Kum Hou

[Kum Hou @kumhou92]

@peternoszek Thanks!🥳

[Kum Hou @kumhou92]

@elonmusk @bot Any chance of getting one-month trial subscription of Grok Bot? Followed for quite some times but did not know whether it's worth to subscribe.

[lauren @poteto]

Day 28/ tibo's game has landed

i spent $5.99 on this domain, but i had a blast making it with grok bot and cursor! i hope you enjoy it and find all the easter eggs 🤍

if i did it right you should be able to play it inside of X too

https://t.co/diV7PyPaqv

[Kum Hou @kumhou92]

this is unexpected ☠️

[lifcc @mylifcc]

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。

[Kum Hou @kumhou92]

@16vchq Just enough to made ourself survive, all people has their own situation.

[Kum Hou @kumhou92]

@anuragdotdev I'm more curious about how did your friend answer to this?

[Kum Hou @kumhou92]

@itsnotryan Hey! It's great for your success. Do you mind to give some advices to newcomers for the ventures?

[Kum Hou @kumhou92]

@Nehakathare To be specific, it's about AI memory system that runs locally to handle the memories of each session while we're working with coding agents like Claude Code and Codex, and the memories can be recalled seamlessly in between.

Any feedbacks are welcomed!
https://t.co/Ed21BNaX2E

[Kum Hou @kumhou92]

@yoeven Applying to W27 with MnemoSeed, building toward memory for AI in any form, starting with a working local MVP that helps coding agents remember project context across sessions. https://t.co/VWpbYWyLsL

[Kum Hou @kumhou92]

@daleverett https://t.co/Ed21BNaX2E
Here you go!

[Kum Hou @kumhou92]

@HisashiSpace @MaximLevoshin Sure! Nice to meet you!

[Kum Hou @kumhou92]

@kushmergedeck Hey Kush! 👋

I'm working on MnemoSeed, a local-first memory layer that helps AI agents carry useful context across sessions and tools.

Still in the early stages, but would love to connect with other builders, exchange feedback, and see what everyone's shipping! 🚀

[Kum Hou @kumhou92]

@thara023 @X Looks like the algorithm worked! 👋

I'm building MnemoSeed, a local-first AI memory project.

Still early days, but always happy to connect with fellow developers and indie hackers, exchange ideas, and learn from what others are building! 🤝

[Kum Hou @kumhou92]

@ammar_nassri Hey Ammar! 👋

I'm building MnemoSeed, a local-first memory layer for AI agents. Still in the early stages, but excited to connect with others building in AI and exchange ideas!

Here's what I'm working on:
https://t.co/Ed21BNaX2E

[Kum Hou @kumhou92]

@b1rkh0ff The reach is nice, but connecting with people who actually build is even better!

I'm working on MnemoSeed, a local-first AI memory layer. Still early, but excited to share ideas, learn from other builders, and see what everyone is shipping.

Let's connect! 🚀

[Kum Hou @kumhou92]

@SemenovID Hey Ivan! Builder here 👋

I'm working on MnemoSeed, a local-first AI memory layer that helps AI agents remember across tools and sessions.

Still early days, but I'd love to connect with fellow founders and devs, share ideas, and learn from what others are building!

[Kum Hou @kumhou92]

@MaximLevoshin If you're curious, here's the repo for what I'm building (still very early!):

https://t.co/Ed21BNaX2E

Would love to hear your thoughts!

[Kum Hou @kumhou92]

@MaximLevoshin I'm also building something in the AI memory space! Still in the early stages, with a local-first approach.

Would love to connect, share what we're building, and exchange ideas. Maybe even explore some collaboration if there's a good fit!