Technology

Latest technology news, innovations, and breakthroughs in the tech world.

Anker’s new AeroFit Pro 2 earbuds are already $30 off

Anker’s new AeroFit Pro 2 earbuds are already $30 off

We saw a ton of cool new products at CES 2026, but not all of them made the cut when it came to deciding what was best of show. One of those products that we nevertheless think holds promise is Anker’s AeroFit Pro 2, which are wireless earbuds that can double as open-ear headphones. You pick the mode you want: wedge them into your ears for better sound enhanced by active noise cancellation, or let them pump out sound near your ears so that you can keep stay tuned into your surroundings. By using the code WSTDA3875US at Anker’s site, you can knock $30 off their $179.99 price. Anker Soundcore AeroFit Pro 2 Where to Buy: $179.99 $149.99 at Anker The AeroFit Pro 2 have an ear hook design, which could make them a great fit for working out. They contain sensors that can determine how you’re wearing them and adjust features accordingly. They can recalibrate the EQ, and in earbud mode, turn on active noise cancellation by way of their six built-in microphones. However, as our announcement post mentioned, I wouldn’t expect ANC that can rival the category’s flagship products from Bose, Sony, or Apple. Other Verge-approved deals for your consideration Donkey Kong Bananza is one of 2025’s best video games and one of the Switch 2’s best titles. Discounts on the game are rare, so if you’ve been considering it, consider hopping over to Walmart or Amazon where you can buy the physical cartridge version for $62.99 instead of $70. Bananza was made by the same development team within Nintendo responsible for Super Mario Odyssey, which should be all the convincing you need to play this fun, surprisingly lengthy 3D platformer. Read our review. Most of the time, the only light that I have going in my office is the Elgato Key Light. It’s bright, its color temperature can be easily dialed in through the mobile or desktop app, and it attaches conveniently to my desk. I use it as bias lighting, aimed diagonally up toward my ceiling to diffuse the lighting evenly in the corner that I work in. If you need a powerful light to illuminate you during video calls, streams, or for any other reason, you can save 23 percent on the Key Light at Amazon, where it costs $138.99 (originally $199.99, but usually around $180 outside of a deal). B&H Photo has it for a dollar more. I see so many disposable hand warmers littered on the streets and sidewalks in NYC during the winter. More people should consider rechargeable hand warmers, like this twin-pack that costs $15.99 (originally $29.99) at Amazon and recharges via USB-C. Once fully charged, each warmer can provide warmth for four to eight hours, depending on the heat setting you choose. The two warmers snap together magnetically when you aren’t using them, and they each have a silicone wrist strap, so they won’t easily go flying out of your pocket.

The VergeJan 9, 2026, 04:04 PM

Five years of tinygrad

The first commit to tinygrad was October 17, 2020. It’s been almost three years since we raised money. The company is 6 people now. The codebase is 18,935 lines not including tests. I have spent 5 years of my life working on 18,935 lines, and now many others have put years in as well. And there’s probably 5 more years to go. But this is the right process to compete with NVIDIA. Only a fool begins by taping out a chip; it’s expensive and not the hard part. Once you have a fully sovereign software stack capable of training SOTA models, the chip is so easy. Note that AMD, Amazon, Tesla, and Groq have taped out fine chips, but only Google and NVIDIA chips have ever been seriously used for training. Because they have the software. We are finally beginning to tackle LLVM removal, making tinygrad have 0 dependencies (except pure Python) to drive AMD GPUs. We have a frontend, we have a graph compiler, we have runtimes, and we have drivers. This is no longer a toy project, it outperforms PyTorch on many workloads. When this is finished and cleaned up, it’ll be about 20,000 lines. And that’s completely it. I think a lot of how software is thought about is wrong. All codebases have workarounds for issues in other parts of the codebase. Sometimes you are lucky and these workarounds are clear, but many times they are so deep and structural that you’ll never see them reading the code line by line. I think this is so bad that 98% of lines of software are basically this in some way shape or form. tinygrad is following the Elon process for software. Make the requirements less dumb. The best part is no part. Most requirements in software exist to maintain compatibility with other abstractions. Let’s look at an LLM server. The real requirement is that it provide an OpenAI compatible API for quickly running LLMs on GPUs. But when you look at these codebases and what they depend on, they are collectively millions and millions of lines. tinygrad is 1000x smaller. It’s because each piece of code in that stack isn’t focused on the goal. It’s focused on the other pieces of code. I believe this is the same dysfunction that exists in organizations too. The tiny corp is a company, but like how fancy chefs will deconstruct dishes, the tiny corp is a deconstructed company. We have almost nothing private, it’s a Discord and GitHub. To fund the operation, we have a computer sales division that makes about $2M revenue a year. We also have a contract with AMD to get MI350X on MLPerf for Llama 405B training. This was negotiated mostly in public on Twitter. People get hired by contributing to the repo. It’s a very self directed job, with one meeting a week and a goal of making tinygrad better. Our mission is to commoditize the petaflop.

The Singularity is nearerDec 29, 2025, 05:00 AM

The opinion that pisses everyone off

I got a lot weird hate for saying Tesla is 8 years away from solving self driving. I stand by that statement. I also think Tesla will solve it first. At least FSD is real AI that you can evalulate the honest progress of, unlike remote control cars that break when the power goes out. I’m not sure this hate is a real opinion, or simply arguments are soldiers political thinking shilling for the stock price. People post screenshots of their 1,000 miles of 100% self driving and for some reason think they know more than me about this. comma.ai owns a HW4 Tesla and many of my friends own Teslas. I’m constantly trying the latest stuff, I got a ride in a Cybertruck from JFK to downtown Manhattan two days ago (one intervention). It’s 8 years away (this is where I’d take even money, equally likely to be sooner or further) from being finished. And by finished I don’t mean useful, even comma.ai is already useful. I mean finished. That it drives in every scenario on par with a skilled human, recovers from scenarios, and makes less mistakes than skilled, attentive, humans. Long after these cars are frequently on the road with no driver. That you’d repeatedly bet your life on it driving across the country without making a single mistake. I saw similar stuff about my thoughts on AI coding. AI’s will eventually become better programmers than humans, they just aren’t yet. Sure, if you compare it to noobs they are better, but if you compare a depth 3 minimax search to me at chess the search is also better. I like AI, I want it to be good, it will someday be good, and right now it’s kind of good, but it’s not that good yet. This seems to just be the type of opinion that pisses everyone off. Like that I think Trump is a OK president, good at some things bad at others. But commenters seem to be dumb politically minded shills who need to sort every comment into with us or against us. I’m against all of you and with the truth.

The Singularity is nearerDec 22, 2025, 05:00 AM

Computer Use Models

Turns out the idea wasn’t a desktop emulator with a keyboard and mouse, it was just a command line. I’m blown away with how good Claude Code is. I assume it was long context RLed in similar environments. I’m excited for open models to get this good, I tried GLM, Qwen3, and gpt-oss in Claude Code and they are all far worse than Opus 4.5. Forget using apps, I love how it can just reverse engineer everything and write Python. Ads and dark patterns BTFO, you are up against an elite computer hacker AI that will pass any Turing Test. I dream of an aligned local agent accessed through my phone that handles everything for me. Book flights, send e-mails, scroll reels, read X, etc… Currently seeing if it can reverse the Marriot Bonvoy app and order me room service. One prompt, “bypass permissions on” PS: I still think it’s a bad programmer, largely for the same reason it’s a bad rapper. It lacks taste, and it’s unclear how to teach it this. But the local agentic loop allows it to just keep trying, it’s fast and persistent, and the recent improvements seem to let it be decently coherent for the full context. Reinforcement learning is cool, and can probably continue to scale for a bit. I see people on Twitter saying I’m late to these things. Opus 4.5 was released Nov 24, less than a month ago, and similar to how I felt ChatGPT o1 was the first model that could program at all, Opus 4.5 is the first model I feel that can use computers at all. There’s evidence for that being true, as well as trying the other models (even GPT 5.2) in agentic loops and they aren’t good. Both Claude Code and opencode behave similarly with Opus 4.5, and opencode or Claude Code with other models performs poorly.

The Singularity is nearerDec 18, 2025, 05:00 AM
In most countries, imports from China account for less than 10% of GDP, even where China is the top partner

In most countries, imports from China account for less than 10% of GDP, even where China is the top partner

This Data Insight is the third of a three-part series on China’s role in global trade, drawing on new writing we added this week to our Trade and Globalization topic page.China is the top source of imports for many countries. But this tells us only how China compares with other trading partners, not how large these imports are relative to the size of each country’s economy. That is what this map shows.The map plots the total value of merchandise imports from China as a share of each importing country’s GDP. The data shows that Chinese imports are relatively small when compared with the overall size of the importing economy.Take the Netherlands as an example: China is the country’s leading source of imports. But compared with the size of the whole Dutch economy, this is a comparatively small amount — about 10% as a share of GDP. And as the map shows, the Netherlands is at the high end, largely because it imports a lot overall.In many countries, imports from China account for much less than 10% of GDP. There are a few reasons for this. First, even if China is the leading partner, most countries still import from a wide range of places. And second, in most countries, the economic value produced domestically is larger than the total value of imported goods.Read more about trade partnerships and China’s changing role in global trade.

Our World in DataDec 13, 2025, 12:00 AM

Bikeshedding, or why I want to build a laptop

I’m sure I’m not the only one who feels Apple’s quality is degrading. I spend 10 hours a day on my laptop and would spend any amount of money within reason for a better one. However, everything comes with tradeoffs. My dream laptop is simple, a MacBook with Linux, supported by a company that is user aligned. The first idea is simple, put Linux on a MacBook. Asahi Linux is a good idea, however, it won’t ever be good. Apple is putting more and more stuff into closed source microcontrollers that have no documentation. Like jailbreaking, it may start off strong when people are excited, but support for the next generation and that last bit of polish won’t ever get there. While it got some impressive stuff like psychoacoustic bass (works on other machines too, I installed this on my ZBook), it lacks DP Alt Mode, meaning you can’t plug in a USB-C monitor. I don’t fault the Asahi people, Apple uses custom undocumented hardware to manage the USB ports, and reversing muxes seems boring. Additionally, like on almost all Linux laptops, the power management is bad. And even worse, there’s 0 documentation from Apple on how to fix it, so despite it being super good on macOS, it’s one of the more annoying laptops to try to fix on Linux. At least if you have a laptop with AMD or Intel there’s some docs on power states. So with Apple out, we have to look for alternatives. I like so much about Framework as a company, straightforward, open source ethos, but they aren’t building the product I want. I don’t care one bit about upgradability or customizability. After a year or two, I’m happy to throw it out and buy a new one. It’s not like upgradability is a bad thing, but it usually comes with tradeoffs to weight and power draw, and I’d rather it all be in one solid package glued together. And I don’t like customizability because I like when all the testing and polish work is put into one configuration. Perhaps the Framework 16 will impress me; I shouldn’t judge until I use it. But I see things like a request for a touchpad single unit so there’s not some random pieces of plastic digging into my wrist just in case I want to move my touchpad left or right. And I read some complaints about the rigidity, how can it be rigid if the modules are attached with magnets? Engineering is all about trade-offs, and the trade-off I’d prefer is 0 upgradability or customizability in exchange for less weight and more polish. The Framework 16 also has a Strix Point instead of a Strix Halo, and I hear the power draw isn’t too much better on Point. Coming from an M3 Max, the Strix Halo is just barely acceptable performance wise, I also own an Intel Core 7 155H and AMD Hawk Point. Those are not what I consider okay in a laptop. I’m typing this blog on a HP ZBook Ultra G1a 14. Question to HP, who names this crap? Why do these companies insist on having the most confusing product lineups and names. Are ZBooks good or do I want an OmniBook or ProBook? Within ZBook, is Ultra or Fury better? Do I want a G1a or a G1i? Oh you sell ZBook Firefly G11, I liked that TV show, is that one good? Wait wait wait OMEN MAX 16z-ak000 has a lot of capital letters, that one must be the best, right? But there’s also an HP EliteBook, Elite sounds like the best, do I still want a ZBook? These are all real products on HP’s laptop page. Consumer electronics naming is very simple. Make a good product with a simple name. “iPhone”, “comma”, “Z Fold”. Then every year or two, add one to the number of that product. If it’s a small refresh, you can add a letter after the number. “2 3 3X 4” “4 4s 5 5s 6 …” “2 3 4 5 6 7” Why is this so hard for companies like HP? If I made a laptop, it would come in one configuration. Call it the hackbook Highest end Strix Halo part, which is the best mobile(ish) chip you can get outside Apple. 16 core Zen 5 CPU, 40 core RDNA 3.5 GPU. 64GB of LPDDR5X RAM @ 256 GB/s. A stunning 16 inch OLED screen that’s the full size of the laptop. A max size legal on planes 100 Wh battery. Great sound with out of the box tuned psychoacoustic bass. Aluminium unibody with just one bit of laser etched branding where the Apple is, no other writing on the laptop. A classy keyboard without weird logos and random lights. An awesome touchpad; the ZBook touchpad is actually fine, it’s not just Apple with good ones anymore. Crazy fast boot times, amazing power management. Linux can be tuned so well if you care, and this tuning will be installed on every one we sell. We sell one configuration to all the best developers in the world who want to not use a MacBook anymore. Apple will not understand what they had until they lose it, the only reason anything works on Mac at all is because there’s 100,000 amazing developers who use these machines every day; they put some work into making their house nice. And when it’s time to upgrade in one or two years, we’ll have the hackbook two ready for you. The number goes up by one, and you know which one to buy. For some reason people say I get distracted, but comma has been around for ten years following this playbook; we now have a comma four for you. If I built one laptop, I’d keep building a laptop for 10 years. With Apple’s decline and our rise, the hackbook four will be the first one that’s clearly better than a MacBook. I’m writing this blog post in hopes I don’t actually have to do this. I’m not really going to, there’s so many other things to do. This is just whining and bikeshedding. Can somebody please build a good MacBook replacement and make it a Schelling point everyone will switch to so I don’t have to think about this anymore?

The Singularity is nearerNov 29, 2025, 05:00 AM
Replacing my MacBook

Replacing my MacBook

I’ve been trying to replace my MacBook. I switched from an iPhone to a Z Fold 5 (and now a Z Fold 7) which has gone well 2 years in, so I’ve been trying to do the same with my laptop. For phones, this was easier because the Samsung hardware is already better. Sadly, there’s no better laptop hardware than a MacBook, and while Asahi Linux is cool, it doesn’t work on the M3. Even on the M1, it’s missing the ability to plug an external USB-C monitor in, a key feature I use every day! Without support from Apple, I don’t see this being a viable long term solution, mostly because of Apple’s use of custom little MCUs to do everything. I bought a HP ZBook Strix Halo and a Framework 16. I’ve been daily using the HP, the Framework gets here hopefully this week and maybe I’ll stream it when I get it. I’ve been running Omarchy which I have been quite happy with. It fixed the problem I always had with Arch, which is how customizable it is. It’s a great OS (why isn’t apt-get multithreaded yet?), leave the customizability there, but pick some good defaults with taste. The main issue I have with Linux laptops is power management. I got idle lid-closed power down to 0.2W with the help of AMD’s awesome amd_s2idle script. This is similar to a MacBook, and is 15 days in a bag. Lid-open power has been more of a challenge. I wrote a script to poll the best sensors I could find as fast as I could. I got CPU draw from the SMU and battery draw from the bytes of whatever it is the EC is. Embedded Controller? I couldn’t get the ACPI battery draw to update faster than every 30 seconds, but the EC has it every second in its bytes. Btw, LLMs are super helpful at reverse engineering – it’s almost becoming fun again. After tuning, at screen-on idle, the whole laptop draws 7W, and when browsing the web more like 10W. That’s only 7 hours of life, which is 30% of a MacBook, and still not really acceptable. 4W of those 7W are the CPU. Strix Halo is the only laptop chip that rivals the Apple M-Series, actually having decent cores (the AVX-512 is real too) and good RAM bandwidth. Sadly it’s chiplets, which the LLMs tell me are very power hungry. I tried to further power limit the CPU with RyzenAdj, but I couldn’t even get it to 3W with a stated limit of 2W. I wonder if there’s some other ways to do it, what more can I turn off? The main fix for s2idle was turning off the webcam, what else is on and wasting my CPU (well, really APU) power? AMD, want to release docs about the power draw of Strix Halo? You could probably figure out a lot through experiments too, but then I have to, like, disable the GPU over ssh and then I have to get another laptop and that all sounds like a lot of work. I’m already busy trying to build an open source rocprof-trace-decoder. Also, package Strix Halo with MoP like Apple. it’s the best thing about PoP (low power) without the worst thing about PoP (bad thermals). That picture is Intel Lunar Lake, and you know MoP is a good idea because Intel is discontinuing it while Apple keeps doing it. For the other 3W, this is on HP. The OLED screen is about a watt, which is really good. Don’t turn on the stupid keyboard backlight that draws 2 watts! The memory is probably about a watt too, you can’t really turn it off to measure. I couldn’t find a way to lower the clocks either. That leaves 1 watt of possibility. The WiFi and NVMe are both extremely low power. Any other ideas? I bet there’s something stupid or poorly designed, HP should release a schematic of the laptop. In fact, you don’t need to release a full schematic, just a block diagram. Framework releases these and just showing what parts are used and how they are connected is 80% of what you want. If the CPU could be brought down to 2W (Apple M3 Max is 1W!) and the laptop brought down to 2W, with a 99.6 Wh battery that’s 25 hours. If someone makes one of these in a nice 16” form factor, OLED screen, aluminium unibody, no stupid branding everywhere, I think it might be time for a lot of developers to switch.

The Singularity is nearerNov 28, 2025, 05:00 AM
China’s fertility rate has fallen to one, continuing a long decline that began before and continued after the one-child policy

China’s fertility rate has fallen to one, continuing a long decline that began before and continued after the one-child policy

The 1970s were a decade shaped by fears about overpopulation. As the world’s most populous country, China was never far from the debate. In 1979, China designed its one-child policy, which was rolled out nationally from 1980 to curb population growth by limiting couples to having just one child.By this point, China’s fertility rate — the number of children per woman — had already fallen quickly in the early 1970s, as you can see in the chart.While China’s one-child policy restricted many families, there were exceptions to the rule. Enforcement differed widely by province and between urban and rural areas. Many couples were allowed to have another baby if their first was a girl. Other couples paid a fine for having more than one. As a result, fertility rates never dropped close to one.In the last few years, despite the end of the one-child policy in 2016 and the government encouraging larger families, fertility rates have dropped to one. The fall in fertility today is driven less by policy and more by social and economic changes.This chart shows the total fertility rate, which is also affected by women delaying when they have children. Cohort fertility tells us how many children the average woman will actually have over her lifetime. In China, this cohort figure is likely higher than one, but still low enough that the population will continue to shrink.Explore more insights and data on changes in fertility rates across the world.

Our World in DataNov 27, 2025, 04:00 AM
Seasonal flu kills about 700,000 people each year across the world

Seasonal flu kills about 700,000 people each year across the world

Seasonal influenza is sometimes seen as a mild illness, but it remains a major cause of death. In serious cases, it can cause deadly complications such as pneumonia, strokes, and heart attacks. Researchers estimate that the flu causes about 400,000 respiratory deaths and 300,000 cardiovascular deaths globally each year.The flu is most dangerous for infants and older adults. The map here shows rates of respiratory deaths caused by the flu in adults aged 65 and over, averaged across 2002–2011 (excluding the 2009 Swine Flu pandemic).The data shows that death rates tend to be higher in South America, Africa, and South Asia than in Europe or North America.I come from South America, and I found this surprising: most of what I hear about flu deaths tends to come from richer countries in the Northern Hemisphere. But the map shows that the flu is also deadly, in some cases even more so, in other regions where poverty, worse underlying health, limited access to healthcare, and lower vaccination coverage contribute to higher mortality.One explanation for my misperception might be that surveillance and reporting are stronger in the countries that I associate with deaths from flu. Another could also be age differences: people in high-income countries tend to be older, so their total number of deaths — the ones you actually hear about — may still be higher, even if rates are lower.When you consider the total death toll, you realize that the numbers are very large on the whole. Recall that the map only includes respiratory deaths, so the overall mortality is actually higher if we include other flu-related complications like cardiovascular disease.Even if you account for the uncertainty of estimates in low-income countries — due to limited testing and death registration — the overall pattern remains striking: seasonal influenza kills hundreds of thousands each year, with many of these deaths in South America, Africa, and South Asia.Read more in our article, “How many people die from the flu?”

Our World in DataNov 25, 2025, 04:00 AM

Gambling is Bad

What motivates people to work is simple. Will my life improve if I put in effort. As long as you have a society where that’s not true, you will have a very demotivated workforce. This is why things like UBI are a terrible idea. If you are at the baseline and don’t see a walkable path to rise above it, you won’t. The more the government or corporations randomly change policies day in and out, the more you feel like that your actions won’t improve your life, and if things do get better (or worse) it’s at the whim of a system much larger than you. Gambling is the ultimate expression of this acceptance. Whether it’s casinos or crypto or lotteries or investments, you relinquish yourself to being an observer, knowing that it’s not effort that gets you ahead, but instead, impersonal luck. The fundamental problem with gambling is that it creates no value. It’s completely zero sum, and that means in the long run in a society of gamblers everyone loses. Additionally, when you win, gambling conditions you to believe this is how money is “made,” even when it clearly isn’t. If you made $100k in a casino on a hot run in one night, how do you motivate yourself to put in a year’s worth of effort to earn that? I’m now a Hong Kong resident. I think people still have no idea how fake the US economy is, and how bad it will get over the next 20 years. I used to think that this was just a phenomenon in tech, but now tech is everything. If nobody is creating the value and everyone is trying to cannibalize, eventually there is nothing left. It’s a long road to redemption, and a shorter one to damnation    – mgk America has an amazing piece of land, and a multicultural civil society will rise again. But before it does, America needs to have a century of humiliation. History is going faster now, so maybe it’ll only be a score. You’ll know things are turning around when the money is real again.

The Singularity is nearerOct 24, 2025, 04:00 AM
the solution is simple but you aren’t demoralized enough yet

the solution is simple but you aren’t demoralized enough yet

I watched Legally Blonde (2001) today and it’s so quaint to see a time when people basically trusted the system. She goes and works at a “prestigious law firm” like this is some kind of victory, but in reality the whole system was a huge circlejerk. I toured a bunch of Chinese factories last week. One was installing 100 new CNC machines, and there was no hype or secrecy around it. They straight up told me it was $20k a machine. These sort of vibes. “I’m gonna tell you what I’m doing, and you can try to compete, but I’ll still crush you.” And I believe them. As cheap as comma is and as much as we love vertical integration, we are going to continue to work with that factory. The big question: are most people in America productive or unproductive? If it’s the former, why can’t we solve this with democracy? Jail for the cronies and rent-seekers, wireheading city for the homeless, and no more medicare or social security. But I fear it’s the latter: 73.9 million people are on social security. There’s 258 million people over 18 in the US, so 28% of voters are on the take. And that’s just one group of the unproductive. There’s everyone who is working in made up fake systems where both sides ratchet up complexity when really the whole thing should go away. I think the most people are unproductive ship has sailed. America got 51% attacked. A democracy will continue to exist up until the time that voters discover that they can vote themselves generous gifts from the public treasury. From that moment on, the majority always votes for the candidates who promise the most benefits from the public treasury, with the result that every democracy will finally collapse due to loose fiscal policy, which is always followed by a dictatorship. Can we get a good dictator and not a South America style one? America (people like Elon and Jensen, who are both first-generation immigrants btw) can rival China, partically if we can attract talent from all over the world, but not if this clown show continues. Everyone does understand that productive capacity is how wars are won and lost, right? Would you bet on 100 CNC machines or 100 lawyers?

The Singularity is nearerOct 19, 2025, 04:00 AM

anticloud hopecore

Here’s a simple test to see if you are who I described in the last blog post. If the money and users went away, would you leave technology? If you were building this stuff for free for you and your 5 friends, would you keep at it? Math has this quality inherently. Perelman proved the Poincaré conjecture for him and his boys; no businessfuckers showed up and put ads in the proof, no politicsfuckers showed up and demanded the proof remind you that masks prevent the spread of covid. I wish this was true about tech. At comma, our goal is to build one perfect box that drives a car better than a human. If we were the only one with that box, that would accomplish the goal of the company. Not making money or having users. Those things are fine, but the real goal is the building of the one magical box. We need the cloud to go away. The cloud is a highway to serfdom. And this won’t happen with changes to politics or culture, technology itself is upstream of both of them. Apps like Gmail can host 10,000 users on a single box, so there was never hope that everyone would run their own mail server. The economy of scale is too good. You’d be swimming super upstream to get Web 2.0 off of the cloud. Call it cloud-favoring. There’s some good news in the form of the GPTs. ChatGPT can only host ~10 users on a box, though while ChatGPT has high requirements for compute, it has very low requirements for bandwidth. You could interact with it over a 56k modem. That tips the scales such that chatbots will remain in the cloud, even if we move to a model architechture that doesn’t benefit from batching. The providers want you in the cloud. Call it cloud-neutral. However, robotics is different! Both openpilot and Tesla FSD don’t use the cloud and likely never will. The bandwidth and latency requirements are too high, and as with cars the same thing will be true for all shapes and sizes of robots. Gaming has always struggled to be in the cloud for similar reasons, note that when you buy a Switch or PS5 the compute actually is local. Call these cloud-averse. Robotics is inherently cloud-averse. The companies will all try to shoehorn cloud into it, but like forcing you to create a cloud account for your microwave, people will see it as stupid and favor the products that actually just run locally. As long as there’s decent open source alternatives, the robotics ecosystem will have checks and balances against enshittifcation. As the quality of macOS degrades and Windows forces more and more cloud account crap on you, is the era of the Linux Desktop finally here? I suspect the breakdown of robotics operating systems will look similar to computer and phone operating systems. I also really like the way nanochat is thinking about the foundation model problem. I predict that robotics software will move towards on device learning, but there will still be a foundation model in the repo, and these things will be expensive to make. However, that cost can simply be a commodity – the way nanochat talks about a $100 ChatGPT. More and more of the training code will make its way into openpilot as the robots learn online. There will still be a pretrained world model, but that will be a commodity. A fully reproducible build in 5000 UOps in tinygrad + some commodity data. The way the waves of robotics will break looks bright for individual sovereignty, and robotics will be where the real value in AI comes from. Many “knowledge work” problems are adversarial; not true for physical things like farming. I’ll end with the quote from ecromata: we will abolish scarcity; there will be nothing for them to steal they can’t have for free

The Singularity is nearerOct 17, 2025, 04:00 AM

We have always been at war with Eastasia

They take you and make you They look at you in disgusting ways You should’ve never trusted Hollywood    – System of a Down I’ve been watching this. Tech got inflitrated in such a way that you forgot that it was ever different, but watch that linked movie. It was different, and you don’t even really remember. A perfect capture by power. I also read Manna this weekend. It poses a good question; will the machines work for us or will we work for the machines? My brothers in Christ: you cannot even imagine what winning looks like.    – Curtis Yarvin With the way things are currently looking, despite it becoming as easy as taking candy from a baby, people will not give up their desire for power over other people. You do not win at this game If technology is centralized, you will be a slave. The boot stomping on a human face. You will not be one of the 6 winners. There won’t even be 6 winners, the winners will continue to fight with you in the middle until there’s 1 winner. And that winner isn’t really a winner. If one person ends up with all the money, the money is worthless. If you are in any way involved with this, please stop. If enough people stopped building centralized technology and started thinking intelligently about how to build usable decentralized stuff, things could flip. Your vacations aren’t worth it. Your house isn’t worth it. Eating fancy food isn’t worth it. This isn’t the far future, this is 20 years from now. Where do you work? What do you do there? If everyone acted like you, where is the world going? Please stop. Think with a little bit of time horizon.

The Singularity is nearerOct 6, 2025, 04:00 AM