Inertia is an odd thing, I want to achieve something, but it feels like every fiber of my being is attempting to stop me
4
AI can solve hard coding problems, and it can do it quick, correct and check itself. It's lacking a lot when it comes to UX and a bit with UI tho.
It's very visible on the web where UI/UX is still fairly simple, however translate those error rates to games where it's nothing but UX, it's not wonder that AI games feel so sloppy.
All in due time, but I think the idea of 1 shotting a project is dumb, I think we'll see curated games, where all aspects have been though about thoroughly created wtih AI.
18
It's crazy how much inertia there is to starting learning something a little outside your comfort zone
1
3
Just bought the two templates for this, not sure I'm gonna use it, but it's damn cool and I want more stuff like this out in the world
Asemulator v1.0.0 is OUT NOW! 🎉
Preview, test and share your animated sprites in context, directly in @aseprite, starting today!
(Link in replies)
1
7
Thanks microsoft and anthropic, this feels coordinated, why do I get logged out of claude while ms is having a breakdown so I can't get back in.... stop vibing your products
1
330
The kinds of ideas I get now that my time isn't the limiting factor in executing on projects are kinda bizarre.
I've made a gitbutler language server for an extension in zed for a use case so niche that I doubt it's a problem that anyone but me has.
17
UX designers should thrive in this age of AI. Since the LLMs rarely know how to design a good experience. They're very good at making something workable, but if you don't know how to ask for a good experience, you wont get it.
1
1
10
Currently building an audit page into collagen to highlight the usage of the program. It's two fold, both a way to see token usage as well as user behavior in the app
1
1
47
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。
What's · up · with · claude · and · using · these · as separators · for · everything
It's so annoying and hella ugly. I had to write a linting rule to ban it from our project because it just introduced it everywhere
23







