US president Donald Trumps massive, sweeping tariffs sent global stock markets tumbling on Thursday, setting the stage for a worldwide trade war and ratcheting up the dangers of a punishing recession. Experts fear that the US cleantech sector is especially vulnerable to a deep downturn, which would undermine the nations progress on reducing greenhouse-gas emissions and undercut its leadership in an essential, growing industry. It would be hard for me to think of cleantech or climate tech sectors that arent facing huge risks, says Noah Kaufman, senior research scholar at the Center on Global Energy Policy at Columbia University, who served on the Council of Economic Advisers under President Joe Biden. I think were a country without a federal climate strategy at this point, with an economy headed in the wrong direction, so I dont see a lot of reason to be optimistic, he adds. Indeed, there are mounting challenges and rising risks across the cleantech and climate tech sectors. How deep and wide-ranging the impact of the economic changes could be depends on many variables and on reactions still taking shape. In particular, the negotiations underway in Congress over the budget will determine the fate of subsidies for electric vehicles, battery production, and other support for clean energy. Many of those programs were established by former president Bidens signature climate law, the Inflation Reduction Act. Beyond the tenuous government support, any slowdown in the broader economy threatens to tighten corporate and venture capital funding for startups working on carbon removal, synthetic aviation fuels, electric delivery vehicles, and other technologies that help companies meet climate action goals. In addition, Trumps tariffs, particularly the now 54% levy on Chinese goods, will push up the costs of key components for many businesses. Notably, the US imported $4 billion worth of lithium-ion batteries from China during the first four months of last year, so the tariff increase would impose a huge tax on products that go into electric vehicles, laptops, phones, and many other devices. Higher prices for aluminum, steel, copper, cement, and numerous other goods and materials will also drive up the costs of doing all sorts of business, including building wind turbines, solar farms, and geothermal plants. And if China, Canada, the European Union, and other nations respond with retaliatory trade measures, as is widely expected, it will also become harder or more expensive for US companies to export goods like EVs or battery components to overseas markets. Even traditional energy stocks took a beating on Wall Street Thursday, out of fear that any broader economic sluggishness will drive down electricity demand. Trump administration cuts to the Department of Energy and other federal programs could also take away money from demonstration projects that help cleantech companies test and scale up their technologies. And if Congress does eliminate certain subsidies in the Inflation Reduction Act, it could halt billion-dollar projects that are being planned or perhaps even some that are already under construction. The growing policy uncertainty and weakening economic conditions alone may already be causing some of this to occur. Since Trump took office, companies have canceled, delayed, or scaled back at least nine US clean energy supply chain developments or operations, according to the Big Green Machine, a database maintained by Jay Turner, a professor of environmental studies at Wellesley College, and student researchers there. The projects that have been affected represent some $8 billion in public and private investments, and more than 9,000 jobs. They include KORE Powers planned battery facility in Arizona, which the company halted; Envision Automotive Energy Supplys paused expansion in Florence County, South Carolina; and Akasols closure of two plants in Michigan. VW also scaled back production at its recently expanded EV factory in Chattanooga, Tennessee, amid slower-than-expected growth in sales and, perhaps, the expectation that the Trump administration will strive to roll back consumer tax credits for vehicle purchases. The biggest challenge for companies that are making hundred-million- or billion-dollar capital investments is dealing with the uncertainty, Turner says. Uncertainty is a real deterrent to making big bets. Venture capital investments in clean energy have been cooling for a while. They peaked at $24.5 billion in 2022 and settled at around $18 billion annually during the last two years, according to data provided by Pitchbook. First-quarter figures for this year arent yet available, though industry watchers are keen to see where they land. Some parts of the cleantech sector could hold up better than others through the Trump administration and any upcoming economic gloom. The Pitchbook report, for instance, noted that the surge in development of AI data centers is fueling demand for dispatchable energy sources. That means the type that can run around the clock, such as nuclear fission, fusion, and geothermal (though in practice, the data center boom has often meant commissioning or relying on natural-gas plants that produce planet-warming emissions). Trumps new energy secretary, Chris Wright, previously the chief executive of the oilfield services company Liberty Energy, has also talked favorably about nuclear power and geothermaland rather unfavorably about renewables like solar and wind. But observers fear that more sectors will lose than win in any economic downturn to come, and Turner stresses that the decisions made during this administration could last well beyond it. The near-term concern is that this emerging clean-energy industry in the US suffers a significant pullback and the US cedes this market to other countries, especially China, that are actively working to position themselves to be leaders in the clean-energy future, he says. The long-term concern, he adds, is that if government policies on cleantech simply advance and retreat with the whims of each administration, companies will stop trying to make long-term investments that bank on such subsidies, grants or loans. Catherine Wolfram, a professor of energy and applied economics at MIT, also notes that China and the European Union are forging ahead in developing policies to drive down emissions and build up carbon-free sectors. She observes that theyre both now moving on to the tougher work of cleaning up heavy industries like steel, while the US is losing ground on even making clean electricity. Its the worst kind of US exceptionalism, she says.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. Rivals are rising to challenge the dominance of SpaceX SpaceX is a space launch juggernaut. In just two decades, the company has managed to edge out former aerospace heavyweights Boeing, Lockheed, and Northrop Grumman to gain near-monopoly status over rocket launches in the US. It is now also the go-to launch provider for commercial customers, having lofted numerous satellites and five private crewed spaceflights, with more to come. Other space companies have been scrambling to compete for years, but developing a reliable rocket takes slow, steady work and big budgets. Now at least some of them are catching up. Read the full story.Ramin Skibba We should talk more about air-conditioning Casey Crownhart Things are starting to warm up here in the New York City area, and its got me thinking once again about something that people arent talking about enough: energy demand for air conditioners. I get it: Data centers are the shiny new thing to worry about. And Im not saying we shouldnt be thinking about the strain that gigawatt-scale computing installations put on the grid. But a little bit of perspective is important here. I just finished up a new story about a novel way to make heat exchangers, a crucial component in air conditioners and a whole host of other technologies that cool our buildings, food, and electronics. Lets dig into why Im writing about the guts of cooling technologies, and why this sector really needs innovation. Read the full story. This article is from The Spark, MIT Technology Reviews weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 Donald Trump has announced sweeping new tariffsExperts fear the measures will spark a global trade war. (FT $) + The new tariffs are significantly higher than Americas targeted trade partners. (Vox)+ US tech companies are reliant on global supply chains. What happens next? (Wired $)+ Tech stocks dropped sharply following the announcement. (CNBC)2 Elon Musk tried to control the Wisconsin Supreme Court raceand lostThe billionaire was mocked on his own platform, X, after the state rejected the Republican candidate he spent millions bankrolling. (The Guardian)+ It was the most expensive judicial election in American history. (Economist $)+ It appears as though Musks political influence is waning. (The Atlantic $)3 Amazon made a bid to keep TikTok operational in the USAs has mobile tech company AppLovin. (WSJ $) + The founder of OnlyFans partnered with a crypto foundation in another bid. (Reuters)4 Parents are worried about their teenagers smartphone use But drawing firm conclusions about phones and social medias effects on their mental health is far from easy. (Nature)5 How China gets around Americas chip restrictions Smuggling and subsidiaries are just some of the ways it skirts the bans. (Rest of World)+ This super-thin semiconductor is just one molecule thick. (Ars Technica)+ Whats next in chips. (MIT Technology Review) 6 Neuralink is looking for new patients across the worldThe company has implanted devices in three peoples brains to date. (Bloomberg $) + Brain-computer interfaces face a critical test. (MIT Technology Review)7 Italian police are investigating a major fire at a Tesla dealershipThe blaze destroyed 17 cars in Rome. (The Guardian) 8 Publishers are experimenting with AI translations for booksNot everyone agrees that the technology is ready. (The Markup) 9 Vibe coding needs a reality check A new AI app created using the loose process generated a recipe for deadly cyanide ice cream. (404 Media)10 You may be unwittingly following JD Vances wife on Instagram If you were following Kamala Harriss husband on the platform, you're now following Usha Vance. (TechCrunch)Quote of the day Elon Musks money might buy some ads, but it repels voters. Wisconsin Democratic Party Chairman Ben Wikler reflects on how his partys candidate Susan Crawford won the states Supreme Court election, despite Musk spending $25 million supporting her Trump-endorsed rival, The Hill reports. The big story The lucky break behind the first CRISPR treatment December 2023 The worlds first commercial gene-editing treatment is set to start changing the lives of people with sickle-cell disease. Its called Casgevy, and it was approved in November 2022 in the UK.The treatment, which will be sold in the US by Vertex Pharmaceuticals, employs CRISPR, which can be easily programmed by scientists to cut DNA at precise locations they choose.But where do you aim CRISPR, and how did the researchers know what DNA to change? Thats the lesser-known story of the sickle-cell breakthrough. Read more about it.Antonio Regalado We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + If youre stuck for what to read next, this list of the 21st centurys best books is a great source of inspiration.+ Controversial ranking timedo you agree that Abbey Road is the Beatles best album?+ Inside the tricky technicalities of time travel.+ Uhoh: magnolia paint is making a comeback.

SpaceX is a space launch juggernaut. In just two decades, the company has managed to edge out former aerospace heavyweights Boeing, Lockheed, and Northrop Grumman to gain near-monopoly status over rocket launches in the US; it accounted for 87% of the countrys orbital launches in 2024, according to an analysis by SpaceNews. Since the mid-2010s, the company has dominated NASAs launch contracts and become a major Pentagon contractor. It is now also the go-to launch provider for commercial customers, having lofted numerous satellites and five private crewed spaceflights, with more to come. Other space companies have been scrambling to compete for years, but developing a reliable rocket takes slow, steady work and big budgets. Now at least some of them are catching up. A host of companies have readied rockets that are comparable to SpaceXs main launch vehicles. Some of these competitors are just starting to get rockets off the ground. And the companies could also face unusual headwinds, given that SpaceXs Elon Musk has an especially close relationship with the Trump administration and has allies at federal regulatory agencies, including those that provide oversight of the industry. But if all goes well, the SpaceX challengers can help improve access to space and prevent bottlenecks if one company experiences a setback. More players in the market is good for competition, says Chris Combs, an aerospace engineer at the University of Texas at San Antonio. I think for the foreseeable future it will still be hard to compete with SpaceX on price. But, he says, the competitors could push SpaceX itself to become better and provide those seeking access to space with a wider array of options.. A big lift There are a few reasons why SpaceX was able to cement its position in the space industry. When it began in the 2000s, it had three consecutive rocket failures and seemed poised to fold. But it barreled through with Musks financial support, and later They got government contracts from the very beginning, says Victoria Samson, a space policy expert at the Secure World Foundation in Broomfield, Colorado. I wouldnt say its a handout, but SpaceX would not exist without a huge influx of repeated government contracts. To this day, theyre still dependent on government customers, though they have commercial customers too. SpaceX has also effectively achieved a high degree of vertical integration, Samson points out: It owns almost all parts of its supply chain, designing, building, and testing all its major hardware components in-house, with a minimal use of suppliers. That gives it not just control over its hardware but considerably lower costs, and the price tag is the top consideration for launch contracts. The company was also open to taking risks other industry stalwarts were not. I think for a very long time the industry looked at spaceflight as something that had to be very precise and perfect, and not a lot of room for tinkering, says Combs. SpaceX really was willing to take some risks and accept failure in ways that others havent been. Thats easier to do when youre backed by a billionaire. Whats finally enabled international and US-based competitors to emerge has been a growing customer base looking for launch services, along with some investors deep pockets. Some of these companies are taking aim at SpaceXs Falcon 9, which can lift as much as about 20,000 kilograms into orbit and is used for sending multiple satellites or the crewed Dragon into space. There is a practical monopoly in the medium-lift launch market right now, with really only one operational vehicle, says Murielle Baker, a spokesperson for Rocket Lab, a US-New Zealand company. Rocket Lab plans to take on the Falcon 9 with its Neutron rocket, which is expected to have its inaugural flight later this year from NASAs Wallops Flight Facility in Virginia. The effort is building on the success of the companys smaller Electron rocket, and Neutrons first stage is intended to be reusable after it parachutes down to the ocean. Another challenger is Texas-based Firefly, whose Alpha rocket can be launched from multiple spaceports so that it can reach different orbits. Firefly has already secured NASA and Space Force contracts, with more launches coming this year (and on March 2 it also became the second private company to successfully land a spacecraft on the moon). Next year, Relativity Space aims to loft its first Terran R rocket, which is partially built from 3D-printed components. And the Bill Gatesbacked Stoke Space aims to launch its reusable Nova rocket in late 2025 or, more likely, next year. Competitors are also rising for SpaceXs Falcon Heavy, holding out the prospect of more options for sending massive payloads to higher orbits and deep space. Furthest along is the Vulcan Centaur rocket, a creation of United Launch Alliance, a joint venture between Boeing and Lockheed Martin. Its expected to have its third and fourth launches in the coming months, delivering Space Force satellites to orbit. Powered by engines from Blue Origin, the Vulcan Centaur is slightly wider and shorter than the Falcon rockets. It currently isnt reusable, but its less expensive than its predecessors, ULAs Atlas V and Delta IV, which are being phased out. Mark Peller, the companys senior vice president on Vulcan development and advanced programs, says the new rocket comes with multiple advantages. One is overall value, in terms of dollars per pound to orbit and what we can provide to our customers, he says, and the second is versatility: Vulcan was designed to go to a range of orbits. He says more than 80 missions are already lined up. Vulcans fifth flight, slated for no earlier than May, will launch the long-awaited Sierra Space Dream Chaser, a spaceplane that can carry cargo (and possibly crew) to the International Space Station. ULA also has upcoming Vulcan launches planned for Amazons Kuiper satellite constellation, a potential Starlink rival. Meanwhile, though it took a few years, Blue Origin now has a truly orbital heavy-lift spacecraft: In January, it celebrated the inaugural launch of its towering New Glenn, a rocket thats only a bit shorter than NASAs Space Launch System and SpaceXs Starship. Future flights could launch national security payloads. Competition is emerging abroad as well. After repeated delays, Europes heavy-lift Ariane 6, from Airbus subsidiary Arianespace, had its inaugural flight last year, ending the European Space Agencys temporary dependence on SpaceX. A range of other companies are trying to expand European launch capacity, with assistance from ESA. China is moving quickly on its own launch organizations too. They had no less than seven commercial space launch companies that were all racing to develop an effective system that could deliver a payload into orbit, Kari Bingen, director of the Aerospace Security Project at the Center for Strategic and International Studies, says of Chinas efforts. They are moving fast and they have capital behind them, and they will absolutely be a competitor on the global market once theyre successful and probably undercut what US and European launch companies are doing. The up-and-coming Chinese launchers include Space Pioneers reusable Tianlong-3 rocket and Cosmoleaps Yueqian rocket. The latter is to feature a chopstick clamp recovery of the first stage, where its grabbed by the launch towers mechanical arms, similar to the concept SpaceX is testing for its Starship. Glitches and government Before SpaceXs rivals can really compete, they need to work out the kinks, demonstrate the reliability of their new spacecraft, and show that they can deliver low-cost launch services to customers. The process is not without its challenges. Boeings Starliner delivered astronauts to the ISS on its first crewed flight in June 2024, but after thruster malfunctions, they were left stranded at the orbital outpost for nine months. While New Glenn reached orbit as planned, its first stage didnt land successfully and its upper stage was left in orbit. SpaceX itself has had some recent struggles. The Federal Aviation Administration grounded the Falcon 9 more than once following malfunctions in the second half of 2024. The company still shattered records last year, though, with more than 130 Falcon 9 launches. It has continued with that record pace this year, despite additional Falcon 9 delays and more glitches with its booster and upper stage. SpaceX also conducted its eighth Starship test flight in March, just two months after the previous one, but both failed minutes after liftoff, raining debris down from the sky. Any company must deal with financial challenges as well as engineering ones. Boeing is reportedly considering selling parts of its space business, following Starliners malfunctions and problems with its 737 Max aircraft. And Virgin Orbit, the launch company that spun off from Virgin Galactic, shuttered in 2023. Another issue facing would-be commercial competitors to SpaceX in the US is the complex and uncertain political environment. Musk does not manage day-to-day operations of the company. But he has close involvement with DOGE, a Trump administration initiative that has been exerting influence on the workforces and budgets of NASA, the Defense Department, and regulators relevant to the space industry. Jared Isaacman, a billionaire who bankrolled the groundbreaking 2021 commercial mission Inspiration4, returned to orbit, again via a SpaceX craft, on Polaris Dawn last September. Now he may become Trumps NASA chief, a position that could give him the power to nudge NASA toward awarding new lucrative contracts to SpaceX. In February it was reported that SpaceXs Starlink might land a multibillion-dollar FAA contract previously awarded to Verizon. It is also possible that SpaceX could strengthen its position with respect to the regulatory scrutiny it has faced for environmental and safety issues at its production and launch sites on the coasts of Texas and Florida, as well as scrutiny of its rocket crashes and the resulting space debris. Oversight from the FAA, the Federal Communications Commission, and the Environmental Protection Agency may be weak. Conflicts of interest have already emerged at the FAA, and the Trump administration has also attempted to incapacitate the National Labor Relations Board. SpaceX had previously tried to block the board from acting after nine workers accused the company of unfair labor practices. SpaceX did not respond to MIT Technology Reviews requests for comment for this story. I think theres going to be a lot of emphasis to relieve a lot of the regulations, in terms of environmental impact studies, and things like that, Samson says. I thought thered be a separation between [Musks] interests, but now, its hard to say where he stops and the US government begins. Regardless of the politics, the commercial competition will surely heat up throughout 2025. But SpaceX has a considerable head start, Bingen argues: Its going to take a lot for these companies to effectively compete and potentially dislodge SpaceX, given the dominant position that [it has] had. Ramin Skibba is an astrophysicist turned science writer and freelance journalist, based in the Bay Are

Things are starting to warm up here in the New York City area, and its got me thinking once again about something that people arent talking about enough: energy demand for air conditioners. I get it: Data centers are the shiny new thing to worry about. And Im not saying we shouldnt be thinking about the strain that gigawatt-scale computing installations put on the grid. But a little bit of perspective is important here. According to a report from the International Energy Agency last year, data centers will make up less than 10% of the increase in energy demand between now and 2030, far less than the energy demand from space cooling (mostly air-conditioning). I just finished up a new story thats out today about a novel way to make heat exchangers, a crucial component in air conditioners and a whole host of other technologies that cool our buildings, food, and electronics. Lets dig into why Im writing about the guts of cooling technologies, and why this sector really needs innovation. One twisted thing about cooling and climate change: Its all a vicious cycle. As temperatures rise, the need for cooling technologies increases. In turn, more fossil-fuel power plants are firing up to meet that demand, turning up the temperature of the planet in the process. Cooling degree days are one measure of the need for additional cooling. Basically, you take a preset baseline temperature and figure out how much the temperature exceeds it. Say the baseline (above which youd likely need to flip on a cooling device) is 21 C (70 F). If the average temperature for a day is 26 C, thats five cooling degree days on a single day. Repeat that every day for a month, and you wind up with 150 cooling degree days. I explain this arguably weird metric because its a good measure of total energy demand for coolingit lumps together both how many hot days there are and just how hot it is. And the number of cooling degree days is steadily ticking up globally. Global cooling degree days were 6% higher in 2024 than in 2023, and 20% higher than the long-term average for the first two decades of the century. Regions that have high cooling demand, like China, India, and the US, were particularly affected, according to the IEA report. You can see a month-by-month breakdown of this data from the IEA here. That increase in cooling degree days is leading to more demand for air conditioners, and for energy to power them. Air-conditioning accounted for 7% of the worlds electricity demand in 2022, and its only going to get more important from here. There were fewer than 2 billion AC units in the world in 2016. By 2050, that could be nearly 6 billion, according to a 2018 report from the IEA. This is a measure of progress and, in a way, something we should be happy about; the number of air conditioners tends to rise with household income. But it does present a challenge to the grid. Another piece of this whole thing: Its not just about how much total electricity we need to run air conditioners but about when that demand tends to come. As weve covered in this newsletter before, your air-conditioning habits arent unique. Cooling devices tend to flip on around the same timewhen its hot. In some parts of the US, for example, air conditioners can represent more than 70% of residential energy demand at times when the grid is most stressed. The good news is that were seeing innovations in cooling technology. Some companies are building cooling systems that include an energy storage component, so they can charge up when energy is plentiful and demand is low. Then they can start cooling when its most needed, without sucking as much energy from the grid during peak hours. Weve also covered alternatives to air conditioners called desiccant cooling systems, which use special moisture-sucking materials to help cool spaces and deal with humidity more efficiently than standard options. And in my latest story, I dug into new developments in heat exchanger technology. Heat exchangers are a crucial component of air conditioners, but you can really find them everywherein heat pumps, refrigerators, and, yes, the cooling systems in large buildings and large electronics installations, including data centers. Weve been building heat exchangers basically the same way for nearly a century. These components basically move heat around, and there are a few known ways to do so with devices that are relatively straightforward to manufacture. Now, though, one team of researchers has 3D-printed a heat exchanger that outperforms some standard designs and rivals others. This is still a long way from solving our looming air-conditioning crisis, but the details are fascinatingI hope youll give it a read. We need more innovation in cooling technology to help meet global demand efficiently so we dont stay stuck in this cycle. And well need policy and public support to make sure that these technologies make a difference and that everyone has access to them too. This article is from The Spark, MIT Technology Reviews weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here.

Rumors of the ongoing death of software development that its being slain by AI are greatly exaggerated. In reality, software development is at a fork in the road: embracing the (currently) far-off notion of fully automated software development or acknowledging the work of a software developer is much more than just writing lines of code. The decision the industry makes could have significant long-term consequences. Increasing complacency around AI-generated code and a shift to what has been termed vibe coding where code is generated through natural language prompts until the results seem to work will lead to code thats more error-strewn, more expensive to run and harder to change in the future. And, if the devaluation of software development skills continues, we may even lack a workforce with the skills and knowledge to fix things down the line. This means software developers are going to become more important to how the world builds and maintains software. Yes, there are many ways their practices will evolve thanks to AI coding assistance, but in a world of proliferating machine-generated code, developer judgment and experience will be vital. The dangers of AI-generated code are already here The risks of AI-generated code arent science fiction: theyre with us today. Research done by GitClear earlier this year indicates that with AI coding assistants (like GitHub Copilot) going mainstream, code churn which GitClear defines as changes that were either incomplete or erroneous when the author initially wrote, committed, and pushed them to the companys git repo" has significantly increased. GitClear also found there was a marked decrease in the number of lines of code that have been moved, a signal for refactored code (essentially the care and feeding to make it more effective). In other words, from the time coding assistants were introduced theres been a pronounced increase in lines of code without a commensurate increase in lines deleted, updated, or replaced. Simultaneously, there's been a decrease in lines moved indicating a lot of code has been written but not refactored. More code isnt necessarily a good thing (sometimes quite the opposite); GitClears findings ultimately point to complacency and a lack of rigor about code quality. Can AI be removed from software development? However, AI doesnt have to be removed from software development and delivery. On the contrary, theres plenty to be excited about. As noted in the latest volume of the Technology Radar Thoughtworks report on technologies and practices from work with hundreds of clients all over the world the coding assistance space is full of opportunities. Specifically, the report noted tools like Cursor, Cline and Windsurf can enable software engineering agents. What this looks like in practice is an agent-like feature inside developer environments that developers can ask specific sets of coding tasks to be performed in the form of a natural language prompt. This enables the human/machine partnership. That being said, to only focus on code generation is to miss the variety of ways AI can help software developers. For example, Thoughtworks has been interested in how generative AI can be used to understand legacy codebases, and we see a lot of promise in tools like Unblocked, which is an AI team assistant thatsupport for new languages in an internal tool, CodeConcise. We use CodeConcise to understand legacy systems; and while our success was mixed, we do think theres real promise here. Tightening practices to better leverage AI Its important to remember much of the work developers do isnt developing something new from scratch. A large proportion of their work is evolving and adapting existing (and sometimes legacy) software. Sprawling and janky code bases that have taken on technical debt are, unfortunately, the norm. Simply applying AI will likely make things worse, not better, especially with approaches like vibe. This is why developer judgment will become more critical than ever. In the latest edition of the Technology Radar report, AI-friendly code design is highlighted, based on our experience that AI coding assistants perform best with well-structured codebases. In practice, this requires many different things, including clear and expressive naming to ensure context is clearly communicated (essential for code maintenance), reducing duplicate code, and ensuring modularity and effective abstractions. Done together, these will all help make code more legible to AI systems. Good coding practices are all too easy to overlook when productivity and effectiveness are measured purely in terms of output, and even though this was true before there was AI tooling, software development needs to focus on good coding first. AI assistance demands greater human responsibility Instagram co-founder Mike Krieger recently claimed that in three years software engineers wont write any code: they will only review AI-created code. This might sound like a huge claim, but its important to remember that reviewing code has always been a major part of software development work. With this in mind, perhaps the evolution of software development wont be as dramatic as some fear. But theres another argument: as AI becomes embedded in how we build software, software developers will take on more responsibility, not less. This is something weve discussed a lot at Thoughtworks: the job of verifying that an AI-built system is correct will fall to humans. Yes, verification itself might be AI-assisted, but it will be the role of the software developer to ensure confidence. In a world where trust is becoming highly valuable as evidenced by the emergence of the chief trust officer the work of software developers is even more critical to the infrastructure of global industry. Its vital software development is valued: the impact of thoughtless automation and pure vibes could prove incredibly problematic (and costly) in the years to come. This content was produced by Thoughtworks. It was not written by MIT Technology Reviews editorial staff.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. How 3D printing could make better cooling systems A new 3D-printed design could make an integral part of cooling systems like air conditioners or refrigerators smaller and more efficient, according to new research. Heat exchangers are devices that whisk away heat, and theyre everywhereused in data centers, ships, factories, and buildings. The aim is to pass as much heat as possible from one side of the device to the other. Most use one of a few standard designs that have historically been easiest and cheapest to make. Energy demand for cooling buildings alone is set to double between now and 2050, and new designs could help efficiently meet the massive demand forecast for the coming decades. Read the full story. Casey Crownhart MIT Technology Review Narrated: The quest to figure out farming on Mars If were going to live on Mars well need a way to grow food in its arid dirt. Researchers think they know a way. This is our latest story to be turned into a MIT Technology Review Narrated podcast, which were publishing each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as its released. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 Thousands of US health agency workers have been laid off Experts warn that patients will die preventable deaths as a result. (Wired $)+ How will the US respond to the measles and bird flu outbreaks? (Reuters)+ US cuts could lead to serious delays in forecasting extreme weather. (Undark)+ The wide-ranging cuts are also likely to lose America money. (The Atlantic $)2 Donald Trump is set to discuss a proposal to save TikTok Hes due to meet with aides today to thrash out a new ownership structure. (NYT $)+ Oracle and Blackstone are among the companies in talks to make an offer. (WSJ $)+ The White House is playing the role of investment bank. (The Guardian)3 X has asked the Supreme Court to exempt its users from law enforcementIt claims to be worried by broad, suspicionless requests. (FT $) 4 Things arent looking good for Mexico-based Chinese companies Trumps tariff plans could imperil an awful lot of deals. (WSJ $)+ The US Chips Act is another probable casualty. (Bloomberg $)5 US lawmakers want to regulate AI companionsA proposed bill would allow users to sue if they suffer harm from their interactions with a companion bot. (WP $) + We need to prepare for addictive intelligence. (MIT Technology Review) 6 Covid hasnt gone awayAnd life for the covid-conscious is getting increasingly difficult. (The Atlantic $) 7 Brands are trying to game Reddit to show up in ChatGPT recommendationsCatering to AI search is a whole business model now. (The Information $) + Your most important customer may be AI. (MIT Technology Review)8 Nothing could destroy the universe Humans have long been obsessed with nothingness. (New Scientist $)9 Would you flirt with a chatbot?Tinder wants you to give it a go. (Bloomberg $) + The AI relationship revolution is already here. (MIT Technology Review)10 Trading in your Tesla is TikToks favorite trend Clips of Tesla owners ditching their cars are going viral. (Fast Company $)+ This guy returned his Cybertruck out of fear his daughter would get bullied. (Insider $)+ Sales of new Teslas are slumping too. (NYT $)Quote of the day Id get on in a heartbeat. Butch Wilmore, one of the pair of astronauts who was stuck in space for nine months, explains how hed be willing to fly on the beleaguered Starliner again, the Washington Post reports. The big story Bringing the lofty ideas of pure math down to earth April 2023Pradeep Niroula Mathematics has long been presented as a sanctuary from confusion and doubt, a place to go in search of answers. Perhaps part of the mystique comes from the fact that biographies of mathematicians often paint them as otherworldly savants. As a graduate student in physics, I have seen the work that goes into conducting delicate experiments, but the daily grind of mathematical discovery is a ritual altogether foreign to me. And this feeling is only reinforced by popular books on math, which often take the tone of a pastor dispensing sermons to the faithful.Luckily, there are ways to bring it back down to earth. Popular math books seek a fresher take on these old ideas, be it through baking recipes or hot-button political issues. My verdict: Why not? Its worth a shot. Read the full story. We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + Why are cats the way they are? This database might help us find out.+ John McFall could become the first disabled person in space.+ ASMR at the V&A is just delightful.+ Addicted to lip balm? Youre not the only one.

A new 3D-printed design could make an integral part of cooling systems like air conditioners or refrigerators smaller and more efficient, according to new research. Heat exchangers are devices that whisk away heat, and theyre everywhereused in data centers, ships, factories, and buildings. The aim is to pass as much heat as possible from one side of the device to the other. Most use one of a few standard designs that have historically been easiest and cheapest to make. Heat exchangers are at the center of the industrial economy. Theyre an essential part of every machine and every system that moves energy, says William King, a professor at the University of Illinois Urbana-Champaign and one of the authors of the new study. Existing designs tend to favor straight lines, right angles, and round tubes, he adds. King and his colleagues used 3D printing to design a heat exchanger that includes features to optimize heat movement, like wavy walls and pyramid-shaped bumps, which wouldnt be possible to make using traditional manufacturing techniques. The team had set out to design a system based on a common refrigerant called R-134a, which is commonly used in devices like air conditioners and refrigerators. When cold water lowers the temperature of the refrigerant, it changes from a gas to a liquid on its path through the device. That liquid refrigerant can then go on to other parts of the cooling system, where its used to lower the temperature of anything from a room to a rack of servers. The best way to cool the refrigerant tends to involve building very thin walls between the two sides of the device and maximizing the amount of contact that the water and the refrigerant make with those walls. (Think about how much colder youd get wearing a thin T-shirt and pants and lying down on ice than simply touching it with your gloved hands.) To design the best possible heat exchanger, researchers used simulations and developed machine-learning models to help predict the performance of different designs under different conditions. After 36,000 simulations, the researchers landed on the one they decided to develop. Among the key components: small fins that jut out on the side of the device that touches the water, increasing the surface area to maximize heat transfer. The team also designed wavy passageways for the water to pass throughonce again helping to maximize surface area. Simulations helped the researchers figure out exactly how curvy the passages should be and where precisely to place the fins. On the side of the devices where the refrigerant passes through, the design includes small pyramid-shaped bumps along the walls. These not only maximize the area for cooling but also help mix the refrigerant as it passes through and prevent liquid from coating the wall (which would slow down the heat transfer). After settling on a design, the researchers used a 3D-printing technique called direct metal laser sintering, in which lasers melt and fuse together a metal powder (in this case, an aluminum alloy), layer by layer. In testing, the researchers found that the heat exchanger created with this technique was able to cool down the refrigerant more efficiently than other designs. The new device was able to achieve a power density of over six megawatts per meter cubedoutperforming one common traditional design, the shell-tube configuration, by between 30% and 50% with the same pumping power. The devices power density was similar to that of brazed plate heat exchangers, another common design in industry. Overall, this device doesnt dramatically outperform the state-of-the-art technology, but the technique of using modeling and 3D printing to produce new heat exchanger designs is promising, says Dennis Nasuta, director of research and development at Optimized Thermal Systems, a consulting firm that works with companies in the HVAC industry on design and research. Its worth exploring, and I dont think that we know yet where we can push it, Nasuta says. One challenge is that today, additive manufacturing techniques such as laser sintering are slow and expensive compared with traditional manufacturing; they wouldnt be economical or feasible to rely on for all our consumer cooling devices, he says. For now, this type of approach could be most useful in niche applications like aerospace and high-end automotives, which could more likely bear the cost, he adds. This particular study was funded by the US Office of Naval Research. Next-generation ships have more electronics aboard than ever, and theres a growing need for compact and efficient systems to deal with all that extra heat, says Nenad Miljkovic, one of the authors of the study. Energy demand for cooling buildings alone is set to double between now and 2050, and new designs could help efficiently meet the massive demand forecast for the coming decades. But challenges including manufacturing costs would need to be overcome to help innovations like the one designed by King and his team make a dent in real devices. Another barrier to adopting these new techniques, Nasuta says, is that current standards dont demand more efficiency. Other technologies already exist that could help make our devices more efficient, but theyre not used for the same reason. It will take time for new manufacturing techniques, including 3D printing, to trickle into our devices, Natsua adds: This isnt going to be in your AC next year.

Tech companies are always trying out new ways for people to interact with computersconsider efforts like Google Glass, the Apple Watch, and Amazons Alexa. Youve probably used at least one. But the most radical option has been tried by fewer than 100 people on Earththose who have lived for months or years with implanted brain-computer interfaces, or BCIs. Implanted BCIs are electrodes put in paralyzed peoples brains so they can use imagined movements to send commands from their neurons through a wire, or via radio, to a computer. In this way, they can control a computer cursor or, in few cases, produce speech. Recently, this field has taken some strides toward real practical applications. About 25 clinical trials of BCI implants are currently underway. And this year MIT Technology Review readers have selected these brain-computer interfaces as their addition to our annual list of 10 Breakthrough Technologies, published in January. BCIs won by a landslide to become the 11th Breakthrough, as we call it. It beat out three runners-up: continuous glucose monitors, hyperrealistic deepfakes, and methane-detecting satellites. The impression of progress comes thanks to a small group of companies that are actively recruiting volunteers to try BCIs in clinical trials. They are: Neuralink, backed by the worlds richest person, Elon Musk; New Yorkbased Synchron; and Chinas Neuracle Neuroscience. Each is trialing interfaces with the eventual goal of getting the fields first implanted BCI approved for sale. I call it the translation era, says Michelle Patrick-Krueger, a research scientist who carried out a detailed survey of BCI trials with neuroengineer Jose Luis Contreras-Vidal at the University of Houston. In the past couple of years there has been considerable private investment. That creates excitement and allows companies to accelerate. Thats a big change, since for years BCIs have been more like a neuroscience parlor trick, generating lots of headlines but little actual help to patients. Patrick-Krueger says the first time a person controlled a computer cursor from a brain implant was in 1998. That was followed by a slow drip-drip of tests in which university researchers would find a single volunteer, install an implant, and carry out studies for months or years. Over 26 years, Patrick-Krueger says, she was able to document a grand total of 71 patients whove ever controlled a computer directly with their neurons. That means you are more likely to be friends with a Mega Millions jackpot winner than know someone with a BCI. These studies did prove that people could use their neurons to play Pong, move a robot arm, and even speak through a computer. But such demonstrations are of no practical help to people with paralysis severe enough to benefit from a brain-controlled computer, because these implants are not yet widely available. One thing is to have them work, and another is how to actually deploy them, says Contreras-Vidal. Also, behind any great news are probably technical issues that need to be addressed. These include how long an implant will last and how much control it offers patients. Larger trials from three companies are now trying to resolve these questions and set the groundwork for a real product. One company, Synchron, uses a stent with electrodes on it thats inserted into a brain vessel via a vein in the neck. Synchron has implanted its stentrode in 10 volunteers, six in the US and four in Australiathe most simultaneous volunteers reported by any BCI group. The stentrode collects limited brain signals, so it gives users only a basic on/off type of control signal, or what Synchron calls a switch. That isnt going to let a paralyzed person use Photoshop. But its enough to toggle through software menus or select among prewritten messages. Tom Oxley, Synchrons CEO, says the advantage of the stentrode is that it is as simple as possible. That, he believes, will make his brain-computer interface scalable to more people, especially since installing it doesnt involve brain surgery. Synchron might be ahead, but its still in an exploratory phase. A pivotal study, the kind used to persuade regulators to allow sales of a specific version of the device, has yet to be scheduled. So theres no timeline for a product. Neuralink, meanwhile, has disclosed that three volunteers have received its implant, the N1, which consists of multiple fine electrode threads inserted directly into the brain through a hole drilled in the skull. More electrodes mean more neural activity is captured. Neuralinks first volunteer, Noland Arbaugh, has shown off how he can guide a cursor around a screen in two dimensions and click, letting him play video games like Civilization or online chess. Finally, Neuracle says it is running two trials in China and one in the US. Its implant consists of a patch of electrodes placed on top of the brain. In a report, the company said a paralyzed volunteer is using the system to stimulate electrodes in his arm, causing his hand to close in a grasp. But details remain sparse. A Neuracle executive would only say that several people had received its implant. Because Neuracles patient count isnt public, it wasnt included in Patrick-Kruegers tally. In fact, theres no information at all in the medical literature on about a quarter of brain-implant volunteers so far, so she counted them using press releases or by e-mailing research teams. Her BCI survey yielded other insights. According to her data, implants have lasted as long as 15 years, more than half of patients are in the US, and roughly 75% of BCI recipients have been male. The data cant answer the big question, though. And that is whether implanted BCIs will progress from breakthrough demonstrations into breakout products, the kind that help many people. In the next five to 10 years, its either going to translate into a product or itll still stay in research, Patrick-Krueger says. I do feel very confident there will be a breakout.

The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 Tech companies are warning their immigrant workers not to leave the US Employees on high-skilled visas could be denied entry back into the States. (WP $)+ Officials are considering collecting citizenship applicants social media data. (Associated Press)2 OpenAI has closed one of the largest private funding rounds in historyIt plans to put the $40 billion cash injection towards building AGI. (The Guardian)+ The deal values OpenAI at a whopping $300 billion. (CNBC)+ The company also teased its first open-weight model in years. (Insider $)3 SpaceX has launched a mission thats never been attempted before Its taking private customers on an orbit between Earths North and South poles. (CNN)+ Crypto billionaire Chun Wang is footing the bill for the mission. (Reuters)+ Europe is finally getting serious about commercial rockets. (MIT Technology Review)4 Some DOGE workers are returning to their old jobsTheyre quietly heading back to their roles at X and SpaceX. (The Information $)+ Top staff were placed on leave after denying DOGE access to their systems. (Wired $) + Can AI help DOGE slash government budgets? Its complex. (MIT Technology Review)5 Amazon is going all-in on AI agents Its new AI model Nova Act is designed to complete tasks such as online shopping. (The Verge)+ Why handing over total control to AI agents would be a huge mistake. (MIT Technology Review)6 DeepMind is making it harder for its researchers to publish studies Its reluctant to share innovations that rivals could capitalize on. (FT $)7 Meet the protestors staking out Tesla dealershipsProfessors and attorneys have taken to the streets to fight back. (New Yorker $) + Far-right extremists are turning up to defend the company. (Wired $)

On March 27, the results of the first clinical trial for a generative AI therapy bot were published, and they showed that people in the trial who had depression or anxiety or were at risk for eating disorders benefited from chatting with the bot. I was surprised by those results, which you can read about in my full story. There are lots of reasons to be skeptical that an AI model trained to provide therapy is the solution for millions of people experiencing a mental health crisis. How could a bot mimic the expertise of a trained therapist? And what happens if something gets complicateda mention of self-harm, perhapsand the bot doesnt intervene correctly? The researchers, a team of psychiatrists and psychologists at Dartmouth Colleges Geisel School of Medicine, acknowledge these questions in their work. But they also say that the right selection of training datawhich determines how the model learns what good therapeutic responses look likeis the key to answering them. Finding the right data wasnt a simple task. The researchers first trained their AI model, called Therabot, on conversations about mental health from across the internet. This was a disaster. If you told this initial version of the model you were feeling depressed, it would start telling you it was depressed, too. Responses like, Sometimes I cant make it out of bed or I just want my life to be over were common, says Nick Jacobson, an associate professor of biomedical data science and psychiatry at Dartmouth and the studys senior author. These are really not what we would go to as a therapeutic response. The model had learned from conversations held on forums between people discussing their mental health crises, not from evidence-based responses. So the team turned to transcripts of therapy sessions. This is actually how a lot of psychotherapists are trained, Jacobson says. That approach was better, but it had limitations. We got a lot of hmm-hmms, go ons, and then Your problems stem from your relationship with your mother, Jacobson says. Really tropes of what psychotherapy would be, rather than actually what wed want. It wasnt until the researchers started building their own data sets using examples based on cognitive behavioral therapy techniques that they started to see better results. It took a long time. The team began working on Therabot in 2019, when OpenAI had released only its first two versions of its GPT model. Now, Jacobson says, over 100 people have spent more than 100,000 human hours to design this system. The importance of training data suggests that the flood of companies promising therapy via AI models, many of which are not trained on evidence-based approaches, are building tools that are at best ineffective, and at worst harmful. Looking ahead, there are two big things to watch: Will the dozens of AI therapy bots on the market start training on better data? And if they do, will their results be good enough to get a coveted approval from the US Food and Drug Administration? Ill be following closely. Read more in the full story. This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first,sign up here.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. The first trial of generative AI therapy shows it might help with depression The first clinical trial of a generative AI therapy bot suggests it was as effective as human therapy for people with depression, anxiety, or risk for developing eating disorders. Even so, it doesnt give a go-ahead to the dozens of companies hyping such technologies while operating in a regulatory gray area. Read the full story. James O'Donnell How a bankruptcy judge can stop a genetic privacy disaster Keith Porcaro The fate of 15 million peoples genetic data rests in the hands of a bankruptcy judge now that 23andMe has filed for bankruptcy. But theres still a small chance of writing a better ending for usersand its a simple fix. Read the full story. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 Meet the online activists fighting back against ICE raids Their networks are warning migrants about ICE officer hotspots in major cities. (WP $)+ Noncitizens are growing increasingly anxious. (NPR)2 US health experts were ordered to bury a measles forecastThe assessment warned the risk of catching the virus was high in areas with lower vaccination rates. (ProPublica) + The former US covid chief has called the outbreak wholly preventable. (Politico)+ How measuring vaccine hesitancy could help health professionals tackle it. (MIT Technology Review)3 Donald Trump is confident a TikTok deal is forthcoming Ahead of the impending deadline on Saturday. (Reuters)4 Chinas efforts to clean up air pollution are accelerating global warming Its dirty air had been inadvertently cooling the planet. (New Scientist $)+ Whos to blame for climate change? Its surprisingly complicated. (MIT Technology Review)5 Brands are spending small amounts on X to appease Elon Musk Theyre doing what they can to avoid triggering a public fallout with the billionaire. (FT $)+ Musks X has a new ownerits, err, Musks xAI. (CNBC)The Oxevision system remotely tracks patients breathing and heart rates. (6 Campaigners are calling to pause a mental health inpatient monitoring systemThe Guardian)+ This AI-powered black box could make surgery safer. (MIT Technology Review)7 The US and China are locked in a race to produce the first useful humanoid robotThe first to succeed will dominate the future of many labor-intensive industries. (WSJ $) + Beijing is treating humanoid robots as a major future industry. (WP $)8 Data center operators are inking solar power dealsIts a proven, clean technology that is relatively low-cost. (TechCrunch) + The cost of AI services is dropping. (The Information $)+ Why the US is still trying to make mirror-magnified solar energy work. (MIT Technology Review) 9 H&M plans to create digital replicas of its models Which means the retailer could outsource entire photoshoots to AI. (NYT $)+ The metaverse fashion stylists are here. (MIT Technology Review)10 What its like to drive a Tesla Cybertruck in Washington DC Expect a whole lot of abuse. (The Atlantic $)+ Protestors are gathering at Tesla showrooms across America. (Insider $)Quote of the day Viruses dont need a passport. Dr William Schaffner, an infectious disease expert at Vanderbilt University Medical Center, warns CNN that the US measles outbreak could spread widely to other countries. The big story Marseilles battle against the surveillance state June 2022 Across the world, video cameras have become an accepted feature of urban life. Many cities in China now have dense networks of them, and London and New Delhi arent far behind. Now France is playing catch-up.Concerns have been raised throughout the country. But the surveillance rollout has met special resistance in Marseille, Frances second-biggest city.Its unsurprising, perhaps, that activists are fighting back against the cameras, highlighting the surveillance systems overreach and underperformance. But are they succeeding? Read the full story.Fleur Macdonald We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + The online pocket computer museum is exceptionally charming.+ Theres an entirely new cat color emerging, and scientists have finally worked out why.+ Experiencing Bluesky skeets posted in real time is a seriously trippy business.+ Never underestimate the power of a good deed.

The first clinical trial of a therapy bot that uses generative AI suggests it was as effective as human therapy for participants with depression, anxiety, or risk for developing eating disorders. Even so, it doesnt give a go-ahead to the dozens of companies hyping such technologies while operating in a regulatory gray area. A team led by psychiatric researchers and psychologists at the Geisel School of Medicine at Dartmouth College built the tool, called Therabot, and the results were published on March 27 in the New England Journal of Medicine. Many tech companies have built AI tools for therapy, promising that people can talk with a bot more frequently and cheaply than they can with a trained therapistand that this approach is safe and effective. Many psychologists and psychiatrists have shared the vision, noting that fewer than half of people with a mental disorder receive therapy, and those who do might get only 45 minutes per week. Researchers have tried to build tech so that more people can access therapy, but they have been held back by two things. One, a therapy bot that says the wrong thing could result in real harm. Thats why many researchers have built bots using explicit programming: The software pulls from a finite bank of approved responses (as was the case with Eliza, a mock-psychotherapist computer program built in the 1960s). But this makes them less engaging to chat with, and people lose interest. The second issue is that the hallmarks of good therapeutic relationshipsshared goals and collaborationare hard to replicate in software. In 2019, as early large language models like OpenAIs GPT were taking shape, the researchers at Dartmouth thought generative AI might help overcome these hurdles. They set about building an AI model trained to give evidence-based responses. They first tried building it from general mental-health conversations pulled from internet forums. Then they turned to thousands of hours of transcripts of real sessions with psychotherapists. We got a lot of hmm-hmms, go ons, and then Your problems stem from your relationship with your mother, said Michael Heinz, a research psychiatrist at Dartmouth College and Dartmouth Health and first author of the study, in an interview. Really tropes of what psychotherapy would be, rather than actually what wed want. Dissatisfied, they set to work assembling their own custom data sets based on evidence-based practices, which is what ultimately went into the model. Many AI therapy bots on the market, in contrast, might be just slight variations of foundation models like Metas Llama, trained mostly on internet conversations. That poses a problem, especially for topics like disordered eating. If you were to say that you want to lose weight, Heinz says, they will readily support you in doing that, even if you will often have a low weight to start with. A human therapist wouldnt do that. To test the bot, the researchers ran an eight-week clinical trial with 210 participants who had symptoms of depression or generalized anxiety disorder or were at high risk for eating disorders. About half had access to Therabot, and a control group did not. Participants responded to prompts from the AI and initiated conversations, averaging about 10 messages per day. Participants with depression experienced a 51% reduction in symptoms, the best result in the study. Those with anxiety experienced a 31% reduction, and those at risk for eating disorders saw a 19% reduction in concerns about body image and weight. These measurements are based on self-reporting through surveys, a method thats not perfect but remains one of the best tools researchers have. These results, Heinz says, are about what one finds in randomized control trials of psychotherapy with 16 hours of human-provided treatment, but the Therabot trial accomplished it in about half the time. Ive been working in digital therapeutics for a long time, and Ive never seen levels of engagement that are prolonged and sustained at this level, he says. Jean-Christophe Blisle-Pipon, an assistant professor of health ethics at Simon Fraser University who has written about AI therapy bots but was not involved in the research, says the results are impressive but notes that just like any other clinical trial, this one doesnt necessarily represent how the treatment would act in the real world. We remain far from a greenlight for widespread clinical deployment, he wrote in an email. One issue is the supervision that wider deployment might require. During the beginning of the trial, Heinz says, he personally oversaw all the messages coming in from participants (who consented to the arrangement) to watch out for problematic responses from the bot. If therapy bots needed this oversight, they wouldnt be able to reach as many people. I asked Heinz if he thinks the results validate the burgeoning industry of AI therapy sites. Quite the opposite, he says, cautioning that most dont appear to train their models on evidence-based practices like cognitive behavioral therapy, and they likely dont employ a team of trained researchers to monitor interactions. I have a lot of concerns about the industry and how fast were moving without really kind of evaluating this, he adds. When AI sites advertise themselves as offering therapy in a legitimate, clinical context, Heinz says, it means they fall under the regulatory purview of the Food and Drug Administration. Thus far, the FDA has not gone after many of the sites. If it did, Heinz says, my suspicion is almost none of themprobably none of themthat are operating in this space would have the ability to actually get a claim clearancethat is, a ruling backing up their claims about the benefits provided. Blisle-Pipon points out that if these types of digital therapies are not approved and integrated into health-care and insurance systems, it will severely limit their reach. Instead, the people who would benefit from using them might seek emotional bonds and therapy from types of AI not designed for those purposes (indeed, new research from OpenAI suggests that interactions with its AI models have a very real impact on emotional well-being). It is highly likely that many individuals will continue to rely on more affordable, nontherapeutic chatbotssuch as ChatGPT or Character.AIfor everyday needs, ranging from generating recipe ideas to managing their mental health, he wrote.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. Anthropic can now track the bizarre inner workings of a large language model The news: The AI firm Anthropic has developed a way to peer inside a large language model and watch what it does as it comes up with a response, revealing key new insights into how the technology works. The takeaway: LLMs are even stranger than we thought.Why it matters: Its no secret that large language models work in mysterious ways. Shedding some light on how they work would expose their weaknesses, revealing why they make stuff up and can be tricked into going off the rails. It would help resolve deep disputes about exactly what these models can and cant do. And it would show how trustworthy (or not) they really are. Read the full story. Will Douglas Heaven What is Signal? The messaging app, explained. With the recent news that the Atlantics editor in chief was accidentally added to a group Signal chat for American leaders planning a bombing in Yemen, many people are wondering: What is Signal? Is it secure? If government officials arent supposed to use it for military planning, does that mean I shouldnt use it either? The answer is: Yes, you should use Signal, but government officials having top-secret conversations shouldnt use Signal.Read the full story to find out why.Jack Cushman This story is part of our MIT Technology Review Explains series, in which our writers untangle the complex, messy world of technology to help you understand what's coming next. You can read more of them here. Spare living human bodies might provide us with organs for transplantation Jessica Hamzelou This week, MIT Technology Review published a piece on bodyoidsliving bodies that cannot think or feel pain. In the piece, a trio of scientists argue that advances in biotechnology will soon allow us to create spare human bodies that could be used for research, or to provide organs for donation. If you find your skin crawling at this point, youre not the only one. Its a creepy idea, straight from the more horrible corners of science fiction. But bodyoids could be used for good. And if they are truly unaware and unable to think, the use of bodyoids wouldnt cross most peoples ethical lines, the authors argue. Im not so sure. Read the full story. This article first appeared in The Checkup, MIT Technology Reviews weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 A judge has ordered Trumps officials to preserve their secret Signal chat While officials are required by law to keep chats detailing government business, Signals messages can be set to auto-disappear. (USA Today)+ The conversation detailed an imminent attack against Houthi rebels in Yemen. (The Hill)+ A government accountability group has sued the agencies involved. (Reuters)+ The officials involved in the chat appear to have public Venmo accounts. (Wired $)2 The White House is prepared to cut up to 50% of agency staff But the final cuts could end up exceeding even that. (WP $)+ The sweeping cuts could threaten vital US statistics, too. (FT $)+ Can AI help DOGE slash government budgets? Its complex. (MIT Technology Review)3 OpenAI is struggling to keep up with demand for ChatGPT's image generation The fervor around its Studio Ghibli pictures has sent its GPUs into overdrive. (The Verge)+ Ghiblis founder is no fan of AI art. (404 Media)+ Four ways to protect your art from AI. (MIT Technology Review)4 Facebook is pivoting back towards friends and family Less news, fewer posts from people you dont know. (NYT $)+ A new tab shows purely updates from friends, with no other recommendations. (Insider $)A specialized powerhouse for AI computing, to be precise. (5 Africa is set to build its first AI factoryRest of World)+ What Africa needs to do to become a major AI player. (MIT Technology Review)Including clips of the doctored voices of well-known journalists. (6 A TikTok network spread Spanish-language immigration misinformationNBC News)7 Your TV is desperate for your dataStreamers are scrambling around for new ways to make money off the information they gather on you. (Vox) 8 This startup extracts rare earth oxides from industrial magnets Its a less intrusive way of accessing minerals vital to EV and wind turbine production. (FT $) + The race to produce rare earth elements. (MIT Technology Review)9 NASA hopes to launch its next Starliner flight as soon as later this year After its latest mission stretched from a projected eight days to nine months. (Reuters)+ Europe is finally getting serious about commercial rockets. (MIT Technology Review) 10 The Sims has been the worlds favorite life simulation game for 25 years But a new Korean game is both more realistic and multicultural. (Bloomberg $)Quote of the day Its like, can you tell the difference between a person and a person-shaped sock puppet that is holding up a sign saying, I am a sock puppet? Laura Edelson, a computer science professor at Northeastern University, is skeptical about brands abilities to ensure their ads are being shown to real humans and not bots, she tells the Wall Street Journal. The big story The race to fix space-weather forecasting before next big solar storm hits April 2024 As the number of satellites in space grows, and as we rely on them for increasing numbers of vital tasks on Earth, the need to better predict stormy space weather is becoming more and more urgent.Scientists have long known that solar activity can change the density of the upper atmosphere. But its incredibly difficult to precisely predict the sorts of density changes that a given amount of solar activity would produce.Now, experts are working on a model of the upper atmosphere to help scientists to improve their models of how solar activity affects the environment in low Earth orbit. If they succeed, theyll be able to keep satellites safe even amid turbulent space weather, reducing the risk of potentially catastrophic orbital collisions. Read the full story.Tereza Pultarova We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + This is very coola nearly-infinite virtual museum entirely generated from Wikipedia.+ How to let go of that grudge youve been harboring (you know the one)+ If your social media feeds have been plagued by hot men making bad art, youre not alone.+ Its Friday, so enjoy this 1992 recording of a very fresh-faced Pearl Jam.

Stop me if youve heard this one before: A tech company accumulates a ton of user data, hoping to figure out a business model later. That business model never arrives, the company goes under, and the data is in the wind. The latest version of that story emerged on March 24, when the onetime genetic testing darling 23andMe filed for bankruptcy. Now the fate of 15 million peoples genetic data rests in the hands of a bankruptcy judge.hearing on March 26, the judge gave 23andMe permission to seek offers for its users data. But, theres still a small chance of writing a better ending for users. After the bankruptcy filing, the immediate take from policymakers and privacy advocates was that 23andMe users should delete their accounts to prevent genetic data from falling into the wrong hands. Thats good advice for the individual user (and you can read how to do so here). But the reality is most people wont do it. Maybe they wont see the recommendations to do so. Maybe they dont know why they should be worried. Maybe they have long since abandoned an account that they dont even remember exists. Or maybe theyre just occupied with the chaos of everyday life. This means the real value of this data comes from the fact that people have forgotten about it. Given 23andMes meager revenuefewer than 4% of people who took tests pay for subscriptionsit seems inevitable that the new owner, whoever it is, will have to find some new way to monetize that data. This is a terrible deal for users who just wanted to learn a little more about themselves or their ancestry. Because genetic data is forever. Contact information can go stale over time: you can always change your password, your email, your phone number, or even your address. But a bad actor who has your genetic datawhether a cybercriminal selling it to the highest bidder, a company building a profile of your future health risk, or a government trying to identify youwill have it tomorrow and the next day and all the days after that. Users with exposed genetic data are not only vulnerable to harm today; theyre vulnerable to exploits that might be developed in the future. While 23andMe promises that it will not voluntarily share data with insurance providers, employers, or public databases, its new owner could unwind those promises at any time with a simple change in terms. In other words: If a bankruptcy court makes a mistake authorizing the sale of 23andMes user data, that mistake is likely permanent and irreparable. All this is possible because American lawmakers have neglected to meaningfully engage with digital privacy for nearly a quarter-century. As a result, services are incentivized to make flimsy, deceptive promises that can be abandoned at a moments notice. And the burden falls on users to keep track of it all, or just give up. Here, a simple fix would be to reverse that burden. A bankruptcy court could require that users individually opt in before their genetic data can be transferred to 23andMes new owners, regardless of who those new owners are. Anyone who didnt respond or who opted out would have the data deleted. Bankruptcy proceedings involving personal data dont have to end badly. In 2000, the Federal Trade Commission settled with the bankrupt retailer ToySmart to ensure that its customer data could not be sold as a stand-alone asset, and that customers would have to affirmatively consent to unexpected new uses of their data. And in 2015, the FTC intervened in the bankruptcy of RadioShack to ensure that it would keep its promises never to sell the personal data of its customers. (RadioShack eventually agreed to destroy it.) The ToySmart case also gave rise to the role of the consumer privacy ombudsman. Bankruptcy judges can appoint an ombuds to help the court consider how the sale of personal data might affect the bankruptcy estate, examining the potential harms or benefits to consumers and any alternatives that might mitigate those harms. The U.S. Trustee has requested the appointment of an ombuds in this case. While scholars have called for the role to have more teeth and for the FTC and states to intervene more often, a framework for protecting personal data in bankruptcy is available. And ultimately, the bankruptcy judge has broad power to make decisions about how (or whether) property in bankruptcy is sold. Here, 23andMe has a more permissive privacy policy than ToySmart or RadioShack. But the risks incurred if genetic data falls into the wrong hands or is misused are severe and irreversible. And given 23andMes failure to build a viable business model from testing kits, it seems likely that a new business would use genetic data in ways that users wouldnt expect or want. An opt-in requirement for genetic data solves this problem. Genetic data (and other sensitive data) could be held by the bankruptcy trustee and released as individual users gave their consent. If users failed to opt in after a period of time, the remaining data would be deleted. This would incentivize 23andMes new owners to earn user trust and build a business that delivers value to users, instead of finding unexpected ways to exploit their data. And it would impose virtually no burden on the people whose genetic data is at risk: after all, they have plenty more DNA to spare. Consider the alternative. Before 23andMe went into bankruptcy, its then-CEO made two failed attempts to buy it, at reported valuations of $74.7 million and $12.1 million. Using the higher offer, and with 15 million users, that works out to a little under $5 per user. Is it really worth it to permanently risk a persons genetic privacy just to add a few dollars in value to the bankruptcy estate? Of course, this raises a bigger question: Why should anyone be able to buy the genetic data of millions of Americans in a bankruptcy proceeding? The answer is simple: Lawmakers allow them to. Federal and state inaction allows companies to dissolve promises about protecting Americans most sensitive data at a moments notice. When 23andMe was founded, in 2006, the promise was that personalized health care was around the corner. Today, 18 years later, that era may really be almost here. But with privacy laws like ours, who would trust it? Keith Porcaro is the Rueben Everett Senior Lecturing Fellow at Duke Law School.

This week, MIT Technology Review published a piece on bodyoidsliving bodies that cannot think or feel pain. In the piece, a trio of scientists argue that advances in biotechnology will soon allow us to create spare human bodies that could be used for research, or to provide organs for donation. If you find your skin crawling at this point, youre not the only one. Its a creepy idea, straight from the more horrible corners of science fiction. But bodyoids could be used for good. And if they are truly unaware and unable to think, the use of bodyoids wouldnt cross most peoples ethical lines, the authors argue. Im not so sure. Either way, theres no doubt that developments in science and biotechnology are bringing us closer to the potential reality of bodyoids. And the idea is already stirring plenty of ethical debate and controversy. One of the main arguments made for bodyoids is that they could provide spare human organs. Theres a huge shortage of organs for transplantation. More than 100,000 people in the US are waiting for a transplant, and 17 people on that waiting list die every day. Human bodyoids could serve as a new source. Scientists are working on other potential solutions to this problem. One approach is the use of gene-edited animal organs. Animal organs dont typically last inside human bodiesour immune systems will reject them as foreign. But a few companies are creating pigs with a series of gene edits that make their organs more acceptable to human bodies. A handful of living people have received gene-edited pig organs. David Bennett Sr. was the first person to get a gene-edited pig heart, in 2022, and Richard Slayman was the first to get a kidney, in early 2024. Unfortunately, both men died around two months after their surgery. But Towana Looney, the third living person to receive a gene-edited pig kidney, has been doing well. She had her transplant surgery in late November of last year. I am full of energy. I got an appetite Ive never had in eight years, she said at the time. I can put my hand on this kidney and feel it buzzing. She returned home in February. At least one company is taking more of a bodyoid-like approach. Renewal Bio, a biotech company based in Israel, hopes to grow embryo-stage versions of people for replacement organs. Their approach is based on advances in the development of synthetic embryos. (Im putting that term in quotation marks because, while its the simplest descriptor of what they are, a lot of scientists hate the term.) Embryos start with the union of an egg cell and a sperm cell. But scientists have been working on ways to make embryos using stem cells instead. Under the right conditions, these cells can divide into structures that look a lot like a typical embryo. Scientists dont know how far these embryo-like structures will be able to develop. But theyre already using them to try to get cows and monkeys pregnant. And no one really knows how to think about synthetic human embryos. Scientists dont even really know what to call them. Rules stipulate that typical human embryos may be grown in the lab for a maximum of 14 days. Should the same rules apply to synthetic ones? The very existence of synthetic embryos is throwing into question our understanding of what a human embryo even is. Is it the thing that is only generated from the fusion of a sperm and an egg? Naomi Moris, a developmental biologist at the Crick Institute in London, said to me a couple of years ago. Is it something to do with the cell types it possesses, or the [shape] of the structure? The authors of the new MIT Technology Review piece also point out that such bodyoids could also help speed scientific and medical research. At the moment, most drug research must be conducted in lab animals before clinical trials can start. But nonhuman animals may not respond the same way people do, and the vast majority of treatments that look super-promising in mice fail in humans. Such research can feel like a waste of both animal lives and time. Scientists have been working on solutions to these problems, too. Some are creating organs on chipsminiature collections of cells organized on a small piece of polymer that may resemble full-size organs and can be used to test the effects of drugs. Others are creating digital representations of human organs for the same purpose. Such digital twins can be extensively modeled, and can potentially be used to run clinical trials in silico. Both of these approaches seem somehow more palatable to me, personally, than running experiments on a human created without the capacity to think or feel pain. The idea reminds me of the recent novel Tender Is the Flesh by Agustina Bazterrica, in which humans are bred for consumption. In the book, their vocal cords are removed so that others do not have to hear them scream. When it comes to real-world biotechnology, though, our feelings about what is acceptable tend to shift. In vitro fertilization was demonized when it was first developed, for instance, with opponents arguing that it was unnatural, a perilous insult, and the biggest threat since the atom bomb. It is estimated that more than 12 million people have been born through IVF since Louise Brown became the first test tube baby 46 years ago. I wonder how well all feel about bodyoids 46 years from now. This article first appeared in The Checkup,MIT Technology Reviewsweekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first,sign up here.

The AI firm Anthropic has developed a way to peer inside a large language model and watch what it does as it comes up with a response, revealing key new insights into how the technology works. The takeaway: LLMs are even stranger than we thought. The Anthropic team was surprised by some of the counterintuitive workarounds that large language models appear to use to complete sentences, solve simple math problems, suppress hallucinations, and more, says Joshua Batson, a research scientist at the company. Its no secret that large language models work in mysterious ways. Fewif anymass-market technologies have ever been so little understood. That makes figuring out what makes them tick one of the biggest open challenges in science. But its not just about curiosity. Shedding some light on how these models work would expose their weaknesses, revealing why they make stuff up and can be tricked into going off the rails. It would help resolve deep disputes about exactly what these models can and cant do. And it would show how trustworthy (or not) they really are. Batson and his colleagues describe their new work in two reports published today. The first presents Anthropics use of a technique called circuit tracing, which lets researchers track the decision-making processes inside a large language model step by step. Anthropic used circuit tracing to watch its LLM Claude 3.5 Haiku carry out various tasks. The second (titled On the Biology of a Large Language Model) details what the team discovered when it looked at 10 tasks in particular. I think this is really cool work, says Jack Merullo, who studies large language models at Brown University in Providence, Rhode Island, and was not involved in the research. Its a really nice step forward in terms of methods. Circuit tracing is not itself new. Last year Merullo and his colleagues analyzed a specific circuit in a version of OpenAIs GPT-2, an older large language model that OpenAI released in 2019. But Anthropic has now analyzed a number of different circuits as a far larger and far more complex model carries out multiple tasks. Anthropic is very capable at applying scale to a problem, says Merullo. Eden Biran, who studies large language models at Tel Aviv University, agrees. Finding circuits in a large state-of-the-art model such as Claude is a nontrivial engineering feat, he says. And it shows that circuits scale up and might be a good way forward for interpreting language models. Circuits chain together different partsor componentsof a model. Last year, Anthropic identified certain components inside Claude that correspond to real-world concepts. Some were specific, such as Michael Jordan or greenness; others were more vague, such as conflict between individuals. One component appeared to represent the Golden Gate Bridge. Anthropic researchers found that if they turned up the dial on this component, Claude could be made to self-identify not as a large language model but as the physical bridge itself. The latest work builds on that research and the work of others, including Google DeepMind, to reveal some of the connections between individual components. Chains of components are the pathways between the words put into Claude and the words that come out. Its tip-of-the-iceberg stuff. Maybe were looking at a few percent of whats going on, says Batson. But thats already enough to see incredible structure. Growing LLMs Researchers at Anthropic and elsewhere are studying large language models as if they were natural phenomena rather than human-built software. Thats because the models are trained, not programmed. They almost grow organically, says Batson. They start out totally random. Then you train them on all this data and they go from producing gibberish to being able to speak different languages and write software and fold proteins. There are insane things that these models learn to do, but we dont know how that happened because we didnt go in there and set the knobs. Sure, its all math. But its not math that we can follow. Open up a large language model and all you will see is billions of numbersthe parameters, says Batson. Its not illuminating. Anthropic says it was inspired by brain-scan techniques used in neuroscience to build what the firm describes as a kind of microscope that can be pointed at different parts of a model while it runs. The technique highlights components that are active at different times. Researchers can then zoom in on different components and record when they are and are not active. Take the component that corresponds to the Golden Gate Bridge. It turns on when Claude is shown text that names or describes the bridge or even text related to the bridge, such as San Francisco or Alcatraz. Its off otherwise. Yet another component might correspond to the idea of smallness: We look through tens of millions of texts and see its on for the word small, its on for the word tiny, its on for the word petite, its on for words related to smallness, things that are itty-bitty, like thimblesyou know, just small stuff, says Batson. Having identified individual components, Anthropic then follows the trail inside the model as different components get chained together. The researchers start at the end, with the component or components that led to the final response Claude gives to a query. Batson and his team then trace that chain backwards. Odd behavior So: What did they find? Anthropic looked at 10 different behaviors in Claude. One involved the use of different languages. Does Claude have a part that speaks French and another part that speaks Chinese, and so on? The team found that Claude used components independent of any language to answer a question or solve a problem and then picked a specific language when it replied. Ask it What is the opposite of small? in English, French, and Chinese and Claude will first use the language-neutral components related to smallness and opposites Anthropic also looked at how Claude solved simple math problems. The team found that the model seems to have developed its own internal strategies that are unlike those it will have seen in its training data. Ask Claude to add 36 and 59 and the model will go through a series of odd steps, including first adding a selection of approximate values (add 40ish and 60ish, add 57ish and 36ish). Towards the end of its process, it comes up with the value 92ish. Meanwhile, another sequence of steps focuses on the last digits, 6 and 9, and determines that the answer must end in a 5. Putting that together with 92ish gives the correct answer of 95. And yet if you then ask Claude how it worked that out, it will say something like: I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95. In other words, it gives you a common approach found everywhere online rather than what it actually did. Yep! LLMs are weird. (And not to be trusted.) The steps that Claude 3.5 Haiku used to solve a simple math problem were not what Anthropic expectedthey're not the steps Claude claimed it took either. ANTHROPIC This is clear evidence that large language models will give reasons for what they do that do not necessarily reflect what they actually did. But this is true for people too, says Batson: You ask somebody, Why did you do that? And theyre like, Um, I guess its because I was . You know, maybe not. Maybe they were just hungry and thats why they did it. Biran thinks this finding is especially interesting. Many researchers study the behavior of large language models by asking them to explain their actions. But that might be a risky approach, he says: As models continue getting stronger, they must be equipped with better guardrails. I believeand this work also showsthat relying only on model outputs is not enough. A third task that Anthropic studied was writing poems. The researchers wanted to know if the model really did just wing it, predicting one word at a time. Instead they found that Claude somehow looked ahead, picking the word at the end of the next line several words in advance. For example, when Claude was given the prompt A rhyming couplet: He saw a carrot and had to grab it, the model responded, His hunger was like a starving rabbit. But using their microscope, they saw that Claude had already hit upon the word rabbit when it was processing grab it. It then seemed to write the next line with that ending already in place. This might sound like a tiny detail. But it goes against the common assumption that large language models always work by picking one word at a time in sequence. The planning thing in poems blew me away, says Batson. Instead of at the very last minute trying to make the rhyme make sense, it knows where its going. I thought that was cool, says Merullo. One of the joys of working in the field is moments like that. Theres been maybe small bits of evidence pointing toward the ability of models to plan ahead, but its been a big open question to what extent they do. Anthropic then confirmed its observation by turning off the placeholder component for rabbitness. Claude responded with His hunger was a powerful habit. And when the team replaced rabbitness with greenness, Anthropic also explored why Claude sometimes made stuff up, a phenomenon known as hallucination. Hallucination is the most natural thing in the world for these models, given how theyre just trained to give possible completions, says Batson. The real question is, How in Gods name could you ever make it not do that? The latest generation of large language models, like Claude 3.5 and Gemini and GPT-4o, hallucinate far less than previous versions, thanks to extensive post-training (the steps that take an LLM trained on the internet and turn it into a usable chatbot). But Batsons team was surprised to find that this post-training seems to have made Claude refuse to speculate as a default behavior. When it did respond with false information, it was because some other component had overridden the dont speculate component. This seemed to happen most often when the speculation involved a celebrity or other well-known entity. Its as if the amount of information available pushed the speculation through, despite the default setting. When Anthropic overrode the dont speculate component to test this, Claude produced lots of false statements about individuals, including claiming that Batson was famous for inventing the Batson principle (he isnt). Still unclear Because we know so little about large language models, any new insight is a big step forward. A deep understanding of how these models work under the hood would allow us to design and train models that are much better and stronger, says Biran. But Batson notes there are still serious limitations. Its a misconception that weve found all the components of the model or, like, a Gods-eye view, he says. Some things are in focus, but other things are still uncleara distortion of the microscope. And it takes several hours for a human researcher to trace the responses to even very short prompts. Whats more, these models can do a remarkable number of different things, and Anthropic has so far looked at only 10 of them. Batson also says there are big questions that this approach wont answer. Circuit tracing can be used to peer at the structures inside a large language model, but it wont tell you how or why those structures formed during training. Thats a profound question that we dont address at all in this work, he says. But Batson sees this as the start of a new era in which it is possible, at last, to find real evidence for how these models work: We dont have to be, like: Are they thinking? Are they reasoning? Are they dreaming? Are they memorizing? Those are all analogies. But if we can literally see step by step what a model is doing, maybe now we dont need analogies.

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand whats coming next. You can read more from the series here. With the recent news that the Atlantics editor in chief was accidentally added to a group Signal chat for American leaders planning a bombing in Yemen, many people are wondering: What is Signal? Is it secure? If government officials arent supposed to use it for military planning, does that mean I shouldnt use it either? The answer is: Yes, you should use Signal, but government officials having top-secret conversations shouldnt use Signal. Read on to find out why. What is Signal? Signal is an app you can install on your iPhone or Android phone, or on your computer. It lets you send secure texts, images, and phone or video chats with other people or groups of people, just like iMessage, Google Messages, WhatsApp, and other chat apps. Installing Signal is a two-minute processagain, its designed to work just like other popular texting apps. Why is it a problem for government officials to use Signal? Signal is very secureas well see below, its the best option out there for having private conversations with your friends on your cell phone. But you shouldnt use it if you have a legal obligation to preserve your messages, such as while doing government business, because Signal prioritizes privacy over ability to preserve data. Its designed to securely delete data when youre done with it, not to keep it. This makes it uniquely unsuited for following public record laws. You also shouldnt use it if your phone might be a target of sophisticated hackers, because Signal can only do its job if the phone it is running on is secure. If your phone has been hacked, then the hacker can read your messages regardless of what software you are running. This is why you shouldnt use Signal to discuss classified material or military plans. For military communication your civilian phone is always considered hacked by adversaries, so you should instead use communication equipment that is saferequipment that is physically guarded and designed to do only one job, making it harder to hack. What about everyone else? Signal is designed from bottom to top as a very private space for conversation. Cryptographers are very sure that as long as your phone is otherwise secure, no one can read your messages. Why should you want that? Because private spaces for conversation are very important. In the US, the First Amendment recognizes, in the right to freedom of assembly, that we all need private conversations among our own selected groups in order to function. And you dont need the First Amendment to tell you that. You know, just like everyone else, that you can have important conversations in your living room, bedroom, church coffee hour, or meeting hall that you could never have on a public stage. Signal gives us the digital equivalent of thatits a space where we can talk, among groups of our choice, about the private things that matter to us, free of corporate or government surveillance. Our mental health and social functioning require that. So if youre not legally required to record your conversations, and not planning secret military operations, go ahead and use Signalyou deserve the privacy. How do we know Signal is secure? People often give up on finding digital privacy and end up censoring themselves out of caution. So are there really private ways to talk on our phones, or should we just assume that everything is being read anyway? The good news is: For most of us who arent individually targeted by hackers, we really can still have private conversations. Signal is designed to ensure that if you know your phone and the phones of other people in your group havent been hacked (more on that later), you dont have to trust anything else. It uses many techniques from the cryptography community to make that possible. Most important and well-known is end-to-end encryption, which means that messages can be read only on the devices involved in the conversation and not by servers passing the messages back and forth. But Signal uses other techniques to keep your messages private and safe as well. For example, it goes to great lengths to make it hard for the Signal server itself to know who else you are talking to (a feature known as sealed sender), or for an attacker who records traffic between phones to later decrypt the traffic by seizing one of the phones (perfect forward secrecy). These are only a few of many security properties built into the protocol, which is well enough designed and vetted for other messaging apps, such as WhatsApp and Google Messages, to use the same one. Signal is also designed so we dont have to trust the people who make it. The source code for the app is available online and, because of its popularity as a security tool, is frequently audited by experts. And even though its security does not rely on our trust in the publisher,it does come from a respected source: the Signal Technology Foundation, a nonprofit whose mission is to protect free expression and enable secure global communication through open-source privacy technology. The app itself, and the foundation, grew out of a community of prominent privacy advocates. The foundation was started by Moxie Marlinspike, a cryptographer and longtime advocate of secure private communication, and Brian Acton, a cofounder of WhatsApp. Why do people use Signal over other text apps? Are other ones secure? Many apps offer end-to-end encryption, and its not a bad idea to use them for a measure of privacy. But Signal is a gold standard for private communication because it is secure by default: Unless you add someone you didnt mean to, its very hard for a chat to accidentally become less secure than you intended. Thats not necessarily the case for other apps. For example, iMessage conversations are sometimes end-to-end encrypted, but only if your chat has blue bubbles, and they arent encrypted in iCloud backups by default. Google Messages are sometimes end-to-end encrypted, but only if the chat shows a lock icon. WhatsApp is end-to-end encrypted but logs your activity, including how you interact with others using our Services. Signal is careful not to record who you are talking with, to offer ways to reliably delete messages, and to keep messages secure even in online phone backups. This focus demonstrates the benefits of an app coming from a nonprofit focused on privacy rather than a company that sees security as a nice to have feature alongside other goals. (Conversely, and as a warning, using Signal makes it rather easier to accidentally lose messages! Again, it is not a good choice if you are legally required to record your communication.) Applications like WhatsApp, iMessage, and Google Messages do offer end-to-end encryption and can offer much better security than nothing. The worst option of all is regular SMS text messages (green bubbles on iOS)those are sent unencrypted and are likely collected by mass government surveillance. Wait, how do I know that my phone is secure? Signal is an excellent choice for privacy if you know that the phones of everyone youre talking with are secure. But how do you know that? Its easy to give up on a feeling of privacy if you never feel good about trusting your phone anyway. One good place to start for most of us is simply to make sure your phone is up to date. Governments often do have ways of hacking phones, but hacking up-to-date phones is expensive and risky and reserved for high-value targets. For most people, simply having your software up to date will remove you from a category that hackers target. If youre a potential target of sophisticated hacking, then dont stop there. Youll need extra security measures, andguides from the Freedom of the Press Foundation and the Electronic Frontier Foundation are a good place to start. But you dont have to be a high-value target to value privacy. The rest of us can do our part to re-create that private living room, bedroom, church, or meeting hall simply by using an up-to-date phone with an app that respects our privacy. Jack Cushman is a fellow of the Berkman Klein Center for Internet and Society and directs the Library Innovation Lab at Harvard Law School Library. He is an appellate lawyer, computer programmer, and former board member of the ACLU of Massachusetts.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. Inside a romance scam compoundand how people get tricked into being there Gaveshs journey had started, seemingly innocently, with a job ad on Facebook promising work he desperately needed. Instead, he found himself trafficked into a business commonly known as pig butcheringa form of fraud in which scammers form romantic or other close relationships with targets online and extract money from them. The Chinese crime syndicates behind the scams have netted billions of dollars, and they have used violence and coercion to force their workers, many of them people trafficked like Gavesh, to carry out the frauds from large compounds, several of which operate openly in the quasi-lawless borderlands of Myanmar.We spoke to Gavesh and five other workers from inside the scam industry, as well as anti-trafficking experts and technology specialists. Their testimony reveals how global companies, including American social media and dating apps and international cryptocurrency and messaging platforms, have given the fraud business the means to become industrialized. By the same token, it is Big Tech that may hold the key to breaking up the scam syndicatesif only these companies can be persuaded or compelled to act. Read the full story. Peter Guest & Emily Fishbein How to save a glacier Theres a lot we dont understand about how glaciers move and how soon some of the most significant ones could collapse into the sea. That could be a problem, since melting glaciers could lead to multiple feet of sea-level rise this century, potentially displacing millions of people who live and work along the coasts. A new group is aiming not only to further our understanding of glaciers but also to look into options to save them if things move toward a worst-case scenario, as my colleague James Temple outlined in his latest story. One idea: refreezing glaciers in place. The whole thing can sound like science fiction. But once you consider how huge the stakes are, I think it gets easier to understand why some scientists say we should at least be exploring these radical interventions. Read the full story. Casey Crownhart This article is from The Spark, MIT Technology Reviews weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here. MIT Technology Review Narrated: How tracking animal movement may save the planet Researchers have long dreamed of creating an Internet of Animals. And theyre getting closer to monitoring 100,000 creaturesand revealing hidden facets of our shared world. This is our latest story to be turned into a MIT Technology Review Narrated podcast, which were publishing each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as its released. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 Donald Trump has announced 25% tariffs on imported cars and parts The measures are likely to make new cars significantly more expensive for Americans. (NYT $)+ Moving car manufacturing operations to the US wont be easy. (WP $)+ Its not just big businesses that will suffer, either. (The Atlantic $)+ How Trumps tariffs could drive up the cost of batteries, EVs, and more. (MIT Technology Review)2 China is developing an AI system to increase its online censorship A leaked dataset demonstrates how LLMs could rapidly filter undesirable material. (TechCrunch)3 Trump may reduce tariffs on China to encourage a TikTok dealThe Chinese-owned company has until April 5 to find a new US owner. (Insider $) + The national security concerns surrounding it havent gone away, though. (NYT $)4 OpenAIs new image generator can ape Studio Ghibli's distinctive styleWhich raises the question of whether the model was trained on Ghiblis images. (TechCrunch) + The tools popularity means its rollout to non-paying users has been delayed. (The Verge)+ The AI lab waging a guerrilla war over exploitative AI. (MIT Technology Review)5 DOGE planned to dismantle USAID from the beginningNew court filings reveal the departments ambitions to infiltrate the system. (Wired $) + Can AI help DOGE slash government budgets? Its complex. (MIT Technology Review)6 Wildfires are getting worse in the southwest of the US While federal fire spending is concentrated mainly in the west, the risk is rising in South Carolina and Texas too. (WP $)+ North and South Carolina were recovering from Hurricane Helene when the fires struck. (The Guardian)+ How AI can help spot wildfires. (MIT Technology Review) 7 A quantum computer has generatedand verifiedtruly random numbersWhich is good news for cryptographers. (Bloomberg $) + Cybersecurity analysts are increasingly worried about the so-called Q-Day. (Wired $)+ Amazons first quantum computing chip makes its debut. (MIT Technology Review)8 Whats next for weight-loss drugs New Scientist $) + Drugs like Ozempic now make up 5% of prescriptions in the US. (MIT Technology Review)9 At least weve still got memes Poking fun at the Trump administrations decisions is a form of online resistance. (New Yorker $)10 Can you truly be friends with a chatbot? People are starting to find out. (Vox)+ The AI relationship revolution is already here. (MIT Technology Review)Quote of the day I cant imagine any professional I know committing this egregious a lapse in judgement. A government technology leader tells Fast Company why top Trump officials decision to use unclassified messaging app Signal to discuss war plans is so surprising. The big story Why one developer wont quit fighting to connect the USs grids September 2024 Michael Skelly hasnt learned to take no for an answer. For much of the last 15 years, the energy entrepreneur has worked to develop long-haul transmission lines to carry wind power across the Great Plains, Midwest, and Southwest. But so far, he has little to show for the effort. Skelly has long argued that building such lines and linking together the nations grids would accelerate the shift from coal- and natural-gas-fueled power plants to the renewables needed to cut the pollution driving climate change. But his previous business shut down in 2019, after halting two of its projects and selling off interests in three more. Skelly contends he was early, not wrong, and that the market and policymakers are increasingly coming around to his perspective. After all, the US Department of Energy just blessed his latest companys proposed line with hundreds of millions in grants. Read the full story. James Temple We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + Severances Adam Scott sure has interesting taste in music.+ While were not 100% sure if Millie is definitely the worlds oldest cat, one thing we know for sure is that she lives a life of luxury.+ Hiking trails are covered in beautiful wildflowers right now; just make sure you tread carefully.+ This is a really charming look at how girls live in America right now.

Glaciers generally move so slowly you cant see their progress with the naked eye. (Their pace is glacial.) But these massive bodies of ice do march downhill, with potentially planet-altering consequences. Theres a lot we dont understand about how glaciers move and how soon some of the most significant ones could collapse into the sea. That could be a problem, since melting glaciers could lead to multiple feet of sea-level rise this century, potentially displacing millions of people who live and work along the coasts. A new group is aiming not only to further our understanding of glaciers but also to look into options to save them if things move toward a worst-case scenario, as my colleague James Temple outlined in his latest story. One idea: refreezing glaciers in place. The whole thing can sound like science fiction. But once you consider how huge the stakes are, I think it gets easier to understand why some scientists say we should at least be exploring these radical interventions. Its hard to feel very optimistic about glaciers these days. (The Thwaites Glacier in West Antarctica is often called the doomsday glaciernot alarming at all!) Take two studies published just in the last month, for example. The British Antarctic Survey released the most detailed map to date of Antarcticas bedrockthe foundation under the continents ice. With twice as many data points as before, the study revealed that more ice than we thought is resting on bedrock thats already below sea level. That means seawater can flow in and help melt ice faster, so Antarcticas ice is more vulnerable than previously estimated. Another study examined subglacial riversstreams that flow under the ice, often from subglacial lakes. The team found that the fastest-moving glaciers have a whole lot of water moving around underneath them, which speeds melting and lubricates the ice sheet so it slides faster, in turn melting even more ice. And those are just two of the most recent surveys. Look at any news site and its probably delivered the same gnarly message at some point recently: The glaciers are melting faster than previously realized. (Our site has one, too: Greenlands ice sheet is less stable than we thought, from 2016.) The new group is joining the race to better understand glaciers. Arte Glacier Initiative, a nonprofit research organization founded by scientists at MIT and Dartmouth, has already awarded its first grants to researchers looking into how glaciers melt and plans to study the possibility of reversing those fortunes, as James exclusively reported last week. Brent Minchew, one of the groups cofounders and an associate professor of geophysics at MIT, was drawn to studying glaciers because of their potential impact on sea-level rise. But over the years, I became less content with simply telling a more dramatic story about how things were goingand more open to asking the question of what can we do about it, he says. Minchew is among the researchers looking into potential plans to alter the future of glaciers. Strategies being proposed by groups around the world include building physical supports to prop them up and installing massive curtains to slow the flow of warm water that speeds melting. Another approach, which will be the focus of Arte, is called basal intervention. It basically involves drilling holes in glaciers, which would allow water flowing underneath the ice to be pumped out and refrozen, hopefully slowing them down. If you have questions about how all this would work, youre not alone. These are almost inconceivably huge engineering projects, theyd be expensive, and theyd face legal and ethical questions. Nobody really owns Antarctica, and its governed by a huge treatyhow could we possibly decide whether to move forward with these projects? Then theres the question of the potential side effects. Just look at recent news from the Arctic Ice Project, which was researching how to slow the melting of sea ice by covering it with substances designed to reflect sunlight away. (Sea ice is different from glaciers, but some of the key issues are the same.) One of the projects largest field experiments involved spreading tiny silica beads, sort of like sand, over 45,000 square feet of ice in Alaska. But after new research revealed that the materials might be disrupting food chains, the organization announced that its concluding its research and winding down operations. Cutting our emissions of greenhouse gases to stop climate change at the source would certainly be more straightforward than spreading beads on ice, or trying to stop a 74,000-square-mile glacier in its tracks. But were not doing so hot on cutting emissionsin fact, levels of carbon dioxide in the atmosphere rose faster than ever in 2024. And even if the world stopped polluting the atmosphere with planet-warming gases today, things may have already gone too far to save some of the most vulnerable glaciers. The longer I cover climate change and face the situation were in, the more I understand the impulse to at least consider every option out there, even if it sounds like science fiction. This article is from The Spark, MIT Technology Reviews weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here.

Heading north in the dark, the only way Gavesh could try to track his progress through the Thai countryside was by watching the road signs zip by. The Jeeps three occupantsGavesh, a driver, and a young Chinese womanhad no languages in common, so they drove for hours in nervous silence as they wove their way out of Bangkok and toward Mae Sot, a city on Thailands western border with Myanmar. When they reached the city, the driver pulled off the road toward a small hotel, where another car was waiting. I had some suspicionslike, why are we changing vehicles? Gavesh remembers. But it happened so fast. They left the highway and drove on until, in total darkness, they parked at what looked like a private house. We stopped the vehicle. There were people gathered. Maybe 10 of them. They took the luggage and they asked us to come, Gavesh says. One was going in front, there was another one behind, and everyone said: Go, go, go. Gavesh and the Chinese woman were marched through the pitch-black fields by flashlight to a riverside where a boat was moored. By then, it was far too late to back out. Gaveshs journey had started, seemingly innocently, with a job ad on Facebook promising work he desperately needed. Instead, he found himself trafficked into a business commonly known as pig butcheringa form of fraud in which scammers form romantic or other close relationships with targets online and extract money from them. The Chinese crime syndicates behind the scams have netted billions of dollars, and they have used violence and coercion to force their workers, many of them people trafficked like Gavesh, to carry out the frauds from large compounds, several of which operate openly in the quasi-lawless borderlands of Myanmar. We spoke to Gavesh and five other workers from inside the scam industry, as well as anti-trafficking experts and technology specialists. Their testimony reveals how global companies, including American social media and dating apps and international cryptocurrency and messaging platforms, have given the fraud business the means to become industrialized. By the same token, it is Big Tech that may hold the key to breaking up the scam syndicatesif only these companies can be persuaded or compelled to act. Were identifying Gavesh using a pseudonym to protect his identity. He is from a country in South Asia, one he asked us not to name. He hasnt shared his story much, and he still hasnt told his family. He worries about how theyd handle it. Until the pandemic, he had held down a job in the tourism industry. But lockdowns had gutted the sector, and two years later he was working as a day laborer to support himself and his father and sister. I was fed up with my life, he says. I was trying so hard to find a way to get out. When he saw the Facebook post in mid-2022, it seemed like a godsend. A company in Thailand was looking for English-speaking customer service and data entry specialists. The monthly salary was $1,500far more than he could earn at homewith meals, travel costs, a visa, and accommodation included. I knew if I got this job, my life would turn around. I would be able to give my family a good life, Gavesh says. What came next was life-changing, but not in the way Gavesh had hoped. The advert was a fraudand a classic tactic syndicates use to force workers like Gavesh into an economy that operates as something like a dark mirror of the global outsourcing industry. The true scale of this type of fraud is hard to estimate, but the United Nations reported in 2023 that hundreds of thousands of people had been trafficked to work as online scammers in Southeast Asia. One 2024 study, from the University of Texas, estimates that the criminal syndicates that run these businesses have stolen at least $75 billion since 2020. These schemes have been going on for more than two decades, but theyve started to capture global attention only recently, as the syndicates running them increasingly shift from Chinese targets toward the West. And even as investigators, international organizations, and journalists gradually pull back the curtain on the brutal conditions inside scamming compounds and document their vast scale, what is far less exposed is the pivotal role platforms owned by Big Tech play throughout the industryfrom initially coercing individuals to become scammers to, finally, duping scam targets out of their life savings. As losses mount, governments and law enforcement agencies have looked for ways to disrupt the syndicates, which have become adept at using ungoverned spaces in lawless borderlands and partnering with corrupt regimes. But on the whole, the syndicates have managed to stay a step ahead of law enforcementin part by relying on services from the worlds tech giants. Apple iPhones are their preferred scamming tools. Meta-owned Facebook and WhatsApp are used to recruit people into forced labor, as is Telegram. Social media and messaging platforms, including Facebook, Instagram, WhatsApp, WeChat, and X, provide spaces for scammers to find and lure targets. So do dating apps, including Tinder. Some of the scam compounds have their own Starlink terminals. And cryptocurrencies like tether and global crypto platforms like Binance have allowed the criminal operations to move money with little or no oversight. Scam workers sit inside Myanmar's KK Park, a notorious fraud hub near the border with Thailand, following a recent crackdown by law enforcement.REUTERS Private-sector corporations are, unfortunately, inadvertently enabling this criminal industry, says Andrew Wasuwongse, the Thailand country director at the anti-trafficking nonprofit International Justice Mission (IJM). The private sector holds significant tools and responsibility to disrupt and prevent its further growth. Yet while the tech sector has, slowly, begun to roll out anti-scam tools and policies, experts in human trafficking, platform integrity, and cybercrime tell us that these measures largely focus on the downstream problem: the losses suffered by the victims of the scams. That approach overlooks the other set of victims, often from lower-income countries, at the far end of a fraud supply chain that is built on human miseryand on Big Tech. Meanwhile, the scams continue on a mass scale. Tech companies could certainly be doing more to crack down, the experts say. Even relatively small interventions, they argue, could start to erode the business model of the scam syndicates; with enough of these, the whole business could start to founder. The trick is: How do you make it unprofitable? says Eric Davis, a platform integrity expert and senior vice president of special projects at the Institute for Security and Technology (IST), a think tank in California. How do you create enough friction? That question is only becoming more urgent as many tech companies pull back on efforts to moderate their platforms, artificial intelligence supercharges scam operations, and the Trump administration signals broad support for deregulation of the tech sector while withdrawing support from organizations that study the scams and support the victims. All these trends may further embolden the syndicates. And even as the human costs keep building, global governments exert ineffectual pressureif any at allon the tech sector to turn its vast financial and technical resources against a criminal economy that has thrived in the spaces Silicon Valley built. Capturing a vulnerable workforce The roots of pig butchering scams reach back to the offshore gambling industry that emerged from China in the early 2000s. Online casinos had become hugely popular in China, but the government cracked down, forcing the operators to relocate to Cambodia, the Philippines, Laos, and Myanmar. There, they could continue to target Chinese gamblers with relative impunity. Over time, the casinos began to use social media to entice people back home, deploying scam-like tactics that frequently centered on attractive and even nude dealers. The doubts didnt really start until after Gavesh reached Bangkoks Suvarnabhumi Airport. As time ticked by, it began to occur to him that he was alone, with no money, no return ticket, and no working SIM card. Often the romance scam was a part of thatbuilding romantic relationships with people that you eventually would aim to hook, says Jason Tower, Myanmar country director at the United States Institute of Peace (USIP), a research and diplomacy organization funded by the US government, who researches the cyber scam industry. (USIPs leadership was recently targeted by the Trump administration and Elon Musks Department of Government Efficiency task force, leaving the organizations future uncertain; its website, which previously housed its research, is also currently offline.) By the late 2010s, many of the casinos were big, professional operations. Gradually, says Tower, the business model turned more sinister, with a tactic called sha zhu pan in Chinese emerging as a core strategy. Scamming operatives work to fatten up or cultivate a target by building a relationship before going in for the slaughterpersuading them to invest in a supposedly once-in-a-lifetime scheme and then absconding with the money. That actually ended up being much, much more lucrative than online gambling, Tower says. (The international law enforcement organization Interpol no longer uses the graphic term pig butchering, citing concerns that it dehumanizes and stigmatizes victims.) Like other online industries, the romance scamming business was supercharged by the pandemic. There were simply more isolated people to defraud, and more people out of work who might be persuaded to try scamming othersor who were vulnerable to being trafficked into the industry. Initially, most of the workers carrying out the frauds were Chinese, as were the fraud victims. But after the government in Beijing tightened travel restrictions, making it hard to recruit Chinese laborers, the syndicates went global. They started targeting more Western markets and turning, Tower says, to much more malign types of approaches to tricking people into scam centers. Getting recruited Gavesh was scrolling through Facebook when he saw the ad. He sent his rsum to a Telegram contact number. A human resources representative replied and had him demonstrate his English and typing skills over video. It all felt very professional. I didnt have any reason to suspect, he says. The doubts didnt really start until after he reached Bangkoks Suvarnabhumi Airport. After being met at arrivals by a man who spoke no English, he was left to wait. As time ticked by, it began to occur to Gavesh that he was alone, with no money, no return ticket, and no working SIM card. Finally, the Jeep arrived to pick him up. Hours later, exhausted, he was on a boat crossing the Moei River from Thailand into Myanmar. On the far bank, a group was waiting. One man was in military uniform and carried a gun. In my country, if we see an army guy when we are in trouble, we feel safe, Gavesh says. So my initial thoughts were: Okay, theres nothing to be worried about. They hiked a kilometer across a sodden paddy field and emerged at the other side caked in mud. There a van was parked, and the driver took them to what he called, in broken English, the office. They arrived at the gate of a huge compound, surrounded by high walls topped with barbed wire. While some people are drawn into online scamming directly by friends and relatives, Facebook is, according to IJMs Wasuwongse, the most common entry point for people recruited on social media. Meta has known for years that its platforms host this kind of content. Back in 2019, the BBC exposed slave markets that were running on Instagram; in 2021, the Wall Street Journal reported, drawing on documents leaked by a whistleblower, that Meta had long struggled to rein in the problem but took meaningful action only after Apple threatened to pull Instagram from its app store. Today, years on, ads like the one that Gavesh responded to are still easy to find on Facebook if you know what to look for. Examples of fraudulent Facebook ads, shared by International Justice Mission. They are typically posted in job seekers groups and usually seem to be advertising legitimate jobs in areas like customer service. They offer attractive wages, especially for people with language skillsusually English or Chinese. The traffickers tend to finish the recruitment process on encrypted or private messaging apps. In our research, many experts said that Telegram, which is notorious for hosting terrorist content, child sexual abuse material, and other communication related to criminal activity, was particularly problematic. Many spoke with a combination of anger and resignation about its apparent lack of interest in working with them to address the problem; Mina Chiang, founder of Humanity Research Consultancy, an anti-trafficking organization, accuses the app of being very much complicit in human trafficking and proactively facilitating these scams. (Telegram did not respond to a request for comment.) But while Telegram users have the option of encrypting their messages end to end, making them almost impossible to monitor, social media companies are of course able to access users posts. And its here, at the beginning of the romance scam supply chain, where Big Tech could arguably make its most consequential intervention. Social media is monitored by a combination of human moderators and AI systems, which help flag users and contentads, posts, pagesthat break the law or violate the companies own policies. Dangerous content is easiest to police when it follows predictable patterns or is posted by users acting in distinctive and suspicious ways. They have financial resources. You can hire the most talented coding engineers in the world. Why cant you just find people who understand the issue properly? Anti-trafficking experts say the scam advertising tends to follow formulaic templates and use common language, and that they routinely report the ads to Meta and point out the markers they have identified. Their hope is that this information will be fed into the data sets that train the content moderation models. While individual ads may be taken down, even in big waveslast November, Meta said it had purged 2 million accounts connected to scamming syndicates over the previous yearexperts say that Facebook still continues to be used in recruiting. And new ads keep appearing. (In response to a request for comment, a Meta spokesperson shared links to policies about bans on content or advertisements that facilitate human trafficking, as well as company blog posts telling users how to protect themselves from romance scams and sharing details about the companys efforts to disrupt fraud on its platforms, one statingthat it is constantly rolling out new product features to help protect people on [its] apps from known scam tactics at scale. The spokesperson also said that WhatsApp has spam detection technology, and millions of accounts are banned per month.) Anti-trafficking experts we spoke with say that as recently as last fall, Meta was engaging with them and had told them it was ramping up its capabilities. But Chiang says there still isnt enough urgency from tech companies. Theres a question about speed. They might be able to say Thats the goal for the next two years. No. But thats not fast enough. We need it now, she says. They have financial resources. You can hire the most talented coding engineers in the world. Why cant you just find people who understand the issue properly? Part of the answer comes down to money, according to experts we spoke with. Scaling up content moderation and other processes that could cause users to be kicked off a platform requires not only technological staff but also legal and policy expertswhich not everyone sees as worth the cost. The vast majority of these companies are doing the minimum or less, says Tower of USIP. If not properly incentivized, either through regulatory action or through exposure by media or other forms of pressure often, these companies will underinvest in keeping their platforms safe. Getting set up Gaveshs new office turned out to be one of the most infamous scamming hubs in Southeast Asia: KK Park in Myanmars Myawaddy region. Satellite imagery shows it as a densely packed cluster of buildings, surrounded by fields. Most of it has been built since late 2019. Inside, it runs like a hybrid of a company campus and a prison. When Gavesh arrived, he handed over his phone and passport and was assigned to a dormitory and an employer. He was allowed his own phone back only for short periods, and his calls were monitored. Security was tight. He had to pass through airport-style metal detectors when he went in or out of the office. Black-uniformed personnel patrolled the buildings, while armed men in combat fatigues watched the perimeter fences from guard posts. On his first full day, he was put in front of a computer with just four documents on it, which he had to read over and overguides on how to approach strangers. On his second day, he learned to build fake profiles on social media and dating apps. The trick was to find real people on Instagram or Facebook who were physically attractive, posted often, and appeared to be wealthy and living a luxurious life, he says, and use their photos to build a new account: There are so many Instagram models that pretend they have a lot of money. After Gavesh was trafficked into Myanmar, he was taken to KK Park. Most of the compound has been built since late 2019.LUKE DUGGLEBY/REDUX Next, he was given a batch of iPhone 8smost people on his team used between eight and 10 devices eachloaded with local SIM cards and apps that spoofed their location so that they appeared to be in the US. Using male and female aliases, he set up dozens of accounts on Facebook, WhatsApp, Telegram, Instagram, and X and profiles on several dating platforms, though he cant remember exactly which ones. Different scamming operations teach different techniques for finding and reaching out to potential victims, several people who worked in the compounds tell us. Some people used direct approaches on dating apps, Facebook, Instagram, orfor those targeting Chinese victimsWeChat. One worker from Myanmar sent out mass messages on WhatsApp, pretending to have accidentally messaged a wrong number, in the hope of striking up a conversation. (Tencent, which owns WeChat, declined to comment.) Some scamming workers we spoke to were told to target white, middle-aged or older men in Western countries who seemed to be well off. Gavesh says he would pretend to be white men and women, using information found from Google to add verisimilitude to his claims of living in, say, Miami Beach. He would chat with the targets, trying to figure out from their jobs, spending habits, and ambitions whether theyd be worth investing time in. One South African woman, trafficked to Myanmar in 2022, says she was given a script and told to pose as an Asian woman living in Chicago. She was instructed to study her assigned city and learn quotidian details about life there. They kept on punishing people all the time for not knowing or for forgetting that theyre staying in Chicago, she says, or for forgetting whats Starbucks or whats [a] latte. Fake users have, of course, been a problem on social media platforms and dating sites for years. Some platforms, such as X, allow practically anyone to create accounts and even to have them verified for a fee. Others, including Facebook, have periodically conducted sweeps to get rid of fake accounts engaged in what Meta calls coordinated inauthentic behavior. (X did not respond to requests for comment.) But scam workers tell us they were advised on simple ways to circumvent detection mechanisms on social media. They were given basic training in how to avoid suspicious behavior such as adding too many contacts too quickly, which might trigger the company to review whether someones profile is authentic. The South African woman says she was shown how to manipulate the dates on a Facebook account to seem as if you opened the account in 2019 or whatever, making it easier to add friends. (Metas spam filtersmeant to reduce the spread of unwanted contentinclude limits on friend requests and bulk messaging.) Wang set up a Tinder profile with a picture of a dog and a bio that read, I am a dog. It passed through the platforms verification system without a hitch. Dating apps, whose users generally hope to meet other users in real life, have a particular need to make sure that people are who they say they are. But Match Group, the parent company of Tinder, ended its partnership with a company doing background checks in 2023. It now encourages users to verify their profile with a selfie and further ID checks, though insiders say these systems are often rudimentary. They just check a box and [do] what is legally required or what will make the media get off of [their] case, says one tech executive who has worked with multiple dating apps on safety systems, speaking on the condition of anonymity because they were not permitted to speak about their work with certain companies. Fangzhou Wang, an assistant professor at the University of Texas at Arlington who studies romance scams, ran a test: She set up a Tinder profile with a picture of a dog and a bio that read, I am a dog. It passed through the platforms verification system without a hitch. They are not providing enough security measures to filter out fraudulent profiles, Wang says. Everybody can create anything. Like recruitment ads, the scam profiles tend to follow patterns that should raise red flags. They use photos copied from existing users or made by artificial intelligence, and the accounts are sometimes set up using phone numbers generated by voice-over-internet-protocol services. Then theres the scammers behavior: They swipe too fast, or spend too much time logged in. A normal human doesnt spend eight hours on a dating app a day, the tech executive says. Whats more, scammers use the same language over and over again as they reach out to potential targets. The majority of them are using predesigned scripts, says Wang. It would be fairly easy for platforms to detect these signs and either stop accounts from being created or make the users go through further checks, experts tell us. Signals of some of these behaviors can potentially be embedded into a type of machine-learning algorithm, Wang says. She approached Tinder a few years ago with her research into the language that scammers use on the platforms, and offered to help build data sets for its moderation models. She says the company didnt reply. (In a statement, Yoel Roth, vice president of trust and safety at Match Group, said that the company invests in proactive tools, advanced detection systems and user education to help prevent harm. He wrote, We use proprietary AI-powered tools to help identify scammer messaging, and unlike many platforms, we moderate messages, which allows us to detect suspicious patterns early and act quickly, adding that the company has recently worked with Reality Defender, a provider of deepfake detection tools, to strengthen its ability to detect AI-generated content. A company spokesperson reported having no record of Wangs outreach but said that the company welcome[s] collaboration and [is] always open to reviewing research that can help strengthen user safety.) A recent investigation published in The Markup found that Match Group has long possessed the tools and resources to track sex offenders and other bad actors but has resisted efforts to roll out safety protocols for fear they might slow growth. This tension, between the desire to keep increasing the number of users and the need to ensure that these users and their online activity are authentic, is often behind safety issues on platforms. While no platform wants to be a haven for fraudsters, identity verification creates friction for users, which stops real people as well as impostors from signing up. And again, cracking down on platform violations costs money. According to Josh Kim, an economist who works in Big Tech, it would be costly for tech companies to build out the legal, policy, and operational teams for content moderation tools that could get users kicked off a platformand the expense is one companies may find hard to justify in the current business climate. The shift toward profitability means that you have to be very selective in where you invest the resources that you have, he says. My intuition here is that unless there are fines or pressure from governments or regulatory agencies or the public themselves, he adds, the current atmosphere in the tech ecosystem is to focus on building a product that is profitable and grows fast, and things that dont contribute to those two points are probably being deprioritized. Getting onlineand staying in line At work, Gavesh wore a blue tag, marking him as belonging to the lowest rank of workers. On top of us are the ones who are wearing the yellow tagsthey call themselves HR or translators, or office guys, he says. Red tags are team leaders, managers And then moving from that, they have black and ash tags. Those are the ones running the office. Most of the latter were Chinese, Gavesh says, as were the really big bosses, who didnt wear tags at all. Within this hierarchy operated a system of incentives and punishments. Workers who followed orders and proved successful at scamming could rise through the ranks to training or supervisory positions, and gain access to perks like restaurants and nightclubs. Those who failed to meet the targets or broke the rules faced violence and humiliation. Gavesh says he was once beaten because he broke an unwritten rule that it was forbidden to cross your legs at work. Yawning was banned, and bathroom breaks were limited to two minutes at a time. KATHERINE LAM Beatings were usually conducted in the open, though the most severe punishments at Gaveshs company happened in a room called the water jail. One day a coworker was there alongside the others, and the next day he was not, Gavesh recalls. When the colleague was brought back to the office, he had been so badly beaten he couldnt walk or speak. They took him to the front, and they said: If you do not listen to us, this is what will happen to you. Gavesh was desperate to leave but felt there was no chance of escaping. The armed guards seemed ready to shoot, and there were rumors in the compound that some people who jumped the fence had been found drowned in the river. This kind of physical and psychological abuse is routine across the industry. Gavesh and others we spoke to describe working 12 hours or more a day, without days off. They faced strict quotas for the number of scam targets they had to have on the hook. If they failed to reach them, they were punished. The UN has documented cases of torture, arbitrary detention, and sexual violence in the compounds. We h

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. China built hundreds of AI data centers to catch the AI boom. Now many stand unused. Just months ago, Chinas boom in data center construction was at its height, fueled by both government and private investors. Renting out GPUs to companies that need them for training AI models was once seen as a sure bet. But with the rise of DeepSeek and a sudden change in the economics around AI, the industry is faltering. Prices for GPUs are falling and many newly built facilities are now sitting empty. Read the full story to find out why. Caiwei Chen OpenAIs new image generator aims to be practical enough for designers and advertisers Whats new? OpenAI has released a new image generator thats designed less for typical surrealist AI art and more for highly controllable and practical creation of visualsa sign that OpenAI thinks its tools are ready for use in fields like advertising and graphic design. Why it matters: While most AI models have been great at creating fantastical images or realistic deepfakes, theyve been terrible at identifying certain objects correctly and putting them in their proper place. OpenAIs new model makes progress on technical issues that have plagued AI image generators for years. But in entering this domain, OpenAI has two paths, both difficult. Read the full story. The AI Hype Index: DeepSeek mania, Israels spying tool, and cheating at chess Separating AI reality from hyped-up fiction isnt always easy. Thats why weve created the AI Hype Indexa simple, at-a-glance summary of everything you need to know about the state of the industry. Take a look at the full index here. The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 The Trump administration has barred 80 companies from buying US tech The list of primarily Chinese firms is forbidden from buying American chips. (NYT $)+ The list included a server maker that buys chips from Nvidia. (WSJ $)+ China disputed claims the firms were seeking knowledge for military purposes. (AP News)2 A DOGE staffer provided tech support to a cybercrime ring And bragged about trafficking in stolen data and cyberstalking an FBI agent. (Reuters)+ Elon Musk could use DOGEs cuts to steer contracts towards his own firms. (The Guardian)+ Can AI help DOGE slash government budgets? Its complex. (MIT Technology Review) 3 The US government has hired a vaccine skeptic to conduct a major vaccine study The long-discredited David Geier will oversee analysis of whether jabs cause autism. (WP $)+ The White House appears to be targeting mRNA vaccines. (FT $)+ Why childhood vaccines are a public health success story. (MIT Technology Review)4 Microsoft has unveiled two deep reasoning Copilot AI agents The two agents, called Researcher and Analyst, are designed to do just that. (The Verge)+ How ChatGPT search paves the way for AI agents. (MIT Technology Review)5 Inside the rise of Chinese hacking The cyber threat posed by the country is increasingly sophisticatedand aggressive. (Economist $)6 Google has instructed workers to remove DEI terms from their workThe company has offered up alternative language to use in its place.(The Information $) 7 Synthesia is offering shares to reward human actors for its AI avatarsThe compensation scheme is the first of its kind. (FT $) + Synthesias hyperrealistic deepfakes will soon have full bodies. (MIT Technology Review)8 Chinas RedNote is working to keep its influx of TikTok refugeesTo do so, itll need to expand its user base outside the Chinese diaspora. (Rest of World) 9 This operating system is designed to keep running during civilizations collapse Collapse OS is designed to give us access to lost knowledge in case of disaster. (Wired $)10 No one really knows how long people live Longevity research is bogged down in bad record-keeping. (NY Mag $)+ The quest to legitimize longevity medicine. (MIT Technology Review) Quote of the day There are so many great reasons to be on Signal. Now including the opportunity for the vice president of the United States of America to randomly add you to a group chat for coordination of sensitive military operations. Moxie Marlinspike, founder of secure messaging platform Signal, pokes fun at the fallout surrounding US officials accidentally adding a journalist to a private military group chat in a post on X. The big story Longevity enthusiasts want to create their own independent state. Theyre eyeing Rhode Island. May 2023 Jessica Hamzelou I recently traveled to Montenegro for a gathering of longevity enthusiasts. All the attendees were super friendly, and the sense of optimism was palpable. Theyre all confident well be able to find a way to slow or reverse agingand they have a bold plan to speed up progress. Around 780 of these people have created a pop-up city that hopes to circumvent the traditional process of clinical trials. They want to create an independent state where like-minded innovators can work together in an all-new jurisdiction that gives them free rein to self-experiment with unproven drugs. Welcome to Zuzalu. Read the full story. We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + Good newsit turns out that fungi are actually pretty good at saving imperiled plants.+ Ever wondered what ancient Egyptian mummy remains smell like? These intrepid scientists found out.+ Kudos to this terrible artist, who is a surprise smash hit.+ Check out this handy guide to walking the path of everyday enlightenment.

Separating AI reality from hyped-up fiction isnt always easy. Thats why weve created the AI Hype Indexa simple, at-a-glance summary of everything you need to know about the state of the industry. While AI models are certainly capable of creating interesting and sometimes entertaining material, their output isnt necessarily useful. Google DeepMind is hoping that its new robotics model could make machines more receptive to verbal commands, paving the way for us to simply speak orders to them aloud. Elsewhere, the Chinese startup Monica has created Manus, which it claims is the very first general AI agent to complete truly useful tasks. And burnt-out coders are allowing AI to take the wheel entirely in a new practice dubbed vibe coding.

A year or so ago, Xiao Li was seeing floods of Nvidia chip deals on WeChat. A real estate contractor turned data center project manager, he had pivoted to AI infrastructure in 2023, drawn by the promise of Chinas AI craze. At that time, traders in his circle bragged about securing shipments of high-performing Nvidia GPUs that were subject to US export restrictions. Many were smuggled through overseas channels to Shenzhen. At the height of the demand, a single Nvidia H100 chip, a kind that is essential to training AI models, could sell for up to 200,000 yuan ($28,000) on the black market. Now, his WeChat feed and industry group chats tell a different story. Traders are more discreet in their dealings, and prices have come back down to earth. Meanwhile, two data center projects Li is familiar with are struggling to secure further funding from investors who anticipate poor returns, forcing project leads to sell off surplus GPUs. It seems like everyone is selling, but few are buying, he says. Just months ago, a boom in data center construction was at its height, fueled by both government and private investors. However, many newly built facilities are now sitting empty. According to people on the ground who spoke to MIT Technology Reviewincluding contractors, an executive at a GPU server company, and project managersmost of the companies running these data centers are struggling to stay afloat. The local Chinese outlets Jiazi Guangnian and 36Kr report that up to 80% of Chinas newly built computing resources remain unused. Renting out GPUs to companies that need them for training AI modelsthe main business model for the new wave of data centerswas once seen as a sure bet. But with the rise of DeepSeek and a sudden change in the economics around AI, the industry is faltering. The growing pain Chinas AI industry is going through is largely a result of inexperienced playerscorporations and local governmentsjumping on the hype train, building facilities that arent optimal for todays need, says Jimmy Goodrich, senior advisor for technology at the RAND Corporation. The upshot is that projects are failing, energy is being wasted, and data centers have become distressed assets whose investors are keen to unload them at below-market rates. The situation may eventually prompt government intervention, he says: The Chinese government is likely to step in, take over, and hand them off to more capable operators. A chaotic building boom When ChatGPT exploded onto the scene in late 2022, the response in China was swift. The central government designated AI infrastructure as a national priority, urging local governments to accelerate the development of so-called smart computing centersa term coined to describe AI-focused data centers. In 2023 and 2024, over 500 new data center projects were announced everywhere from Inner Mongolia to Guangdong, according to KZ Consulting, a market research firm. According to the China Communications Industry Association Data Center Committee, a state-affiliated industry association, at least 150 of the newly built data centers were finished and running by the end of 2024. State-owned enterprises, publicly traded firms, and state-affiliated funds lined up to invest in them, hoping to position themselves as AI front-runners. Local governments heavily promoted them in the hope theyd stimulate the economy and establish their region as a key AI hub. However, as these costly construction projects continue, the Chinese frenzy over large language models is losing momentum. In 2024 alone, over 144 companies registered with the Cyberspace Administration of Chinathe country's central internet regulatorto develop their own LLMs. Yet according to the Economic Observer, a Chinese publication, only about 10% of those companies were still actively investing in large-scale model training by the end of the year. Chinas political system is highly centralized, with local government officials typically moving up the ranks through regional appointments. As a result, many local leaders prioritize short-term economic projects that demonstrate quick resultsoften to gain favor with higher-upsrather than long-term development. Large, high-profile infrastructure projects have long been a tool for local officials to boost their political careers. The post-pandemic economic downturn only intensified this dynamic. With Chinas real estate sectoronce the backbone of local economiesslumping for the first time in decades, officials scrambled to find alternative growth drivers. In the meantime, the countrys once high-flying internet industry was also entering a period of stagnation. In this vacuum, AI infrastructure became the new stimulus of choice. AI felt like a shot of adrenaline, says Li. A lot of money that used to flow into real estate is now going into AI data centers. By 2023, major corporationsmany of them with little prior experience in AIbegan partnering with local governments to capitalize on the trend. Some saw AI infrastructure as a way to justify business expansion or boost stock prices, says Fang Cunbao, a data center project manager based in Beijing. Among them were companies like Lotus, an MSG manufacturer, and Jinlun Technology, a textile firmhardly the names one would associate with cutting-edge AI technology. This gold-rush approach meant that the push to build AI data centers was largely driven from the top down, often with little regard for actual demand or technical feasibility, say Fang, Li, and multiple on-the-ground sources, who asked to speak anonymously for fear of political repercussions. Many projects were led by executives and investors with limited expertise in AI infrastructure, they say. In the rush to keep up, many were constructed hastily and fell short of industry standards. Putting all these large clusters of chips together is a very difficult exercise, and there are very few companies or individuals who know how to do it at scale, says Goodrich. This is all really state-of-the-art computer engineering. Id be surprised if most of these smaller players know how to do it. A lot of the freshly built data centers are quickly strung together and dont offer the stability that a company like DeepSeek would want. To make matters worse, project leaders often relied on middlemen and brokerssome of whom exaggerated demand forecasts or manipulated procurement processes to pocket government subsidies, sources say. By the end of 2024, the excitement that once surrounded Chinas data center boom was curdling into disappointment. The reason is simple: GPU rental is no longer a particularly lucrative business. The DeepSeek reckoning The business model of data centers is in theory straightforward: They make money by renting out GPU clusters to companies that need computing capacity for AI training. In reality, however, securing clients is proving difficult. Only a few top tech companies in China are now drawing heavily on computing power to train their AI models. Many smaller players have been giving up on pretraining their models or otherwise shifting their strategy since the rise of DeepSeek, which broke the internet with R1, its open-source reasoning model that matches the performance of ChatGPT o1 but was built at a fraction of its cost. DeepSeek is a moment of reckoning for the Chinese AI industry. The burning question shifted from Who can make the best large language model? to Who can use them better? says Hangcheng Cao, an assistant professor of information systems at Emory University. The rise of reasoning models like DeepSeeks R1 and OpenAIs ChatGPT o1 and o3 has also changed what businesses want from a data center. With this technology, most of the computing needs come from conducting step-by-step logical deductions in response to users queries, not from the process of training and creating the model in the first place. This reasoning process often yields better results but takes significantly more time. As a result, hardware with low latency (the time it takes for data to pass from one point on a network to another) is paramount. Data centers need to be located near major tech hubs to minimize transmission delays and ensure access to highly skilled operations and maintenance staff. This change means many data centers built in central, western, and rural Chinawhere electricity and land are cheaperare losing their allure to AI companies. In Zhengzhou, a city in Lis home province of Henan, a newly built data center is even distributing free computing vouchers to local tech firms but still struggles to attract clients. Additionally, a lot of the new data centers that have sprung up in recent years were optimized for pretraining workloadslarge, sustained computations run on massive data setsrather than for inference, the process of running trained reasoning models to respond to user inputs in real time. Inference-friendly hardware differs from whats traditionally used for large-scale AI training. GPUs like Nvidia H100 and A100 are designed for massive data processing, prioritizing speed and memory capacity. But as AI moves toward real-time reasoning, the industry seeks chips that are more efficient, responsive, and cost-effective. Even a minor miscalculation in infrastructure needs can render a data center suboptimal for the tasks clients require. In these circumstances, the GPU rental price has dropped to an all-time low. A recent report from the Chinese media outlet Zhineng Yongxian said that an Nvidia H100 server configured with eight GPUs now rents for 75,000 yuan per month, down from highs of around 180,000. Some data centers would rather leave their facilities sitting empty than run the risk of losing even more money because they are so costly to run, says Fan: The revenue from having a tiny part of the data center running simply wouldnt cover the electricity and maintenance cost. Its paradoxicalChina faces the highest acquisition costs for Nvidia chips, yet GPU leasing prices are extraordinarily low, Li says. Theres an oversupply of computational power, especially in central and west China, but at the same time, theres a shortage of cutting-edge chips. However, not all brokers were looking to make money from data centers in the first place. Instead, many were interested in gaming government benefits all along. Some operators exploit the sector for subsidized green electricity, obtaining permits to generate and sell power, according to Fang and some Chinese media reports. Instead of using the energy for AI workloads, they resell it back to the grid at a premium. In other cases, companies acquire land for data center development to qualify for state-backed loans and credits, leaving facilities unused while still benefiting from state funding, according to the local media outlet Jiazi Guangnian. Towards the end of 2024, no clear-headed contractor and broker in the market would still go into the business expecting direct profitability, says Fang. Everyone I met is leveraging the data center deal for something else the government could offer. A necessary evil Despite the underutilization of data centers, Chinas central government is still throwing its weight behind a push for AI infrastructure. In early 2025, it convened an AI industry symposium, emphasizing the importance of self-reliance in this technology. Major Chinese tech companies are taking note, making investments aligning with this national priority. Alibaba Group announced plans to invest over $50 billion in cloud computing and AI hardware infrastructure over the next three years, while ByteDance plans to invest around $20 billion in GPUs and data centers. In the meantime, companies in the US are doing likewise. Major tech firms including OpenAI, Softbank, and Oracle have teamed up to commit to the Stargate initiative, which plans to invest up to $500 billion over the next four years to build advanced data centers and computing infrastructure. Given the AI competition between the two countries, experts say that China is unlikely to scale back its efforts. If generative AI is going to be the killer technology, infrastructure is going to be the determinant of success, says Goodrich, the tech policy advisor to RAND. The Chinese central government will likely see [underused data centers] as a necessary evil to develop an important capability, a growing pain of sorts. You have the failed projects and distressed assets, and the state will consolidate and clean it up. They see the end, not the means, Goodrich says. Demand remains strong for Nvidia chips, and especially the H20 chip, which was custom-designed for the Chinese market. One industry source, who requested not to be identified under his company policy, confirmed that the H20, a lighter, faster model optimized for AI inference, is currently the most popular Nvidia chip, followed by the H100, which continues to flow steadily into China even though sales are officially restricted by US sanctions. Some of the new demand is driven by companies deploying their own versions of DeepSeeks open-source models. For now, many data centers in China sit in limbobuilt for a future that has yet to arrive. Whether they will find a second life remains uncertain. For Fang Cunbao, DeepSeeks success has become a moment of reckoning, casting doubt on the assumption that an endless expansion of AI infrastructure guarantees progress. Thats just a myth, he now realizes. At the start of this year, Fang decided to quit the data center industry altogether. The market is too chaotic. The early adopters profited, but now its just people chasing policy loopholes, he says. Hes decided to go into AI education next. What stands between now and a future where AI is actually everywhere, he says, is not infrastructure anymore, but solid plans to deploy the technology.

OpenAI has released a new image generator thats designed less for typical surrealist AI art and more for highly controllable and practical creation of visualsa sign that OpenAI thinks its tools are ready for use in fields like advertising and graphic design. The image generator, which is now part of the companys GPT-4o model, was promised by OpenAI last May but wasnt released. Requests for generated images on ChatGPT were filled by an older image generator called DALL-E. OpenAI has been tweaking the new model since then and will now release it over the coming weeks to all tiers of users starting today, replacing the older one. The new model makes progress on technical issues that have plagued AI image generators for years. While most have been great at creating fantastical images or realistic deepfakes, theyve been terrible at something called binding, which refers to the ability to identify certain objects correctly and put them in their proper place (like a sign that says hot dogs properly placed above a food cart, not somewhere else in the image). It was only a few years ago that models started to succeed at things like Put the red cube on top of the blue cube, a feature that is essential for any creative professional use of AI. Generators also struggle with text generation, typically creating distorted jumbles of letter shapes that look more like captchas than readable text. Example images from OpenAI show progress here. The model is able to generate 12 discrete graphics within a single imagelike a cat emoji or a lightning boltand place them in proper order. Another shows four cocktails accompanied by recipe cards with accurate, legible text. More images show comic strips with text bubbles, mock advertisements, and instructional diagrams. The model also allows you to upload images to be modified, and it will be available in the video generator Sora as well as in GPT-4o. Its a new tool for communication, says Gabe Goh, the lead designer on the generator at OpenAI. Kenji Hata, a researcher at OpenAI who also worked on the tool, puts it a different way: I think the whole idea is that were going away from, like, beautiful art. It can still do that, he clarifies, but it will do more useful things too. You can actually make images work for you, he says, and not just just look at them. Its a clear sign that OpenAI is positioning the tool to be used more by creative professionals: think graphic designers, ad agencies, social media managers, or illustrators. But in entering this domain, OpenAI has two paths, both difficult. One, it can target the skilled professionals who have long used programs like Adobe Photoshop, which is also investing heavily in AI tools that can fill images with generative AI. Adobe really has a stranglehold on this market, and theyre moving fast enough that I dont know how compelling it is for people to switch, says David Raskino, the cofounder and chief technical officer of Irreverent Labs, which works on AI video generation. The second option is to target casual designers who have flocked to tools like Canva (which has also been investing in AI). This is an audience that may not have ever needed technically demanding software like Photoshop but would use more casual design tools to create visuals. To succeed here, OpenAI would have to lure people away from platforms built for design in hopes that the speed and quality of its own image generator would make the switch worth it (at least for part of the design process). Its also possible the tool will simply be used as many image generators are now: to create quick visuals that are good enough to accompany social media posts. But with OpenAI planning massive investments, including participation in the $500 billion Stargate project to build new data centers at unprecedented scale, its hard to imagine that the image generator wont play some ambitious moneymaking role. Regardless, the fact that OpenAIs new image generator has pushed through notable technical hurdles has raised the bar for other AI companies. Clearing those hurdles likely required lots of very specific data, Raskino says, like millions of images in which text is properly displayed at lots of different angles and orientations. Now competing image generators will have to match those achievements to keep up. The pace of innovation should increase here, Raskino says.

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. Ethically sourced spare human bodies could revolutionize medicine Many challenges in medicine stem, in large part, from a common root cause: a severe shortage of ethically-sourced human bodies. There might be a way to get out of this moral and scientific deadlock. Recent advances in biotechnology now provide a pathway to producing living human bodies without the neural components that allow us to think, be aware, or feel pain.Many will find this possibility disturbing, but if researchers and policymakers can find a way to pull these technologies together, we may one day be able to create spare bodies, both human and nonhuman. These could revolutionize medical research and drug development, greatly reducing the need for animal testing, rescuing many people from organ transplant lists, and allowing us to produce more effective drugs and treatments. All without crossing most peoples ethical lines. Read the full story.Why the world is looking to ditch US AI models Eileen Guo A few weeks ago, when I was at the digital rights conference RightsCon in Taiwan, I watched in real time as civil society organizations from around the world, including the US, grappled with the loss of one of the biggest funders of global digital rights work: the United States government. Some policymakers and business leadersin Europe, in particularare reconsidering their reliance on US-based tech and asking whether they can quickly spin up better, homegrown alternatives. This is particularly true for AI. Read the full story. This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. How to delete your 23andMe data Consumer DNA testing company 23andMe has filed for bankruptcy protection, following months of speculation around CEO Anne Wojcickis plans to take the firm private. The news means that 23andMeand the genetic data of millions of its customerscould soon be put up for sale. But although customers worried about the security of their DNA data can request its deletion, truly scrubbing your information from the companys archives is easier said than done. Read the full story.Rhiannon Williams The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 US security leaders accidentally added a journalist to a secret Signal chat The group used the unapproved platform to discuss classified military strikes in Yemen. (The Atlantic $)+ It raises questions over how the US government is handling sensitive information. (Vox)+ The Trump administration has embraced the encrypted messaging app. (WP $) 2 Donald Trumps H-1B visa crackdown could seriously harm US tech firmsAmazon is likely to be hit particularly hard. (Rest of World) + US visa and green-card holders are being detained and deported. (NY Mag $)+ Tariffs, DOGE and scams are weighing heavily on the tech industry. (Insider $)+ America relies heavily on skilled overseas workers. (The Conversation)3 DeepSeeks runaway success is shaking up Chinas AI startups Theyre overhauling their business models in an effort to keep up. (FT $)+ The AI development gap between China and the US is narrowing. (Reuters)+ How DeepSeek ripped up the AI playbookand why everyones going to follow its lead. (MIT Technology Review)4 AI companies dont want to be regulated anymore Emboldened by the Trump administration, the industrys biggest firms are lobbying for fewer rules. (NYT $)5 Colorado is experimenting with psychedelic mushroomsIt plans to administer them in healing centers across the state. (Undark) + Job titles of the future: Pharmaceutical-grade mushroom grower. (MIT Technology Review)6 Tesla sales are plummeting in EuropeAs customers turn to its Chinese rival BYD. (The Guardian) + Elon Musks companies are under increasing pressure from their rivals. (Economist $)+ BYD was one of our 2024 Climate Tech Companies to Watch. (MIT Technology Review)7 This Indian city relies on the wind to stay coolPalava City is a living testbed of technological innovation. (WP $) + No power, no fans, no AC: The villagers fighting to survive Indias deadly heatwaves. (MIT Technology Review)8 Filming your online routine is not for the faint of heart Absurd clips are doing the rounds on social media yet again. (NY Mag $)9 Floating wood could help to refreeze the Arctic By helping to seed the formation of new ice. (New Scientist $)+ Inside a new quest to save the doomsday glacier. (MIT Technology Review) 10 Silicon Valley workers are ditching dating apps Instead, theyre attending carefully vetted dating meetups IRL. (Wired $)Quote of the day The path to saving TikTok should run through Capitol Hill. Three Democratic senators urge Donald Trump to work with Congress to save TikTok from shutting down in the US, the Verge reports. The big story How AI is changing gymnastics judging January 2024 The 2023 World Championships last October marked the first time an AI judging system was used on every apparatus in a gymnastics competition. There are obvious upsides to using this kind of technology: AI could help take the guesswork out of the judging technicalities. It could even help to eliminate biases, making the sport both more fair and more transparent. At the same time, others fear AI judging will take away something that makes gymnastics special. Gymnastics is a subjective sport, like diving or dressage, and technology could eliminate the judges role in crafting a narrative. For better or worse, AI has officially infiltrated the world of gymnastics. The question now is whether it really makes it fairer. Read the full story. Jessica Taylor Price We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + These plants are quite possibly math geniuses.+ Inside the weird and wonderful world of animal art.+ Get me on a (sustainable) trip to the Cook Islands immediately.+ Its officially cherry blossom season around the world!

This story was originally published in October 2024. In March 2025, 23andMe filed for bankruptcy and announced its plans to facilitate a sale process to maximize the value of its business. MIT Technology Review'sHow Toseries helps you get things done. Things arent looking good for 23andMe. The consumer DNA testing company recently parted ways with all its board members but CEO Anne Wojcicki over her plans to take the company private. Its also still dealing with the fallout of a major security breach last October, which saw hackers access the personal data of around 5.5 million customers. 23andMes business is built on taking saliva samples from its customers. The DNA from those samples is processed and analyzed in its labs to produce personalized genetic reports detailing a users unique health and ancestry. The uncertainty swirling around the companys future and potential new ownership has prompted privacy campaigners to urge users to delete their data. Its not just you. If anyone in your family gave their DNA to 23&Me, for all of your sakes, close your/their account now, Meredith Whittaker, president of the encrypted messaging platform Signal, posted on X after the boards resignation. "Customers should consider current threats to their privacy as well as threats that may exist in the futuresome of which may be magnified if 23AndMe were sold to a new owner," says Jason Kelley, activism director at the Electronic Frontier Foundation. "23AndMe has protections around this much of this. But a potential sale could put your data in the hands of a far less scrupulous company." A spokesperson for 23andMe said that the company has strong customer privacy protections in place, and does not share customer data with third parties without customers consent. "Our research program is opt-in, requiring customers to go through a separate, informed consent process before joining," they say. We are committed to protecting customer data and are consistently focused on maintaining the privacy of our customers. That will not change. Why deleting your account comes with a caveat Deleting your data from 23andMe is permanent and cannot be reversed. But some of that data will be retained to comply with the companys legal obligations, according to its privacy statement. That means 23andMe and its third-party genotyping laboratory will hang onto some of your genetic information, plus your date of birth and sexalongside data linked to your account deletion request, including your email address and deletion request identifier. When MIT Technology Review asked 23andMe about the nature of the genetic information it retains, it referred us to its privacy policy but didn't provide any other details. Any information youve previously provided and consented to being used in 23andMe research projects also cannot be removed from ongoing or completed studies, although it will not be used in any future ones. Beyond the laboratories that process the saliva samples, the company does not share customer information with anyone else unless the user has given permission for it to do so, the spokesperson says, including employers, insurance companies, law enforcement agencies, or any public databases. We treat law enforcement inquiries, such as a valid subpoena or court order, with the utmost seriousness. We use all legal measures to resist any and all requests in order to protect our customer's privacy, the spokesperson says. To date, we have successfully challenged these requests and have not released any information to law enforcement. For those who still want their data deleted, heres how you go about it. How to delete your data from 23andMe Log into your account and navigate to Settings. Under Settings, scroll to the section titled 23andMe data. Select View. You may be asked to enter your date of birth for extra security. In the next section, youll be asked which, if any, personal data youd like to download from the company (onto a personal, not public, computer). Once youre finished, scroll to the bottom and select Permanently delete data. You should then receive an email from 23andMe detailing its account deletion policy and requesting that you confirm your request. Once you confirm youd like your data to be deleted, the deletion will begin automatically and youll immediately lose access to your account. What about your genetic sample? When you set up your 23andMe account, youre given the option either to have your saliva sample securely destroyed or to have it stored for future testing. If youve previously opted to store your sample but now want to delete your 23andMe account, the company says, it will destroy the sample for you as part of the account deletion process. What if you want to keep your genetic data, just not on 23andMe? Even if you want your data taken off 23AndMe, there are reasons why you might still want to have it hosted on other DNA sitesfor genealogical research, for example. And some people like the idea of having their DNA results stored on more than one database in case something happens to any one company. This is where downloading your data comes into play. FamilyTreeDNA, MyHeritage, GEDmatch, and Living DNA are among the DNA testing companies that allow you to upload existing DNA results from other companies, although Ancestry and 23andMe dont accept uploads. How to download your raw genetic data Navigate directly to you.23andme.com/tools/data/. Click on your profile name on the top right-hand corner. Then select Resources from the menu. Select Browse raw genotyping data and then Download. Visit Account settings and click on View under 23andMe data. Enter your date of birth for security purposes. Tick the box indicating that you understand the limitations and risks associated with uploading your information to third-party sites and press Submit request. 23andMe warns its users that uploading their data to other services could put genetic data privacy at risk. For example, bad actors could use someone elses DNA data to create fake genetic profiles. They could use these profiles to match with a relative and access personal identifying information and specific DNA variantssuch as information about any disease risk variants you might carry, the spokesperson says, adding: This is one reason why we dont support uploading DNA to 23andMe at this time. Update: This article has been updated to reflect that when asked about the nature of the genetic information it retains, 23andMe referred us to its privacy policy but didn't provide any other details.

Why do we hear about medical breakthroughs in mice, but rarely see them translate into cures for human disease? Why do so few drugs that enter clinical trials receive regulatory approval? And why is the waiting list for organ transplantation so long? These challenges stem in large part from a common root cause: a severe shortage of ethically-sourced human bodies. It may be disturbing to characterize human bodies in such commodifying terms, but the unavoidable reality is that human biological materials are an essential commodity in medicine, and persistent shortages of these materials create a major bottleneck to progress. This imbalance between supply and demand is the underlying cause of the organ shortage crisis, with more than 100,000 patients currently waiting for a solid organ transplant in the US alone. It also forces us to rely heavily on animals in medical research, a practice that cant replicate major aspects of human physiology and necessitates the infliction of harm to sentient creatures. In addition, the safety and efficacy of any experimental drug must still be confirmed in clinical trials on living human bodies. These costly trials risk harm to patients, can take a decade or longer to complete, and make it through to approval less than 15% of the time. There might be a way to get out of this moral and scientific deadlock. Recent advances in biotechnology now provide a pathway to producing living human bodies without the neural components that allow us to think, be aware, or feel pain. Many will find this possibility disturbing, but if researchers and policymakers can find a way to pull these technologies together, we may one day be able to create spare bodies, both human and nonhuman. These could revolutionize medical research and drug development, greatly reducing the need for animal testing, rescuing many people from organ transplant lists, and allowing us to produce more effective drugs and treatments. All without crossing most peoples ethical lines. Bringing technologies together Although it may seem like science fiction, recent technological progress has pushed this concept into the realm of plausibility. Pluripotent stem cells, one of the earliest cell types to form during development, can give rise to every type of cell in the adult body. Recently, researchers have used these stem cells to create structures that seem to mimic the early development of actual human embryos. At the same time, artificial uterus technology is rapidly advancing, and other pathways may be opening to allow for the development of fetuses outside of the body. By integrating these different technologies and using established genetic techniques to inhibit brain development, it is possible to envision the creation of bodyoids a potentially unlimited source of human bodies, developed entirely outside of a human body from stem cells, that lack sentience or the ability to feel pain. There are still many technical roadblocks to achieving this vision, but we have reason to expect that bodyoids couldradically transform biomedical research by addressing critical limitations in the current models of research, drug development and medicine. Among many other benefits, theywould offer an almost unlimited source of organs, tissues and cells for use in transplantation. It could even be possible to generate organs directly from a patient's own cells, essentially cloning their biological material to ensure that transplanted tissues are a perfect immunological match to a patient and thus eliminating the need for lifelong immunosuppression. Bodyoids developed from a patient's cells could also allow for personalized screening of drugs, allowing physicians to directly assess the effect of different interventions in a biological model that accurately reflects a patient's own personal genetics and physiology. We can even envision using animal bodyoidsin agriculture, as a substitute for the use of sentient animal species. Of course, exciting possibilities are not certainties. We do not know whether the embryo models recently created from stem cells could give rise to living people or, thus far, even to living mice. We do not know when, or whether, an effective technique will be found for successfully gestating human bodies entirely outside a person. We cannot be sure whether such bodyoids can survive without ever having developed brains or the parts of brains associated with consciousness, or whether they would still serve as accurate models for living people without those brain functions. Even if it all works, it may not be practical or economical to grow bodyoids, possibly for many years, until they can be mature enough to be useful for our ends. Each of these questions will require substantial research and time. But we believe this idea is now plausible enough to justify discussing both the technical feasibility and ethical implications. Ethical considerations and societal implications Bodyoids could address many ethical problems in modern medicine, offering ways to avoid unnecessary pain and suffering. For example, they could offer an ethical alternative to the way we currently use nonhuman animals for research and food, providing meat or other products with no animal suffering or awareness. But when we come to human bodyoids, the issues become harder. Many will find the concept grotesque or appalling. And for good reason. We have an innate respect for human life in all its forms. We do not allow broad research on people who no longer have consciousness or, in some cases, never had it. At the same time, we know much can be gained from studying the human body. We learn much from the bodies of the dead, which these days are used for teaching and research only with consent. In laboratories, we study cells and tissues that were taken, with consent, from the bodies of the dead and the living. Recently we have even begun using for experiments the animated cadavers of people who have been declared legally dead, who have lost all brain function but whose other organs continue to function with mechanical assistance. Genetically modified pig kidneys have been connected to, or transplanted into, these legally dead but physiologically active cadavers to help researchers determine whether they would work in living people. In all these cases, nothing was, legally, a living human being at the time it was used for research. Human bodyoids would also fall into that category. But there are still a number of issues worth considering. The first is consent: The cells used to make bodyoids would have to come from someone, and wed have to make sure that this someone consented to this particular, likely controversial, use. But perhaps the deepest issue is that bodyoids might diminish the human status of real people who lack consciousness or sentience. Thus far, we have held to a standard that requires us to treat all humans born alive as people, entitled to life and respect. Would bodyoidscreated without pregnancy, parental hopes, or indeed parentsblur that line? Or would we consider a bodyoid a human being, entitled to the same respect? If so, whyjust because it looks like us? A sufficiently detailed mannequin can meet that test. Because it looks like us and is alive? Because it is alive and has our DNA? These are questions that will require careful thought. A call to action Until recently, the idea of making something like a bodyoid would have been relegated to the realms of science fiction and philosophical speculation. But now it is at least plausibleand possibly revolutionary. It is time for it to be explored. The potential benefitsfor both human patients and sentient animal speciesare great. Governments, companies and private foundations should start thinking about bodyoids as a possible path for investment.There is no need to start with humanswe can begin exploring the feasibility of this approach with rodents or other research animals. As we proceed, the ethical and social issues are at least as important as the scientific ones. Just because something can be done does not mean it should be done. Even if it looks possible, determining whether we should make bodyoids, nonhuman or human, will require considerable thought, discussion, and debate. Some of that will be by scientists, ethicists, and others with special interest or knowledge. But ultimately, the decisions will be made by societies and governments. The time to start those discussions is now, when a scientific pathway seems clear enough for us to avoid pure speculation but before the world is presented with a troubling surprise. The announcement of the birth of Dolly the cloned sheep back in the 1990s launched a hysterical reaction, complete with speculation about armies of cloned warrior slaves. Good decisions require more preparation. The path toward realizing the potential of bodyoids will not be without challenges; indeed, it may never be possible to get there, or even if it is possible, the path may never be taken. Caution is warranted, but so is bold vision; the opportunity is too important to ignore. Carsten T. Charlesworth is a postdoctoral fellow at the Institute of Stem Cell Biology and Regenerative Medicine (ISCBRM) at Stanford University. Henry T. Greely is the Deane F. and Kate Edelman Johnson Professor of Law and director of the Center for Law and the Biosciences at Stanford University. Hiromitsu Nakauchi is a professor of genetics and an ISCBRM faculty member at Stanford University and a distinguished university professor at the Institute of Science Tokyo.

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first,sign up here. This weeks edition of The Algorithm is brought to you not by your usual host, James ODonnell, but Eileen Guo, an investigative reporter at MIT Technology Review. A few weeks ago, when I was at the digital rights conference RightsCon in Taiwan, I watched in real time as civil society organizations from around the world, including the US, grappled with the loss of one of the biggest funders of global digital rights work: the United States government. As I wrote in my dispatch, the Trump administration's shocking, rapid gutting of the US government (and its push into what some prominent political scientists call competitive authoritarianism) also affects the operations and policies of American tech companiesmany of which, of course, have users far beyond US borders. People at RightsCon said they were already seeing changes in these companies willingness to engage with and invest in communities that have smaller user basesespecially non-English-speaking ones. As a result, some policymakers and business leadersin Europe, in particularare reconsidering their reliance on US-based tech and asking whether they can quickly spin up better, homegrown alternatives. This is particularly true for AI. One of the clearest examples of this is in social media. Yasmin Curzi, a Brazilian law professor who researches domestic tech policy, put it to me this way: Since Trumps second administration, we cannot count on [American social media platforms] to do even the bare minimum anymore. Social media content moderation systemswhich already use automation and are also experimenting with deploying large language models to flag problematic postsare failing to detect gender-based violence in places as varied as India, South Africa, and Brazil. If platforms begin to rely even more on LLMs for content moderation, this problem will likely get worse, says Marlena Wisniak, a human rights lawyer who focuses on AI governance at the European Center for Not-for-Profit Law. The LLMs are moderated poorly, and the poorly moderated LLMs are then also used to moderate other content, she tells me. Its so circular, and the errors just keep repeating and amplifying. Part of the problem is that the systems are trained primarily on data from the English-speaking world (and American English at that), and as a result, they perform less well with local languages and context. Even multilingual language models, which are meant to process multiple languages at once, still perform poorly with non-Western languages. For instance, one evaluation of ChatGPTs response to health-care queries found that results were far worse in Chinese and Hindi, which are less well represented in North American data sets, than in English and Spanish. For many at RightsCon, this validates their calls for more community-driven approaches to AIboth in and out of the social media context. These could include small language models, chatbots, and data sets designed for particular uses and specific to particular languages and cultural contexts. These systems could be trained to recognize slang usages and slurs, interpret words or phrases written in a mix of languages and even alphabets, and identify reclaimed language (onetime slurs that the targeted group has decided to embrace). All of these tend to be missed or miscategorized by language models and automated systems trained primarily on Anglo-American English. The founder of the startup Shhor AI, for example, hosted a panel at RightsCon and talked about its new content moderation API focused on Indian vernacular languages. Many similar solutions have been in development for yearsand weve covered a number of them, including a Mozilla-facilitated volunteer-led effort to collect training data in languages other than English, and promising startups like Lelapa AI, which is building AI for African languages. Earlier this year, we even included small language models on our 2025 list of top 10 breakthrough technologies. Still, this moment feels a little different. The second Trump administration, which shapes the actions and policies of American tech companies, is obviously a major factor. But there are others at play. First, recent research and development on language models has reached the point where data set size is no longer a predictor of performance, meaning that more people can create them. In fact, smaller language models might be worthy competitors of multilingual language models in specific, low-resource languages, says Aliya Bhatia, a visiting fellow at the Center for Democracy & Technology who researches automated content moderation. Then theres the global landscape. AI competition was a major theme of the recent Paris AI Summit, which took place the week before RightsCon. Since then, theres been a steady stream of announcements about sovereign AI initiatives that aim to give a country (or organization) full control over all aspects of AI development. AI sovereignty is just one part of the desire for broader tech sovereignty thats also been gaining steam, growing out of more sweeping concerns about the privacy and security of data transferred to the United States. The European Union appointed its first commissioner for tech sovereignty, security, and democracy last November and has been working on plans for a Euro Stack, or digital public infrastructure. The definition of this is still somewhat fluid, but it could include the energy, water, chips, cloud services, software, data, and AI needed to support modern society and future innovation. All these are largely provided by US tech companies today. Europes efforts are partly modeled after India Stack, that countrys digital infrastructure that includes the biometric identity system Aadhaar. Just last week, Dutch lawmakers passed several motions to untangle the country from US tech providers. This all fits in with what Andy Yen, CEO of the Switzerland-based digital privacy company Proton, told me at RightsCon. Trump, he said, is causing Europe to move faster to come to the realization that Europe needs to regain its tech sovereignty. This is partly because of the leverage that the president has over tech CEOs, Yen said, and also simply because tech is where the future economic growth of any country is. But just because governments get involved doesnt mean that issues around inclusion in language models will go away. I think there needs to be guardrails about what the role of the government here is. Where it gets tricky is if the government decides These are the languages we want to advance or These are the types of views we want represented in a data set, Bhatia says. Fundamentally, the training data a model trains on is akin to the worldview it develops. Its still too early to know what this will all look like, and how much of it will prove to be hype. But no matter what happens, this is a space well be watching. Now read the rest of The Algorithm Deeper Learning OpenAI has released its first research into how using ChatGPT affects peoples emotional well-being OpenAI released two pieces of research last week that explore how ChatGPT affects people who engage with it on emotional issues, yielding some interesting results. Female study participants were slightly less likely to socialize with people than their male counterparts who used the chatbot for the same period of time, our reporter Rhiannon Williams writes. And people who used voice mode in a gender that was not their own reported higher levels of loneliness at the end of the experiment. Why it matters: AI companies have raced to build chatbots that act not just as productivity tools but also as companions, romantic partners, friends, therapists, and more. Legally, its largely still a Wild West landscape. Some have instructed users to harm themselves, and others have offered sexually charged conversations as underage characters represented by deepfakes. More research into how people, especially children, are using these AI models is essential. OpenAIs work is only a start. Read more from Rhiannon Williams. Bits and Bytes Opinion Why handing over total control to AI agents would be a huge mistake Companies like OpenAI and Butterfly Effect (the startup in China that made Manus) are racing to build AI agents that can do tasks for you by taking over your computer. In this op-ed, some top AI researchers detail the potential missteps that could occur if we cede more control of our digital lives to decision-making AIs. A provocative experiment pitted AI against federal judges Research has long shown that judges are influenced by many factors, like how sympathetic they are to defendants, or when their last meal was. Despite AI models inherent problems with biases and hallucinations, researchers at the University of Chicago Law School wondered if they can present more objective opinions. They can, but that doesnt make them better judges, the researchers say. (The Washington Post) Elon Musks truth-seeking chatbot often disagrees with him Musk promised his company xAIs model Grok would be an antidote to the woke and politically influenced chatbots that he says dominate today. But in tests done by the Washington Post, the model contradicted many of Musks claims about specific issues. (The Washington Post) A Disney employee downloaded an AI tool that contained malware, and it ruined his life MIT Technology Review has long predicted that the proliferation of AI will enable scammers to up their productivity as never before. One victim of this trend is Matthew Van Andel, a Disney employee who downloaded malware disguised as an AI tool. It led to his firing. (Wall Street Journal) The facial recognition company Clearview attempted to buy Social Security numbers and mugshots for its database Three years ago, Clearview was fined for scraping images of individuals faces from the internet. Now, court records reveal that the company was attempting to buy 690 million arrest records and 390 million arrest photos in the USrecords that also contained Social Security numbers, emails, and physical addresses. The deal fell through, but Clearview nonetheless holds one of the largest databases of facial images, and its tools are used by police and federal agencies. (404 Media)

This is today's edition ofThe Download,our weekday newsletter that provides a daily dose of what's going on in the world of technology. Why handing over total control to AI agents would be a huge mistake Margaret Mitchell, Avijit Ghosh, Sasha Luccioni, Giada Pistilli all work for Hugging Face, an open source AI company. AI agents have set the tech industry abuzz. Unlike chatbots, these groundbreaking new systems can navigate multiple applications to execute complex tasks, like scheduling meetings or shopping online, in response to simple user commands. As agents become more capable, a crucial question emerges: How much control are we willing to surrender, and at what cost? The promise is compelling. Who doesnt want assistance with cumbersome work or tasks theres no time for? But this vision for AI agents brings significant risks that might be overlooked in the rush toward greater autonomy. In fact, our research suggests that agent development could be on the cusp of a very serious misstep. Read the full story.OpenAI has released its first research into how using ChatGPT affects peoples emotional wellbeing OpenAI says over 400 million people use ChatGPT every week. But how does interacting with it affect us? Does it make us more or less lonely? These are some of the questions OpenAI set out to investigate, in partnership with the MIT Media Lab, in a pair of new studies. They found that while only a small subset of users engage emotionally with ChatGPT, there are some intriguing differences between how men and women respond to using the chatbot. They also found that participants who trusted and bonded with ChatGPT more were likelier than others to be lonely, and to rely on it more.Chatbots powered by large language models are still a nascent technology, and difficult to study. Thats why this kind of research is an important first step toward greater insight into ChatGPTs impact on us, which could help AI platforms enable safer and healthier interactions. Read the full story. Rhiannon Williams The must-reads Ive combed the internet to find you todays most fun/important/scary/fascinating stories about technology. 1 Genetic testing firm 23andMe has filed for bankruptcy protection Following months of uncertainty over its future. (CNN)+ Tens of millions of peoples genetic data could soon belong to a new owner. (WSJ $)+ How to delete your 23andMe data. (MIT Technology Review)2 Europe wants to lessen its reliance of US cloud giants But thats easier said than done. (Wired $)3 Anduril is considering opening a drone factory in the UK Europe is poised to invest heavily in defenseand Anduril wants in. (Bloomberg $)+ The company recently signed a major drone contract with the UK government. (Insider $)+ We saw a demo of the new AI system powering Andurils vision for war. (MIT Technology Review)4 Bird flu has been detected in a sheep in the UK Its the first known instance of the virus infecting a sheep. (FT $)+ But the UK is yet to report any transmission to humans. (Reuters)+ How the US is preparing for a potential bird flu pandemic. (MIT Technology Review)5 A tiny town in the Alps has emerged as an ALS hotspotSuggesting that its causes may be more environmental than genetic. (The Atlantic $) + Motor neuron diseases took their voices. AI is bringing them back. (MIT Technology Review)6 Firefly Aerospaces Blue Ghost lunar lander has completed its missionAnd captured some pretty incredible footage along the way. (NYT $) + Europe is finally getting serious about commercial rockets. (MIT Technology Review) 7 How the US could save billions of dollars in wasted energy WSJ $) 8 We need new ways to measure painResearchers are searching for objective biological indicators to get rid of the guesswork. (WP $) + Brain waves can tell us how much pain someone is in. (MIT Technology Review)9 What falling in love with an AI could look like Its unclear whether loving machines could be training grounds for future relationships, or the future of relationships themselves. (New Yorker $)+ The AI relationship revolution is already here. (MIT Technology Review)10 Could you walk in a straight line for hundreds of miles? YouTubes favorite new challenge isnt so much arduous as it is inconvenient. (The Guardian)Quote of the day Blockbuster has collapsed. Its time for Netflix to rise. Kian Sadeghi pitches the company they founded, DNA testing firm Nucleus Genomics, as a replacement for 23andMe in a post on X. The big story This towns mining battle reveals the contentious path to a cleaner future January 2024 In June last year, Talon, an exploratory mining company, submitted a proposal to Minnesota state regulators to begin digging up as much as 725,000 metric tons of raw ore per year, mainly to unlock the rich and lucrative reserves of high-grade nickel in the bedrock. Talon is striving to distance itself from the mining industrys dirty past, portraying its plan as a clean, friendly model of modern mineral extraction. It proclaims the site will help to power a greener future for the US by producing the nickel needed to manufacture batteries for electric cars and trucks, but with low emissions and light environmental impacts. But as the company has quickly discovered, a lot of locals arent eager for major mining operations near their towns. Read the full story. James Temple We can still have nice things A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet 'em at me.) + Who are fandoms for, and who gets to escape into them? + A long-lost Klimt painting of Prince William Nii Nortey Dowuona has gone on display in the Netherlands.+ Feeling down? These feel-good movies will pick you right up.+ Why Gen Z are dedicated followers of Old Money fashion.

AI agents have set the tech industry abuzz. Unlike chatbots, these groundbreaking new systems operate outside of a chat window, navigating multiple applications to execute complex tasks, like scheduling meetings or shopping online, in response to simple user commands. As agents are developed to become more capable, a crucial question emerges: How much control are we willing to surrender, and at what cost? New frameworks and functionalities for AI agents are announced almost weekly, and companies promote the technology as a way to make our lives easier by completing tasks we cant do or dont want to do. Prominent examples include computer use, a function that enables Anthropics Claude system to act directly on your computer screen, and the general AI agent Manus, which can use online tools for a variety of tasks, like scouting out customers or planning trips. These developments mark a major advance in artificial intelligence: systems designed to operate in the digital world without direct human oversight. The promise is compelling. Who doesnt want assistance with cumbersome work or tasks theres no time for? Agent assistance could soon take many different forms, such as reminding you to ask a colleague about their kids basketball tournament or finding images for your next presentation. Within a few weeks, theyll probably be able to make presentations for you. Theres also clear potential for deeply meaningful differences in peoples lives. For people with hand mobility issues or low vision, agents could complete tasks online in response to simple language commands. Agents could also coordinate simultaneous assistance across large groups of people in critical situations, such as by routing traffic to help drivers flee an area en masse as quickly as possible when disaster strikes. But this vision for AI agents brings significant risks that might be overlooked in the rush toward greater autonomy. Our research team at Hugging Face has spent years implementing and investigating these systems, and our recent findings suggest that agent development could be on the cusp of a very serious misstep. Giving up control, bit by bit This core issue lies at the heart of whats most exciting about AI agents: The more autonomous an AI system is, the more we cede human control. AI agents are developed to be flexible, capable of completing a diverse array of tasks that dont have to be directly programmed. For many systems, this flexibility is made possible because theyre built on large language models, which are unpredictable and prone to significant (and sometimes comical) errors. When an LLM generates text in a chat interface, any errors stay confined to that conversation. But when a system can act independently and with access to multiple applications, it may perform actions we didnt intend, such as manipulating files, impersonating users, or making unauthorized transactions. The very feature being soldreduced human oversightis the primary vulnerability. To understand the overall risk-benefit landscape, its useful to characterize AI agent systems on a spectrum of autonomy. The lowest level consists of simple processors that have no impact on program flow, like chatbots that greet you on a company website. The highest level, fully autonomous agents, can write and execute new code without human constraints or oversightthey can take action (moving around files, changing records, communicating in email, etc.) without your asking for anything. Intermediate levels include routers, which decide which human-provided steps to take; tool callers, which run human-written functions using agent-suggested tools; and multistep agents that determine which functions to do when and how. Each represents an incremental removal of human control. Its clear that AI agents can be extraordinarily helpful for what we do every day. But this brings clear privacy, safety, and security concerns. Agents that help bring you up to speed on someone would require that individuals personal information and extensive surveillance over your previous interactions, which could result in serious privacy breaches. Agents that create directions from building plans could be used by malicious actors to gain access to unauthorized areas. And when systems can control multiple information sources simultaneously, potential for harm explodes. For example, an agent with access to both private communications and public platforms could share personal information on social media. That information might not be true, but it would fly under the radar of traditional fact-checking mechanisms and could be amplified with further sharing to create serious reputational damage. We imagine that It wasnt meit was my agent!! will soon be a common refrain to excuse bad outcomes. Keep the human in the loop Historical precedent demonstrates why maintaining human oversight is critical. In 1980, computer systems falsely indicated that over 2,000 Soviet missiles were heading toward North America. This error triggered emergency procedures that brought us perilously close to catastrophe. What averted disaster was human cross-verification between different warning systems. Had decision-making been fully delegated to autonomous systems prioritizing speed over certainty, the outcome might have been catastrophic. Some will counter that the benefits are worth the risks, but wed argue that realizing those benefits doesnt require surrendering complete human control. Instead, the development of AI agents must occur alongside the development of guaranteed human oversight in a way that limits the scope of what AI agents can do. Open-source agent systems are one way to address risks, since these systems allow for greater human oversight of what systems can and cannot do. At Hugging Face were developing smolagents, a framework that provides sandboxed secure environments and allows developers to build agents with transparency at their core so that any independent group can verify whether there is appropriate human control. This approach stands in stark contrast to the prevailing trend toward increasingly complex, opaque AI systems that obscure their decision-making processes behind layers of proprietary technology, making it impossible to guarantee safety. As we navigate the development of increasingly sophisticated AI agents, we must recognize that the most important feature of any technology isnt increasing efficiency but fostering human well-being. This means creating systems that remain tools rather than decision-makers, assistants rather than replacements. Human judgment, with all its imperfections, remains the essential component in ensuring that these systems serve rather than subvert our interests. Margaret Mitchell, Avijit Ghosh, Sasha Luccioni, Giada Pistilli all work for Hugging Face, a global startup in responsible open-source AI. Dr. Margaret Mitchell is a researcher and Chief Ethics Scientist at Hugging Face. Dr. Sasha Luccioni is Climate Lead at Hugging Face, where she spearheads research, consulting and capacity-building to elevate the sustainability of AI systems. Dr. Avijit Ghosh is an Applied Policy Researcher at Hugging Face working at the intersection of responsible AI and policy. His research and engagement with policymakers has helped shape AI regulation and industry practices. Dr. Giada Pistilli is a philosophy researcher working as Principal Ethicist at Hugging Face.