What a great example of an “in principle yes, but in practice no” phenomenon.
omg vindicated
communication via boulder wiggling is real
science.org/doi/10.1126/scia…
1
4
242
Breaking news:
Rich, privileged, powerful white people absolutely *love* drugs.
The idea that there could be police consequences or loss of freedom for them related to drugs just doesn’t apply to them. Drug policing is for people of colour and poor people in the hood, not them.
Whispers swirl about Melbourne’s ‘cocaine mums’—private school elites, inheritance-fueled and hiding secrets behind mahogany doors. Is the glamour masking a wild underbelly? You haven’t heard this version yet.
29
Luke Weston retweeted
This has to be the most absurd case of image fraud in academia I have ever seen. Now I have found three different papers on chicken nuggets with problematic figures. Systematic chicken nugget fraud 🐔🍗
Luke Weston retweeted
I keep forgetting that there was a time in American academia when professors had secretaries
My email is reaching the point where soon it will be the case that >50% of the messages I'm receiving are auto AI generated. The frequency of messages is also increasing rapidly. In lieu of installing an agent to manage this... I'm considering going Knuth. www-cs-faculty.stanford.edu/…
Luke Weston retweeted
“.. lenders have expressed concerns that if the group later decided to shoot the valuable chips that underpin their loans into space, it would prevent them from reclaiming and selling the semiconductors in the event of a default ..”
@FT $SPCX
ft.com/content/33b5643c-2957…
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。
Luke Weston retweeted
This is why computer vision is hard
Readers added context they thought people might want to know
The sprinkler activated due to heat from a real charcoal grill stove behind the sign at a barbecue stall (a customer was grilling skewers), not the flame graphic on it; the resort confirmed normal operation and adjusted the stove angle.
163.com/dy/article/LC4…
k.sina.cn/article_532885…
Did I read this correctly, the F-22 is dependent on PCMCIA and CompactFlash cards?
Retro Computing + Photography Enthusiasts,
The time has come to FREAK OUT and buy AS MANY CompactFlash cards as you can.
Supply is going to be redirected towards the F-22 shortly.
Current Legacy NAND is already in crazy-short supply; it's just way too profitable to move capacity to higher-value nodes. Demand on these "mature" outputs is about to go through the roof...and I don't see suppliers increasing.
It's all going to be redirected to stable, industrial contracts like this one. Consumer is going to be the first to go, because it's super low margin and at the bottom of the allocation pool.
There’s always some new thing the bureaucracy comes up with.
Like “the university has announced that DDT is prohibited and all researchers must check they don’t have any”.
159
The moral is that it’s no good to try and race constantly pushing changes, changes and changes to stay on top of a mountain of “CVE slop”.
The cure is often worse than the disease.
There needs to be triage, and only push changes that correspond to real serious risks in context.
Patch now or go offline? What's the culprit behind the near-simultaneous outages at the FAA, Microsoft, banks, Apple and the NSA, which neatly blend in what looks like one linked event? Officials reportedly blame a rush to patch bugs found by AI. The advice “patch PHP or go offline” was indeed real. And behold the java2shell vulnerability just arriving. Patching explanation is plausible as patching is not a single instant click but a process. Stuff must be tested and deployed, systems restarted. Then checking it everything works. Backlog and rushing shrinks the time for some of these. But if fixing ai-assisted identified bugs takes down broad systems for long and simultaneously then it’s serious in itself. Availability is also part of security. Prepare by keeping a few days' food and cash at home. It always helps.
Luke Weston retweeted
one redditor thinks all these software meltdowns are due to code freezes before Thanksgiving after a year of management forcing vibe coding on orgs:
Did they really just throw reinforcement learning at the fucking inverted pendulum/two wheel robot stabilization problem? That's like the most common control system design project ever, classical control theory will get you something much more stable and entirely predictable🤦♂️
Researchers have used reinforcement learning to teach a 5-tonne John Deere skid-steer loader to perform complex acrobatic manoeuvres autonomously. Trained in simulation and transferred to a real machine, the system can balance on two wheels, navigate rough terrain, spin, and control its bucket before landing safely.
The research demonstrates the potential of AI-driven automation for heavy construction and industrial machinery.
Video Credit: @leggedrobotics / @ETH
Paper: wevlv.co/4kltYys
#engineering #technology
Luke Weston retweeted
Machinist here:
While it is technically possible to machine dimensions to 1 micron of tolerance it isnt when you part is this massive.
And even so it would be completely impractical and irrelevant because even a temperature swing of a single degree will make a part as big as this this expand or contract by a double digit amount of microns.
1 micron mistake on a gear this huge, and the whole workpiece goes straight to the scrap yard
Readers added context they thought people might want to know
Thermal expansion on multi-meter steel gears exceeds ~70 microns per °C in diameter (far more for tooth spacing), and such industrial parts like slewing rings use tolerances/distortions of tens to hundreds of microns with grades like ISO 10, not 1 micron.
calibr.pro/tables/thermal…
glcncmachining.com/mass-productio…
rotis-lj.si/wp-content/upl…
Luke Weston retweeted
It seems undeniable that Microsoft has just completely hollowed itself out.
From an article by a former employee who returned in 2023.
There’s something just … I don’t know how to put my finger on it. “Rolled out a B61 to celebrate new employees” with a photo op.
Like new employees could celebrate a photo op without having to go to the trouble of dragging out the 360 kiloton bomber-dropped thermonuclear weapon
Pantex Plant rolled out a B61-13 bomb to celebrate new employees. The bomb replaces a small number of B61-12s, but with a significantly more powerful warhead (360 kilotons vs. 50 kilotons). Delivery to USAF has begun; B-2 stockpile operations from FY2028. teapawt.net/PantexPlant/status/212…
1
11
352
It feels like Microsoft is just in terminal decline and terminal enshittification.
It’s a shame because over the last four decades they *have* made some good products, they have done some good innovation, for all their controversy, they have left their mark on computing.






































