means the world’s best still fail twice as often as they succeed.
Building products is no different.
Even if you deeply understand your customer, you’ll still be wrong most of the time — it just takes iteration to discover what actually works.
Take two teams:
Team A ships every 2 weeks.
Weeks 2, 4, 6, 8 — all wrong. Week 10 — nailed it.
Team B waits for “perfect.”
Week 8 — wrong. Week 16 — wrong. Week 24 — finally right.
Team A made more mistakes, but found truth faster.
Team B “protected” quality and ended up slower, later, and with less conviction.
In the real world, there’s no limit to your at-bats per inning.
You can swing 100 times if you design your org and culture for it.
Speed isn’t a trade-off with quality. Speed is the way to get to quality.































