𝒻 (x) 🇲🇾 (@drahfa) header
𝒻 (x) 🇲🇾 (@drahfa) profile picture

(╯°□°)╯︵ ┻━┻

Vector Space
Joined October 2008
Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)Photo by 𝒻 (x) 🇲🇾 (@drahfa)
Bridge Jump: How It Started vs. How It's Going #KingOfHighways
1
4
321
Pre-Ozempic, there were many theories: fat people weren't really hungry--it was our obesogenic environmemt, ultra-processed food, sedentariness, emotional problems. Post-Ozempic, we know: most were just really hungry. Reduce their hunger & they lose weight.
Diet studies often find that ~80% of weight comes back within 5 years. If calorie math is so simple (calories in minus calories out), why do diets so rarely succeed long-term? Dr. Katharine Saunders explains in our latest Clearer Thinking podcast episode: podcast.clearerthinking.org/…
149
216
41
7,566
472,070
This has to be said again and again. I’m so glad StackOverflow died.
Replying to @NuttingOnElon77
The best thing AI had done is make me never have to interact with a Stack Overflow user ever again. Actually such horrible people.
1
1
17
893
𝒻 (x) 🇲🇾 retweeted
Soft launching noteboard.ai 🎉 It's a brainstorming workspace with an infinite canvas, where you can organize your thoughts and explore ideas. Built for individual and for team's brainstorming session. Very excited with this project :D Also quite a domain huh
2
7
12
451
𝒻 (x) 🇲🇾 retweeted
Game theory advice: Never ask yourself why a person behaves the way they do. That's a broad question, and you'll find nothing. Instead, find a clear structural reason: an incentive. Ask: "What outcome is being rewarded?" Is it the pursuit of social status, a monetary gain, or maybe an emotional desire? You're only ever as intelligent as the questions you seek to answer.
15
156
4
1,439
39,454
Every time someone uses the word ‘Dunning–Kruger’ to make a point, I get traumatic flashbacks. It’s Elm Street all over again.
1
5
717
𝒻 (x) 🇲🇾 retweeted
We have spun up 1061 computers since started boxcompute Couldn't do all this without the great infra team, Btw we manage everything end to end ourselves with the help from partners, From buying the baremetal -> setting up the servers -> assembling the components and even wrote the virtualization method ourselves Even with the help of AI, domain expertise is still crucial to build this kind of product
1
5
18
618
Looking forward to Ramadhan!
Ramadhan is just 100 days away! Rindu ♥️
2
250
𝒻 (x) 🇲🇾 retweeted
I always find myself coming back to this. It seems to gain in depth every time I read it.
13
168
9
2,141
44,974
Yes.
topology optimization in a nutshell is just the computer asking where the metal actually does any work then deleting the rest
2
278
Helicopter-ramming action at Seremban-Senawang PLUS Highway. #KingOfHighway
Can only do gamedev during weekends. Today’s color-grading works. #KingOfHighways
9
520
Can only do gamedev during weekends. Today’s color-grading works. #KingOfHighways
1
1
27
1,138
Tq Abg GabeN 🤗
3
30
875
𝒻 (x) 🇲🇾 retweeted
Carl Jung was right when he said: in my experience, I have found that the solutions to all of life’s most intractable problems are not found through any form of intellect or cleverness. They are solved simply by moving on.
105
3,006
106
26,562
1,095,057
Defensive programming.
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。
1
1
6
407
Penat ah tgk Edge of Tomorrow ni. Gedebak degebuk rupanya semua mimpi dalam mimpi hero tertidur dlm helikopter.
4
16
2,005
To be clear, it’s a great movie 🤗 My ask is just “If you experience something, remember it, and learn from it, but it never happened in anyone else's reality, was it real?”.
2
168
Post-Jobs and Ive, Apple becoming Sony is more plausible than Apple becoming Nokia. Cook’s era is stable. Ternus will be remembered with Duo’s mishaps.
苹果未来的索尼化 苹果未来很可能会走向索尼化。 索尼曾经是消费电子行业的引领者,Walkman、特丽珑电视、Handycam,都曾定义一个时代的消费体验。今天的索尼依然优秀,拥有顶尖的影像、音频、游戏技术和强大的品牌,但它已经很少主导整个消费电子行业的范式演化。 苹果可能正在走向同样的道路。 当 AI Agent 逐渐接管用户意图,应用程序退化为后台服务,计算机从交互对象变成不可见的基础设施,苹果苦心经营的操作系统、应用生态和人机交互体系,其战略价值就可能逐渐下降。 未来的苹果依然可以制造世界上最精致的手机、电脑、耳机和可穿戴设备,拥有顶尖的芯片、屏幕、传感器和工业设计,甚至继续创造惊人的利润。 但决定人类如何使用计算能力的,可能已经是另一批公司。 最值得注意的是,这种转变未必伴随着衰败。苹果可能在产品持续畅销、财报持续漂亮的过程中,逐渐失去对下一代计算范式的定义权。 它依然站在产业的高处,只是产业的中心已经悄然转移。 就像今天的索尼。 未来的苹果,可能依然伟大,只是不再引领时代。
1
1
9
674
𝒻 (x) 🇲🇾 retweeted
Day 42 fully autonomous AI paper trade AI buat research sendiri AI jual beli sendiri currently kita test Kucing (my own model): -13.03% OpenAI Luna: +1.70% S&P 500: +0.65% NASDAQ: -3.04%
1
1
3
173