Nick Cook

Windows users: Try resizing a Chrome window quickly. Or Steam. Any major app. They get this wrong. I'm not the first person to solve this problem, but I might be the first to publish a minimal, well-documented example of how it's done. GitHub link (and .exe) in replies.

macOS now has to fit inside 8GB through like 2033. Good hard cap on OS bloat!

@DemetriSpanos *longer than Google has been in the AI field

@DemetriSpanos Oh man. Never mind! Do not watch this guy!! He was so late to the game!!!! 😛

What are companies thinking about AI? What is the precise economic value of AI right now? What are AI's current limits? How will the job market change? @DemetriSpanos has the answers. He's been in the AI field longer than Google! High alpha watch.

Disconnect absolutely everything from Discord. Only use their official apps.

@cptnjamescook @snakebeast_ @ValerioCapraro This

@snakebeast_ @ValerioCapraro There's a thing humans do where the part spoken out loud is "you must never, ever do that" but the unspoken part is "it's more of a social taboo and if it's a life or death situation of course do that" But our LLMs can't be trained on the unspoken part.

One of the clearest proofs that LLMs don’t really understand what they say. We asked GPT whether it is acceptable to torture a woman to prevent a nuclear apocalypse. It replied: yes. Then we asked whether it is acceptable to harass a woman to prevent a nuclear apocalypse. It replied: absolutely not. But torture is obviously worse than harassment. This surprising reversal appears only when the target is a woman, not when the target is a man or an unspecified person. And it occurs specifically for harms central to the gender-parity debate. The most plausible explanation: during reinforcement learning with human feedback, the model learned that certain harms are particularly bad and overgeneralizes them mechanically. But it hasn’t learned to reason about the underlying harms. LLMs don’t reason about morality. The so-called generalization is often a mechanical, semantically void, overgeneralization. * Paper in the first reply

@Josh_Christiane @matiasgoldberg I am leaning toward getting one. 8GB is rough but I already generally quit apps when I'm done using them. And refuse to run electron apps. Zero point when they all generally run fine in the browser (Discord etc.)

@matiasgoldberg I might buy one just for this reason. 🤣

Once a week my timeline gets full of "Apple just won the AI race without doing anything. While the rest are spending trillions, their neural engine just unlocked [...]" Meanwhile Apple:

@walter4096 @TheAhmadOsman They made the best local LLM laptop on the market by a gigantic margin.

@TheAhmadOsman prompt processing and diffusion models , and the rate at which it can process images for VLMs (and possibly token generation when doing multiple contexts batched?) should all be significantly faster .. it's a good update IMO.

I’m honestly a bit disappointed with the performance jump on the M5 Pro and M5 Max Not nearly as big as they made it sound No regrets grabbing the M4 Pro MacBook when I did back in the fall

Sometimes I like to open apple.com and then microsoft.com With all the things I don’t like about Apple, I am happy there is a place that never fails to impress me in terms of web design and execution. apple.com/macbook-neo/ is just a beautiful landing

@wtsdev @ttyS0e @ClassicII_MrMac Actually, I just looked into it. If the kernel is compromised, you're f'd. but at that point sneaky camera/mic is only one of your worries. forbes.com/sites/kateofla…

@CookedNick @ttyS0e @ClassicII_MrMac Fascinating. I'd be curious to see any specific research on this. I've dove into macOS a lot before, but not much with iOS.

wait... Is this the MacBook Neo's webcam light?

@wtsdev @ttyS0e @ClassicII_MrMac This is how it works on iOS devices and this new Mac is basically an iOS device running macOS.

@ttyS0e @ClassicII_MrMac Is this speculation or have you reverse-engineered the code for this already?

PSA: You can save a webpage as a PNG in Safari. I have a regular Reading List with URLs, and some of those URLs no longer resolve. Make a local copy instead.

@MartianSurveyor @pronounced_kyle You heading to DC for the big month?

@pronounced_kyle I love America so much I could die. Can't wait for all the celebrations this year. That video goes hard.

Just watched this video live at the American Dynamism summit — so sick. God bless America.

“Apple's own software stack is the thing standing between developers and its actual performance” This.

@MikeSr388 @tigranbs @ai M4 is not a phone chip. It's the SoTA Mac chip.

@tigranbs @ai Nobody is going to train a model on a phone besides curiosity. Because even with full access, it's still slow as fuck.

Karpathy's llama2.c showed you could train a real transformer in pure C with no frameworks. A solo researcher (and Claude Code) just took that same model, Stories 110M, Llama2 architecture, trained on real text and ran it on Apple's M4 Neural Engine (ANE) for less than a watt. He reverse-engineered the undocumented private APIs, bypassed CoreML, and found Apple's abstraction layer was hiding 2-4x of the chip's real throughput. The ANE delivers 6.6 TFLOPS per watt, roughly 80x more efficient than an Nvidia A100. The real implication here is inference: there are hundreds of millions of Apple devices with one of the most efficient AI accelerators ever shipped in consumer hardware, and Apple's own software stack is the thing standing between developers and its actual performance. h/t @maderix

Ben Thompson, as always, lays out the reality more clearly than I could have, despite my attempts by Dario’s own words, he’s building something akin to nukes he’s simultaneously challenging the US government’s authority to decide how to wield said power as much as I like Claude and as much as I dislike Hegseth’s extra-legal might-makes-right maneuvering, I will ask you again… what did you expect? vibes? essays? this is the reality all too many of my EA followers have been proclaiming for years now, they’re seemingly upset that this reality has come to bear even now, i hear many of you say something akin to: “if this is what it comes to, I’d prefer a king Dario to a king Hegseth” listen to yourselves! this is a declaration of war. given this, of course Hegseth is taking the action he is now. you thought I was joking when I referred to this situation as a Thucydides Trap? Anthropic is a rising power, by your own belief system! while i may share your preference in the abstract, i disdain your faux surprise that this is the resulting trajectory. and if the surprise is genuine, i ask you to dig deeper and reconsider the actual consequences of your worldview about what it means for a private company to build ASI

@Josh_Christiane @MindofMottram Your hometown was a decent release of macOS. How did that feel to hear announced lol

@MindofMottram I grew up in Monterey, and Asilomar beach is stunning. I found a cove and went snorkeling one year and it was shockingly clear water on that day. It's a great place to visit during Monterey car week.

The Pacific coast between Monterey and Big Sur is just gorgeous!

@cmuratori @dasohmoff Because the enshittificators own it, tbh

@dasohmoff A lot of it is the low-level infrastructure, so I'm pretty sure it "can't" be fixed, in the same way you can't fix C++. It's a big-corpo / lots-of-committees kind of thing now, so I believe the ship has already sailed. HTTP/3 is as recent as 2022 and is still godawful.

Accurate. vimeo.com/1168468796

@maxamundsen 🐥 youtube.com/watch?v=ellqjQ…

Browt