This goes out to everyone, but specifically the LGBT community, but especially the trans community.
Shit's getting concerning right now, not just in the US but also in the UK, and as you can see with how the UK went from being the best place in Europe for LGBT rights to going to rolling back transgender rights faster than they burnt documents relating to their Imperial Atrocities, I think we need to have a talk.
This goes out to everyone, but I want transgender people and Queer people generally to sit up and listen because the rest of the acronym is next. You need to start caring about your privacy, especially in regards to technology, because that will come back to bite you and it's the reason they're so gung ho on AI: Surveillance.
Here's some shit you should (at least try) and do:
Keep a paper diary and a paper notebook: A paper diary isn't merely theraputic, it's also very good to keep a historical record that isn't electronic because the government and their corporate allies can't fuck with it, and you can say all the things that would get your house raided either now or in the future when their "targets" change.
If you are going to do shit on a computer, keep shit either offline on your own computer or on a NAS if you have one. Replace Google Docs with LibreOffice, which is free and open source and cross platform. Keeping shit local is your friend. If you're keeping notes on your phone, try something like Notally, something that doesn't call back to your phone's servers. CoMaps let's you download OpenStreetMaps on your phone so even when you are offline, you have a fully functioning map app.
Encrypted Comms: No, WhatsApp is not safe, it's run by Meta and they want your fucking data and they will give it to whoever pays them. Get Signal: It's Open Source, Run by a non-profit who's whole goal is secure communications. It's got all the shit you'd want and more.
If you can, get off Windows or Mac and move to Linux: Linux isn't complex, Linux Mint is easy, and it's user friendly. It's not 2006 anymore.
Get a VPN. It's not that scary. The UK and parts of the US are already demanding you give away your ID to view social media and are blocking places that can't/won't comply. Imgur, the Image Hosting and meme sharing website is blocked in the UK and the Online Safety act has also forced the shit down of Support forms for Illnesses and Football supporter forums because they couldn't comply. I recommend Proton, which also have a good Email Service and a suite of things like cloud storage, a Password Manager and even an AI chat bot if you really can't live without it, but again, Local is best.
Keep Tor in your mind: yes that might sound like overkill but you don't know what might happen in the next government cycles.
Shit's getting bad, start thinking about your privacy.
I'm on a tour with my new book, the international bestseller Enshittification: catch me next in Miami, Burbank, Lisbon! Full schedule here.
While "tech exceptionalism" can be a grave sin (as with the "move fast and break things" ethos that wrecked so much of our world, especially its labor markets), there are ways in which tech is truly exceptional, in the sense of bringing forth capabilities and affordances that have never existed before, in all of human history.
One obvious way in which tech is exceptional: its flexibility. Digital computers are "Turing-complete, universal von Neumann machines," which means that they are engines capable of computing every valid program. They are truly general purpose. We have many other general purpose machines, of course, but they are simple things, like wheels. Computers are unique in that they are both complex and universal, and every computer can run every program. Just as we don't know how to make knives that only cut in beneficial ways, we also don't know how to make computers that only run desirable programs.
Every computer can run every program, including ones that the user doesn't want (viruses), or that the manufacturer doesn't want (ad-blockers). No one knows how to make a computer that is almost Turing-complete. There's no such thing as "Turing-complete minus one." We can't make a computer that only runs the programs the manufacturer has authorized – all we can do is criminalize the act modifying your own computer to do what you tell it to, even if the manufacturer objects:
I've devoted a lot of my life to exploring the policy implications of this amazing fact, but that's not the only amazing, exceptional thing about technology. There's at least one other way in which modern digital technology has produced something that is genuinely, civilizationally novel: encryption.
Encryption – scrambling data so that it can only be read by its intended recipient – is an age-old project for both the authorities (who used ciphers to keep their secrets safe since the time of the Caesars) and for those who would overthrow them (revolutionary movements have always used codes to protect themselves from the authorities they sought to dethrone).
But WWII ushered in a new era, in which encryption (and attempts to break it) went digital, as Alan Turing and the codebreakers of Bletchley Park turned themselves to a computer-aided mathematics of scrambling and descrambling. In the decades that followed, a modern form of encryption emerged, one that was powerful beyond the wildest dreams of the Caesars and their revolutionary adversaries.
Modern, computerized encryption can scramble data to the point where it is literally unscramblable by an unauthorized party. In the eyeblink moment between you pressing the camera button on your phone and the resulting image being saved to its mass storage, the bits that make up that image are scrambled so thoroughly that even if every hydrogen atom in the universe were made into a computer, and even if all those computers were put to work guessing at the key, we would run out of time and universe before we ran out of keys.
Even futuristic, experimental technologies like quantum computing that may revolutionize codebreaking are also revolutionizing scrambling itself:
https://signal.org/blog/pqxdh/
The history of encryption is seriously fraught. Until the early 1990s, the NSA classed working encryption as a munition and banned civilian access to a whole branch of mathematics. It wasn't until Cindy Cohn – then a lawyer for the Electronic Frontier Foundation, now its executive director – convinced a court that the First Amendment protected the right to publish computer code, that we were all able to gain access to this essential technology, which today safeguards your messages, files, banking transactions, and the software updates for your car's brakes, your pacemaker, and the informatics on airplanes. Cohn has announced her retirement from EFF in 2026, and while she will be sorely missed, we do have her memoir, Privacy's Defender, to look forward to:
The legalization of encryption was a starting gun for the internet itself, as true information security entered the picture and pervaded every part of service design. Every security crisis, every scandal (e.g. Snowden), jolted the effort to encrypt the internet forward, and in this way, much of the internet lurched into a state we can call "encrypted by default."
But even as this privacy-preserving technology was perfected and made ubiquitous, something weird and contradictory happened: mass surveillance also took off online. The ad-tech industry – and its handmaidens, the data-broker industry – rigged the game so that our private activities were only encrypted in such a way as to defend their privacy, but not ours. Our data is encrypted in transit to the servers we interact with, and when it is at rest on those servers' mass storage devices, but it is not encrypted in a way that prevents companies from data-mining it, or decrypting it and selling it on or giving it away or combining it with surveillance data purchased or traded from others.
This isn't an inevitability: it's a choice. The ubiquity of surveillance in the age of encryption is a policy choice. The reason companies don't encrypt our data so that they can't use it against us is because they don't have to. Congress hasn't updated American consumer privacy law since 1988, when they passed a law that prohibits video store clerks from disclosing our VHS rentals:
Why hasn't Congress updated our privacy rights since Die Hard was in theaters? Because American cops and spies love commercial internet surveillance. Tech companies and data brokers are a source of fine-grained, off-the-books, warrantless surveillance data that the American state is totally dependent on. There is no difference between "commercial surveillance" and "government surveillance" – they are a fused symbiote and neither could survive without the other:
Governments have hated encryption since the Clinton era, and have been attempting to subvert it since computers came in beige boxes and modems screamed in agony every time you tried to look at the internet:
It's no mystery why we don't have federal bans on facial recognition – if we did, ICE wouldn't be able to nonconsensually, warrantlessly steal your face and store it for 15 years (at least):
Why did the EU allow Ireland to facilitate mass surveillance for a decade after the GDPR's passage? Because European authorities also hate encryption and say that it is a "totally erroneous perception that it is everyone's civil liberty to communicate on encrypted messaging services":
The internet could be the most privacy-preserving communications medium in history. Instead, it has ushered in an era of nightmarish surveillance. This isn't a technology problem. It's a policy problem. Criminals spy on us online because our governments wanted to spy on us online, so they let corporations spy on us online.
Imagine what the internet would look like today if, in its early regulatory moments, our elected representatives had demanded privacy, rather than trying to ban it. Sure, some corporations would have spied on us anyway, and criminals would have done their best to compromise our privacy, but criminals and rogue firms wouldn't have been able to attract capital to engage in conduct that was likely to give rise to massive fines and criminal prosecutions for violating the privacy laws Congress never bothered to write for us.
Think of it this way: sure, there are e-commerce sites that are just scams, that take your money and never ship you goods. Those sites don't have IPOs, they're not listed on stock exchanges, and they get shut down or blocked. They exist in the shadows, not in the light. Imagine if that was the kind of commercial surveillance industry we'd gotten: marginal, shadowy, illegal, forever on the run. There would still have been some bad privacy invasions, but these would have been crimes, not Harvard Business Review case-studies:
(And before you email me about that one time Paypal closed your account and kept your money or Ebay wouldn't give you a refund, sure, that's right, those things suck, and the companies should face penalties for them, but their business model isn't stealing money from their customers; but Google and Meta and Apple's business model is 100% stealing data from their customers.)
Instead of treating data theft the way we treat monetary theft, we're now increasingly treating monetary theft like data theft. The legislative formalization of cryptocurrency will now allow companies to steal your money with the same blissful lack of consequence as Google faced for stealing your private information:
https://www.citationneeded.news/issue-89/
We're rounding the corner on a decade since the beginning of the fight against Big Tech, and the efforts to cut it down to size. These keep foundering on the political economy of crushing an all-powerful monopolist – namely, that it is all-powerful.
Breakups, taxes and fines are all forms of redistribution, which seek to address the harms of monopoly after the monopoly has been formed. The failure to make privacy protections as inviolable as financial protections is a missed opportunity for predistribution. Bans on data collection, mining, and sale would have prevented these monopolies from forming in the first place. Predistribution is far more effective than redistribution:
It's amazing to realize that the privacy-invading internet has somehow beaten the encrypted internet. It's crazy that the only entity that will promise to encrypt your data beyond the reach of a data broker, an ad-tech giant, or a government is a ransomware criminal, who will also encrypt your data beyond your reach:
It didn't have to be this way. This wasn't a technology failure. It wasn't a commercial failure. It was a policy failure. Since the 1990s, whenever push came to shove, governments decided that they would rather preserve their ability to spy on us than keep us safe from private spying.
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
🔬 Physicists finally observe magnetic vortices predicted 50 years ago
🔬 Physicists finally observe magnetic vortices predicted 50 years ago
An international team of researchers has experimentally observed magnetic vortices in an atomically thin material — a phenomenon predicted by theory in the 1970s.
The study, led by scientists at the University of Texas at Austin, confirms key predictions of two-dimensional magnetism.
The experiment used an ultrathin crystal of nickel phosphorus trisulfide (NiPS₃) only one atomic layer thick.
When cooled to about –150 °C, the material entered the Berezinskii–Kosterlitz–Thouless (BKT) phase.
In this state:
• atomic magnetic moments form tiny vortex structures
• vortices appear in pairs rotating in opposite directions
• the structures remain surprisingly stable despite being only a few nanometers across
When the temperature was lowered even further, the researchers observed another transition — the six-state clock phase, where magnetic moments can point in only six symmetric directions.
For the first time, scientists captured the full sequence of predicted phase transitions in a single material system.
This discovery could have important technological implications.
Magnetic vortices may enable:
• ultra-dense data storage
• nanoscale electronics
• quantum technologies
• next-generation magnetic sensors
The next challenge is to find materials where similar effects occur closer to room temperature, which could bring vortex-based technologies into real devices.
I was raised with the belief that trust isn’t given—it’s earned. When Turkey, a NATO ally, lost the U.S.’s trust in 2019 by purchasing Russian S-400 air defenses, it was rightfully banned from receiving F-35s due to the risk of allowing Moscow to collect sensitive data on the fighter’s stealth signature. Now, according to The New York Times, Trump is likely to reopen the sale to Ankara, and for the life of me I can’t imagine how the Turks earned it.
Since 2019, they have kept every last unit of their S-400s—up to 200 missiles, many still in their shipping containers. They have remained the only NATO member that refuses to sanction Russia over Ukraine. They have hosted Hamas leadership in Istanbul and laundered billions for Iran, while Erdogan openly threatens Israel with destruction and, in Benjamin Netanyahu’s words this week, “talks openly about conquering Jerusalem.” They continue to occupy half of Cyprus—a fellow EU member—and to menace Greece, another one. Not to mention, this is the same country Trump himself claims he had to stop from joining Iran in the last conflict.
If trust isn’t earned, it seems it can be bought. The salesman on the account is Tom Barrack, the billionaire real estate investor, longtime Trump friend and ambassador to Ankara. Barrack doesn’t talk about Turkey’s conduct; he talks about the deal. The seven-year ban his own boss imposed? “Insane,” he says. The legal requirement that Turkey actually give up its Russian missiles? A technicality to be workshopped—the S-400s are inactive, he assures us, so possession is just paperwork. This is, in his own words, “classic Trump deal-making.” And when trust itself came up, Barrack showed us exactly what the word weighs in his ledger: Describing the Israel-Hezbollah ceasefire, America’s ambassador announced that “everybody has been equally untrustworthy.”
Netanyahu appears to have known about the impending sale and tried his best to derail it. During an interview with Fox yesterday, Netanyahu, unprompted, launched into an attack on selling the technology to “a regime infected by the Muslim Brotherhood, an extreme movement that hates America and chants ‘Death to America.’”
There was already early warning of this turn in late June, when the administration notified Congress of its intent to proceed with the sale of F110 engines to power Turkey’s indigenous fifth-generation fighter, the Kaan. Speaking alongside Vice President JD Vance, the president signaled he was prepared to make Erdogan “very happy” on both the engines and potentially on readmitting Turkey to the F-35 program.
A “very happy” Erdogan sends a chill down my spine; thankfully, Congress is poised to ruin his mood. The law bars F-35 transfers so long as Turkey possesses the S-400, so the administration is hunting for a workaround: hand the systems to some unnamed third party (a mechanism, one official admits, that hasn’t been worked out), or pull a few key parts and declare them “inoperable.”
And then there’s Israel. U.S. law obligates Washington to preserve its qualitative military edge, the guarantee that Israel outflies every rival in the region. Handing the region’s premier stealth fighter to Israel’s next regional rival certainly undermines that supremacy. To preserve the gap, Israel may receive further access or compensation of that nature.
Even if Erdogan gets his announcement at this week’s NATO summit, the sale itself must survive a one-to-two-year gauntlet: some legal sleight of hand to make the S-400s “disappear,” lifting sanctions that remain in place, formal notification to Congress, and bypassing resolutions of disapproval lawmakers are already promising. Then Turkey joins the back of Lockheed’s queue, where recent customers have waited four to five years from contract to first delivery, meaning that beyond the handful of jets Turkey paid for in 2019, which are collecting dust in U.S. storage, no new F-35 is likely to fly over Anatolia before 2030. That gives Israel time—time it can use to throw up roadblocks for Turkey and to get itself into the F-47 program.
China is nearing completion of the world’s largest hybrid pumped hydro storage system at Lianghekou Dam, reinforcing its dominance in long-d
The 4.2-GW hybrid project at Lianghekou combines conventional hydropower and pumped storage, creating the world’s largest system of its kind.
Pumped hydro will absorb excess generation from 7 GW of regional renewables, strengthening grid stability amid surging energy demand.
As lithium-ion batteries dominate short-duration storage, China is positioning long-duration technologies like pumped hydro at the center of its clean energy strategy.
The world's largest pumped hydro energy storage facility is nearing completion in China. The megaproject will bring China one step closer to its goal of being the world's first and preeminent electrostate, and will advance the nascent but rapidly growing long-duration energy storage sector.
The storage project being built at the Lianghekou Dam in Western Sichuan plateau will have four units of 300 megawatts (MW) each when completed, making it the largest facility of its kind. The project will house a conventional hydropower plant in addition to a pumped hydro energy storage facility. Pumped hydro works by moving water uphill when excess energy is being generated. When that energy is later needed, the water is released to flow downhill, spinning a turbine to create electricity in the process.
The hydropower plant and the pumped hydro facility will together have a capacity of 4.2 gigawatts (GW), amounting to "the world's largest hybrid pumped hydro storage system," according to a recent report from Interesting Engineering. "The pumped hydro facility is designed to absorb green energy from the 7 GW of renewable energy plants being built on the plateau, making it the 'super power bank' of the region," the article goes on to describe.
Pumped hydro is a relatively simple technology with a long history in China. It's a proven technology with long-term storage potential, making it indispensable in China's increasingly electrified economy. But the technology also has its downsides. Environmentalists often take a negative view of hydro megaprojects, as dams are extremely disruptive to ecosystems. And hydro systems are vulnerable to climate change, as doubt can severely impact their output - as seen in Sichuan in 2022. But the Chinese government clearly feels that the benefits outweigh the risks.
Even before the addition of this massive energy storage capacity, China has been leading the global energy storage sector for years. The Chinese government has heavily incentivized the sector's expansion as part of a greater strategy to shore up domestic energy security while also building up Beijing's dominance as a global clean energy superpower. Being at the forefront of energy storage development and deployment is a key part of the country's bid to become the world's first electrostate.
On a global level, the energy storage sector is seeing a meteoric rise, driven in large part by demand from utilities and data centers and the rapid growth of variable energies including wind and solar. Soaring rates of energy demand and increasingly complex energy flows are placing unprecedented strain on global energy grids and creating volatility in energy markets. Energy storage is a critical stopgap to fixing these issues and providing backup for greater energy security and grid resilience.
As a result, energy storage is gearing up to be "clean energy's next trillion-dollar business." While the current energy storage landscape is dominated by lithium-ion batteries, this technology has some major drawbacks, including geopolitically fraught value chains (also dominated by China) and short energy storage duration. While lithium-ion batteries can hold onto energy for a matter of hours, balancing energy flows from wind and solar will require storage systems that can hold onto excess energy for full days and even full seasons.
Long-term energy storage, such as pumped hydro, will therefore be increasingly indispensable in the coming years as the global clean energy transition continues to pick up steam. Despite major setbacks in clean energy policy in the United States, the world's largest economy, solar and wind energy are continuing to go gangbusters on a global level as renewables have simply become too cheap to fail.
And even in the United States, the energy storage sector is continuing to show strong performance "despite policy headwinds". Grid battery installations hit a record high in 2025. However, far greater levels of investing and policy support will be needed if the United States has any hope of competing with China, where energy storage innovation, development, and deployment are concerned.
NASA's SpaceX CRS-34 Dragon returns packed with space station science
Scientists await a big splash in the Pacific Ocean as one of the most research-packed Dragon spacecraft to date returns, completing the 34th SpaceX commercial resupply mission to the International Space Station for NASA. Biological and materials samples, along with tested hardware, are heading back to research teams on Earth for further analysis, advancing NASA's work to prepare humans for exploration beyond low Earth orbit and to deliver benefits back home.
Tiny cells, huge health insights
Some samples returning are for NASA's Hematopoietic Stem Cell Expansion in Space: Pathfinder Investigation (InSPA-StemCellEX-H2), which seeks to use the microgravity environment to scale up the production of stem cells. On Earth, lab-produced blood stem cells lose their ability to form different cell types, like red and white blood cells that are critical to treating patients with certain blood diseases and cancers. In microgravity, researchers believe this ability will be better preserved while also growing these stem cells in greater numbers. The returning samples will undergo further analysis to determine whether space-based efforts produce larger quantities of enhanced stem cells suitable for clinical use.
The team behind NASA's Streptococcus pneumoniae (Spn) Infection of Cardiac Tissue (MVP Cell-09) experiment is awaiting the return of stem cell-derived heart tissues that were intentionally infected with a pneumonia-causing bacterium as part of ongoing microgravity research. Pneumonia increases the risk of heart disease, which is not fully understood. Because bacteria tend to become more active and virulent in microgravity, this experiment could amplify their effects, making it possible to detect cellular responses that cannot be observed on Earth.
NASA's Megakaryocyte Flying-One (MeF1) samples are returning to Earth to help researchers understand how large cells found in bone marrow, known as megakaryocytes, and the platelets they produce adapt to spaceflight. Megakaryocytes and platelets play important roles in the formation of blood clots and immune responses. The returning samples, including those taken from astronauts, could show how the human immune system reacts aboard the space station and help prepare for future exploration missions.
Driving design enhancements
Many spacecraft use cryogenic fuels for propulsion, but temperature swings in space can cause these extremely cold fuels to slowly evaporate and escape their tank, reducing fuel efficiency and complicating mission planning. NASA's Zero Boil-Off Tank Noncondensables (ZBOT-NC) investigation aboard the station studies how gases that do not condense into liquids at cold temperatures affect pressure control and fluid behavior in propellant tanks. Hardware returning aboard Dragon, including drives containing fluid-physics data, could help validate models and contribute to the design of more efficient cryogenic fuel storage systems for long-duration missions.
Semiconductor research samples from NASA's In-Space Production of Semimetal-Semiconductor Composite Bulk Crystals in Microgravity (SUBSA-InSPA-SSCug) investigation are returning to Earth for further analysis. This study manufactured semimetal-semiconductor composite alloy crystals in space, which have applications in many electronics, including sensors and lasers. Researchers believe microgravity could enable the production of significantly larger and higher-quality crystals, supporting the development of next-generation semiconductor technologies.
Innovative medical research mix
NASA's DNA Nano Therapeutics-3 research team will receive tiny, space-assembled DNA-inspired materials that are combined with medicines to create active cancer treatments. Producing these treatments in microgravity can improve how well they perform in the body. This research could improve patient outcomes by helping therapies reach tumors more effectively, stay in the body longer and improve medicine release.
Tissue models of the brain, heart, liver and kidney that were tested with novel RNA-based medicines as part of NASA's InSPA-Sachi Nanoligomer investigation are also returning. Microgravity can accelerate aging and disease processes, giving researchers a unique environment to better observe how well these new drugs work on different organs ahead of clinical trials.
Samples from ESA (European Space Agency)'s Green Bone investigation are returning to Earth to help researchers understand how bone cells grow and develop on a new scaffold made from wood. Designed to mimic real bone, this scaffold was tested in microgravity to understand its ability to heal defects and fractures. Because living in microgravity simulates conditions like osteoporosis, a skeletal disorder that affects millions of people worldwide, the results could help treat patients with these fragile bone conditions.
NASA's 3D Bone Marrow Analog research team will analyze the returning 3D-printed tissues that mimic parts of the bone marrow. Spaceflight can cause aging-like changes, including bone and muscle loss. To investigate potential countermeasures, these tissue models were exposed to small vibrations aboard the space station to simulate exercise. After the samples return to Earth, researchers will measure bone-like mineral formations and observe cellular and genetic changes. Findings from this investigation could help develop new strategies to maintain astronaut bone and muscle health during future long-duration missions.
In the United States, more than 900,000 knee cartilage injuries occur annually, with many requiring surgery. NASA's InSPA-Auxilium Bioprinter-Cell Printing is investigating how to treat these injuries and is returning 3D-printed cartilage tissue samples from the space station. This investigation uses the orbiting laboratory's unique microgravity environment to bioprint cartilage tissues with more evenly distributed cells compared with those printed on Earth. The results could help produce higher-quality cartilage prints to treat joint injuries.
Scientists successfully develop half metal material that conducts single-spin electrons
Researchers at Forschungszentrum Jülich have successfully created the world's first experimentally verified two-dimensional half metal—a material that conducts electricity using electrons of just one spin type: either "spin-up" or "spin-down." Their findings, now published as an Editors' Suggestion in Physical Review Letters, mark a milestone in the quest for materials enabling energy-efficient spintronic that go beyond conventional electronics.
Half metals are key to spintronics: Unlike traditional conductors, half metals allow only one spin orientation to pass through. This makes them ideal candidates for spintronics, a next-generation information technology that leverages both the charge and the spin of electrons for data storage and processing. In conventional electronics, on the other hand, only the charge is used.
As they scramble to keep their systems online, AI companies are making things expensive for the rest of us. Large language models such as ChatGPT and Claude are so resource-hungry that tech companies may be purchasing 70 percent of the world’s supply of high-end computer memory, causing a shortage. As a result, the prices of computer memory and storage are skyrocketing: Hard drives that I bought for my reporting two years ago for $350 each were $800 when I checked two weeks ago, and are now out of stock. The prices of some laptops have gone up as much as 50 percent, and low-cost computers are being hit the hardest. Affordable entry-level computers may “disappear by 2028” according to one forecast. And the memory shortage is expected to continue for years.
The memory is being put into data centers, which tech firms are expanding at incredible speed. They are planning to multiply total U.S.-data-center capacity by a factor of eight over the next few years. The demand for electricity at these sites is already so great that some companies are repurposing jet engines to power them.
The problem is not simply that AI is being deployed so widely or quickly. Other computer technologies have seen similarly massive growth without triggering such a large spike in electricity or a shortage of computer components: Video and music are now streamed around the globe, accounting for many terabytes of internet traffic daily; the smartphone boom required the manufacturing of billions of devices that are now transferring huge amounts of data; billions of household devices are also now part of the Internet of Things; and whole industries have moved their operations to cloud software, which is hosted not in the sky but in, yes, data centers.
The problem with generative AI, in the industry’s own jargon, is that it does not scale. The cost of growing from, say, a thousand users to a million is a key factor that venture capitalists examine when they evaluate start-ups. They want to see that the cost of adding each new user decreases over time, so that the company can support millions of users and make increasing profits. This is achieved partly through the careful engineering of computer systems that can efficiently handle more users who want to post photos, hail Ubers, or stream music.
With generative AI, the work of building efficient, scalable systems has not been done. And the problem is exacerbated by the ever-larger generative-AI models, which have grown from 175 billion parameters in 2020 to more than 1 trillion today, according to independent estimates (the actual sizes of the models powering products such as Claude and ChatGPT are secret). The large in large language model should not be a selling point. But the industry’s observation that bigger models tend to outperform smaller ones has given rise to a totemic belief in “scaling laws” that suggest any problem can be solved by simply making models bigger. “Maybe with 10 gigawatts of compute, AI can figure out how to cure cancer,” OpenAI CEO Sam Altman wrote on his blog in September.
Yet the returns are diminishing. The bigger an AI model is, the less it improves with each added parameter, and so it must be made bigger at a faster rate just to sustain steady progress. I asked a few AI researchers whether they could name any other real-world software that scales so poorly. None of them could think of any. Even outside the world of software, it’s hard to find a comparable example, given that economy of scale is the principle that has made light bulbs, cars, and clothing so affordable. By economic and engineering measures, generative AI might be the worst technology ever deployed.
Discover Top Posts Tagged with #next-generation data storage technologies | Tumgag
