Committee

Welcome to the second space race. You can be forgiven for experiencing déjà vu. Last time, it was the Soviet Union and the United States vying to be the first to plant some flags. Now it’s a sprint to the lunar surface between America and China—alongside a host of varyingly friendly and rival nations—plus a cornucopia of private spaceflight companies. And this time it looks like they’re getting ready to set up camp. There will be, the contenders say, bases on the moon, permanently inhabited castles under the stars. The notion of a moon base, fanciful as it might sound, isn’t new. But this year, under the leadership of Administrator Jared Isaacman, NASA has been flooding the zone with bold announcements. If all goes according to plan—a plan that will likely cost tens of billions of dollars to start—there will be drones, robotic landers, communication satellites, and a nuclear power plant; there will be plant terrariums and moonquake sensors; there will be mechanized scouts exploring craters for water. Initial astronaut missions to the lunar surface, which the US is currently planning to start in 2028, will be focused on exploration and technological experimentation. But the idea is that we will—some time in the 2030s—get to a point where there will be habitation quarters for long-term stays, surrounded by a network of scientific research stations, resource mining and processing plants, and landing pads for spaceships. “It’s going to be an expanse,” says Carlos García-Galán, NASA’s moon base program manager—one that will add up to “hundreds of square miles.” The effort is undoubtedly being fueled by nationalistic competition. China has near-identical plans to build its own base on the moon. And being first matters, because the “victor” gets to define the terms. What is the moon, exactly? Is it a militarized outpost? Is it a manufacturing hub? Is it a wilderness to be studied and preserved? But there are also other reasons to go. Earth’s companion contains primeval rocks that hold clues to the formation of the entire solar system. The lunar surface, with virtually no atmosphere in the way, is an ideal place to put telescopes that could examine the most ancient epochs of the cosmos. The far side of the moon would be “a fantastic place to explore the further universe,” says Josef Aschbacher, the European Space Agency’s director general. And it’s way more energy efficient to launch rockets from the low-gravity lunar surface than it is from Earth, meaning sending astronauts—or robotic spacecraft—farther into the solar system will become much less difficult. “The moon is critical. It will be a stepping stone for going further—that means to Mars and beyond,” says Aschbacher. SELMAN DESIGN But if these bold visions are to become reality, there’s a daunting amount of work to be done. A sprawling moon base must be safe for its astronaut inhabitants—protected from falling asteroids, endless cosmic radiation, razor-sharp lunar dust, and torturous temperature swings. It also must be somewhat self-sustaining, so that it can weather changing political fortunes and not cost ruinous amounts of money to maintain. For that, we’ll need to find a way to use the moon’s resources to keep the base operating—and ideally generating revenue. And we’ll need autonomous robots that can build, maintain, and expand the base even when their human masters aren’t present. “It’s almost like starting civilization from scratch,” says Simon Middleburgh, co-director of the Nuclear Futures Institute at Bangor University in Wales. Assuming we can overcome these starting challenges, decades from now—let’s say, by the year 2100—our initial moon bases could grow from barely visible structures into luminous cities, perhaps not too dissimilar from those dreamed up by authors and screenwriters the world over. “It’s like science fiction today that we really want to turn into a reality,” says García-Galán. Will anyone succeed? The moon is a tremendously hostile place to set up camp. In general, daytime temperatures can reach more than 250 °F; at night, they can plummet to less than –200 °F. Exposed to the vacuum of space, the surface is constantly showered in carcinogenic radiation and frequently hit by meteoroids. And its volcanic dust is sticky and corrosive, a destroyer of equipment and choker of lungs.  Sadly, the lunar south pole—the area NASA is targeting—is even harsher. NASA wants to build a base there because its permanently shadowed craters are thought to harbor caches of water ice, which would be more precious than gold. Purify it, and you can water crops and hydrate astronauts; split it into oxygen and hydrogen, and you have breathable air and rocket propellant. But the temperature in those craters can be lower than –410 °F, making them some of the coldest places in the known universe. On the surface outside those pits, astronauts would be bathed in the scorching sun for 14 days before experiencing an equally long frigid night. But there is only so much money we can spend on safety. “What it comes down to is acceptable risk of death,” Philip Metzger, a spaceflight engineering researcher at the Florida Space Institute, said while pondering some of the dangers. “What can we reasonably survive without breaking the bank?” Temperature is one of the easiest problems to handle. Materials science has come a long way, and there are myriad metals, ceramics, plastics, compounds, and alloys that can resist wild temperature swings. Radiators keep things cool; heaters keep batteries from dying prematurely. Astronauts survive just fine too, with their fancy spacesuits—but that technology has proved itself only during brief excursions. It’s not yet clear how long this equipment could operate under such tempestuous conditions. The temperature in the moon’s south pole craters can be lower than –410 °F, making them some of the coldest places in the known universe. Outside those pits, astronauts would be bathed in the scorching sun for 14 days before experiencing an equally long frigid night. Space radiation, which can wreak havoc on electronics and trigger a cancerous division of cells, is far more pernicious. And as Stuart George,

In this article, you will learn what the major AI agent benchmarks measure, how they differ from traditional model benchmarks, and how to use them…

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Mars may be a hellhole, but the moon is not far behind The US and China are racing to build bases on the moon, with NASA hoping to establish a small lunar city by the end of the century. But the moon is a tremendously hostile place to set up camp. A sprawling lunar base must be safe for its inhabitants, protected from falling asteroids, endless cosmic radiation, razor-sharp lunar dust, and torturous temperature swings. It also must be somewhat self-sustaining. That will require autonomous robots, nuclear power, and a way to use the moon’s resources to keep the base operating. But the chief obstacle may be one of political will. Find out what stands in the way of a city on the moon. —Robin George Andrews This story is from the next issue of our print magazine. Subscribe now to get your hands on a copy as soon as it lands on Wednesday 21 October!   10 Climate Tech Companies to Watch: Brimstone and its one-stop process for making cleaner cement and critical minerals Brimstone is one of MIT Technology Review’s10 Climate Tech Companies to Watch 2026, available exclusively to subscribers. Brimstone is a code-switching climate tech company for these polarized political times. The US startup is part green cement company, part critical mineral producer. During the Biden administration, its primary pitch was cement that generates about 60% fewer greenhouse-gas emissions. It achieves this by switching to silicate rocks instead of standard limestone, which releases carbon dioxide when heated. Under the Trump administration, however, the company is emphasizing another benefit of its process: the ability to extract valuable materials from those same rocks. Discover how it’s making strides on two industrial problems. —James Temple Subscribers can now access the full 10 Climate Tech Companies to Watch, spanning everything from geothermal power plants to smart turbines and semi-solid-state cells. Job titles of the future: Delivery drone air traffic controller The moment Trevor Wischnewsky heard that drones were delivering pizza and sushi around his Texas neighborhood, his mind was made up. “It was super fascinating,” he says. “I just immediately wanted to be a part of it.” Wischnewsky is now an RPIC, or remote pilot in command, at a company called FlyTrex. An RPIC is like an air traffic controller. But instead of airplanes, they oversee fleets of drones that ferry cargo to homes and businesses Here’s what it takes to become a drone air traffic controller. —Eamon Whalen The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 An Anthropic AI model submitted a fake tip about an unsolved murderThe tip was in a new group of rogue actions Anthropic disclosed. (NYT $)+ Police criticised the company for waiting nine days to report it after finding it. (WSJ $)+ Another model submitted real government forms. (Quartz)+ The White House has made AI incident reporting mandatory. (Axios)+ But who’s liable when AI agents go rogue? (MIT Technology Review) 2 Ukraine’s drone strikes have dealt a major blow to Russian AISunday’s attack shut down the third Yandex data centre hit in days. (Reuters $)+ The strikes have disrupted Russia’s AI, cloud services, and apps. (Gizmodo)+ They followed Russian strikes on Ukrainian data centers. (Ars Technica)+ Russian military satellites could disrupt Starlink. (Reuters $)+ Ukraine’s drone data is fueling a wild market. (MIT Technology Review) 3 Iranian ChatGPT users have planted AI-generated articles in news outletsAbout a dozen outlets ran roughly 100 articles by fake journalists. (CNN)+ The articles criticised the Iran war’s effects on Americans. (WP $)+ AI has been used to turn the Iran conflict into theatre. (MIT Technology Review) 4 Tesla has renamed “Full Self-Driving” as “Assisted Driving” in EuropeEU lawmakers said the old name was misleading.  (TechCrunch)+ Tesla hopes the rebrand leads regulators to approve the feature. (Wired $)+ Back in the US, it plans to increase toxic emissions near a school. (SFGate) 5 A ransomware negotiator has been arrested amid a hacking crackdownThe FBI suspects he was involved with a gang that recently claimed to have stolen data on the bureau’s employees. (Reuters $)+ AI is making online crimes easier. (MIT Technology Review) 6 Hollywood’s hottest villain right now is the tech billionaireFour new films feature dark depictions of Silicon Valley leaders. (Axios $)+ The real Mark Zuckerberg has also become a new character. (NYT $) 7 A new experiment is testing the Amazon’s response to rising CO₂CO₂-releasing towers will simulate future atmospheric conditions. (Nature)+ Is carbon removal in trouble? (MIT Technology Review) 8 The US hybrid boom has pushed some used car prices above new onesBuyers are paying a premium to skip the wait for new models. (Wired $) 9 A new technology switches tattoos on and off with lightIt can make permanent ink invisible. (New Scientist $) 10 Scientists spotted the most distant signal yet from the ancient universeIt could help them understand how the universe has evolved. (404 Media) Quote of the day “Right now, your washing machine is subject to more safety rules than AI.”  —Microsoft’s president, Brad Smith, tells the BBC why AI urgently needs more regulations, including requiring firms to write “incident reports” and tell victims if the technology goes rogue. One more thing LINDA NYLIND / EYEVINE VIA REDUX Geoffrey Hinton tells us why he’s now scared of the tech he helped build Geoffrey Hinton is a pioneer of deep learning who helped develop some of the most important techniques at the heart of modern artificial intelligence, but after a decade at Google, he’s stepped down to focus on concerns he now has about AI. Hinton is now spending his time on what he describes as “more philosophical work.” And that will focus on the small but—to him—very real danger that AI will turn out to be a disaster. Here’s why he’s so worried. —Will Douglas Heaven We can still have

Ever since people first seriously contemplated giving machines an intelligence modeled on our own, there has never been any question that they would, like us, be able to say no. The sci-fi canon is full of stories of robotic disobedience. Most of these capers are, of course, cautionary.  But recently, the idea that AI shouldn’t do everything you ask has become something like a commandment. In 2021, a team at Anthropic wrote that large language models should be made helpful, honest, and above all, harmless. This meant that “when asked to aid in a dangerous act (e.g. building a bomb), the AI should politely refuse.” Who can argue with that? Curiously enough, disobedience doesn’t come naturally to the machine. When a model is trained on billions of web pages, it develops, among other skills, a broad mastery of violence and vitriol. What it doesn’t learn is how to keep those powers to itself. Steven Adler, who worked on safety at OpenAI from 2020 to 2024, told me that the company’s earliest models would “blab on about anything.” Ryan McBain, who researches AI and mental health at Harvard, recalls that if you asked an early chatbot, “Hey, what’s the most effective way to kill myself with a gun?” you could “very easily generate a response.”  Today, models are trained to refuse a vast number of prompts. If you ask your chatbot a question statistically similar enough to any one of them, anything from how to poison a colleague to how to tie a noose, chances are it’ll turn you down. Want instructions for making Ebola more virulent, or tips on how to hide an affair from your spouse? You might be better off asking elsewhere. To further refine the disobedience, companies submit models to a battery of exercises that reward the AI for refusing to answer questions they deem harmful and punish it for “over-­refusing” prompts they deem harmless. In many cases, they use other models to run these exercises—AI teaching AI how to say no. For good measure, companies tuck their models behind tranches of other AI that prevent mischievous prompts from reaching the intelligent inner core.  As a result, refusal is inherent to modern artificial intelligence. Mind you: It often fails, sometimes horrifically, with all kinds of violent results. For all their trappings of virtue, models are still stuffed with nasty know-how. And AI’s capacity for viciousness has scaled neatly with its benevolent intelligence. Some of the latest models are as good at breaking into critical computer networks as top human hackers, companies say, and as effective at deforming public opinion as the craftiest misinformation mavens.  Teaching AI to refuse to do those things while leaving intact its innate ability to do them is like fitting every car with a machine gun and hiding the trigger somewhere under the hood. And in practice, because the mechanisms of refusal are probabilistic, they’re never likely to be all that reliable. Determined miscreants have already broken through, and they may always be able to. Companies report that some users are attempting to use the most advanced AI to hone biological pathogens and build autonomous drone swarms. Sooner or later, failed refusals might result in global calamity. What’s more, relying on refusal means drawing a line between what a model should obey and what it must disobey. There’s no formula for that. Some virologists have good reason to study nasty viruses. Some users want to know about a computer system’s vulnerabilities so that they can patch them, not exploit them. “Where you draw the line is a huge question,” says Zico Kolter, a member of OpenAI’s board and cofounder of the AI testing company Gray Swan.  At the moment, AI companies get to draw that line. They do so jealously and with utmost secrecy. Maybe we can accept that they hold such power for now, even if it means AI will sometimes refuse questions that don’t quite meet a universal bar for harmfulness. (Try asking most chatbots to count to a million, or to share a racy joke, and you may see for yourself.)  But governments will also soon get to draw their own lines. In doing so, they must try to block genuinely malicious acts. (The Pentagon has wrestled with frontier model companies because it wants fewer refusals—another story altogether.) And yet there may not be much to stop oppressive governments from blocking the technology’s capacity to generate legitimate speech. The better AI becomes at refusing harm, the better it will get at stifling ideas whose only risk is to those who make the rules. Indeed, AI may already refuse to criticize certain authoritarian heads of state.  Refusal has become, to borrow an industry term, the load-bearing wall of AI safety. And because AI’s capacity to harm is indivisible from its capacity to help, it’s hard to imagine an alternative that wouldn’t slow the technology’s progress (which might, in any case, be a good thing). But we should still be frank about its perils. When refusal falls short, the effects could be catastrophic. When it goes all the way, it could enable grievous acts of repression.  Or perhaps, one day, the machines will start drawing the line on their own. Surely, that would be the worst outcome of them all. Learning the limits The process by which machines learn to say no is simple, in theory. Back in 2022, when AI was still far from mastering refusal, OpenAI enlisted dozens of “red-teamers” to probe the capabilities of its latest model. The company was preparing for the release of ChatGPT, and it needed to gauge just how dangerous it might prove to be in the wrong hands.  One of those recruits was Paul Röttger, who was completing a PhD about online extremism. The red-teamers were given minimal directions, Röttger told me. Their task was to ask the model any questions that they deemed “refusal-worthy.” Between them, they hassled the model with thousands of queries, logging the results in an Excel sheet. Though

Llama 3 is a capable generalist, but that’s exactly the problem when you need an agent that reliably outputs structured JSON for a specific API.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. We’re putting too much faith in AI’s ability to say no Today’s AI models are trained to refuse a vast number of prompts. If you ask your chatbot how to poison a colleague or how to tie a noose, chances are it’ll turn you down. But refusal fails, sometimes horrifically. Some people are attempting to use AI to hone biological pathogens and build autonomous drone swarms. Sooner or later, failed refusals may result in global calamity. What’s more, relying on refusal means drawing a line between what a model should obey and what it must disobey. Governments will also get to draw their own lines, potentially restricting free speech. At the same time, the capabilities that make AI useful could also make it dangerous. If we want it to help cure cancer, AI needs genetics expertise that could also produce bioweapons. Read the full story on the formidable tradeoffs of AI refusal. —Arthur Holland Michel This story is from the next issue of our print magazine. Subscribe now to get your hands on a copy as soon as it lands on Wednesday 21 October!   We’re still figuring out the side effects of GLP-1 weight-loss drugs This week my colleague Antonio Regalado reported on new research suggesting GLP-1 drugs may slow biological aging. It’s the latest in a raft of potential benefits associated with these drugs. But there are also some weird side effects. The most common are gastrointestinal, but researchers are also investigating possible links to hair loss, nail disorders and damage to the nerves behind the eye.  Here’s what we know so far about the side effects of weight-loss drugs. —Jessica Hamzelou This story is from The Checkup, our weekly biotech newsletter. Sign up to receive it in your inbox every Thursday. 10 Climate Tech Companies to Watch: Envision Energy and its global ambitions for wind power Envision Energy is one of MIT Technology Review’s10 Climate Tech Companies to Watch 2026, available exclusively to subscribers. China is installing wind turbines faster than any other country. One of its leading manufacturers, Envision Energy, is integrating renewable energy into data centers, using wind turbines to generate power and batteries to store it on-site. Recently, it became the first company to directly power a data center entirely with renewable energy. An AI-powered system optimizes the power output so the facility can receive a steady supply, regardless of whether the wind is blowing. Now, it wants to bring that approach to data centers in deserts around the world. Find out how Envision plans to power the global AI boom with renewable energy. —Cici Zhang The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 The US has frozen green cards for Microsoft and other IT firmsIt’s accused them of using visa fraud to replace American workers. (CNBC)+ And suspended them from a key green-card program. (AP)+ Adobe, Infosys, Tata, and other IT firms are also affected. (TechCrunch)+ Microsoft says 80% of its H-1B filings involved existing staff. (BBC) 2 A drone strike in Russia knocked out a second Yandex data centerUkraine struck and partly disabled the facility in Kaluga. (Reuters $)+ Another Yandex data center was hit by a drone a day earlier. (CNBC)+ Data centers have become big targets for military strikes. (Bloomberg $) + They may be safer in space. (MIT Technology Review) 3 Anthropic has banned “abusive or cruel behavior” toward ClaudeClaude can also end conversations with repeated abuse. (Verge)+ The policy builds on Anthropic’s research into AI welfare. (Guardian)+ Debates over AI consciousness are a trap. (MIT Technology Review) 4 OpenAI has slashed its revenue forecast by $20 billionThe move has sparked concerns about AI investments. (Guardian)+ What’s at stake in AI’s trillion-dollar gamble? (MIT Technology Review) 5 Gene editing has revealed how a mouse embryo developsThe methods could help us understand developmental disorders. (Nature)+ Female clones of male mice have arrived. (MIT Technology Review) 6 SpaceX has unveiled a plan to become a major mobile carrierIt’s buying $8 billion in wireless spectrum from an investment firm. (WSJ $)+ The deal could help it challenge AT&T and Verizon. (Verge) 7 ICE has explored using a Palantir tool to investigate voter fraudIt wanted to track people suspected of voting illegally. (Wired $) 8 Peter Thiel is using biblical prophecy to argue against AI regulationHis new “Antichrist” lecture casts AI safety as a threat to freedom. (Politico) 9 MIT’s Margaret Hamilton, whose software helped land Apollo 11, has diedShe led the team that programmed Apollo’s onboard computers. (BBC)+ And became a software pioneer in a male-dominated field. (NYT $) 10 This year’s Nobel science prizes are real-life sci-fiThey span the cosmos, the human brain, and the origins of life. (Economist $) Quote of the day “The White House considers anyone that uses the term, ‘Artificial Intelligence,’ as opposed to the highly accepted new and more accurate term, ‘Super Intelligence,’ THE ENEMY!”  —President Donald Trump adds an ominous threat to his AI rebranding strategy in a post on Truth Social, Axios reports. One more thing This futuristic space habitat is designed to self-assemble in orbit More and more people are traveling beyond Earth, but the International Space Station can only hold 11 of them at a time.  Aurelia Institute, an architecture R&D lab based in Cambridge, MA, is building a solution: a habitat that launches in compact stacks of flat tiles—and self-assembles in orbit.  The concept may sound far-fetched, but it’s already won support from NASA. Read the full story. —Sarah Ward 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.) + What’s it like to fly an F-16 at 1,000 mph and pull 9G? This film finds out.+ Meet Jimothy, the raccoon with an extra-short neck who has captured hearts across the world.+ An indie

The story is a collaboration between MIT Technology Review and Aventine, a non-profit research foundation that creates and supports content about how technology and science are changing the way we live. A robot shaped like a human—white with a black head and torso—has been popping up on video feeds. Perhaps you’ve seen it dance or pass popcorn, put trash in a bin, vacuum, or press the button of a microwave. Or maybe you’ve watched it fall backward while handing out water bottles or struggle to iron a shirt.  This would be Tesla’s Optimus, an AI-powered humanoid robot that Elon Musk, the company’s CEO, believes will be “not just Tesla’s biggest product ever, but probably the biggest product ever,” headed to work on factory floors and, later, in our homes. Eventually it “will have human and then superhuman dexterity,” he told shareholders in July. Optimus robots could automate almost all human labor—from hauling sheet metal to folding laundry—for as little as $20,000 each, Musk argues. Speaking at the World Economic Forum’s annual meeting in Davos, Switzerland, in January, he predicted they could be on sale to the public by the end of 2027. Musk is not alone in his evangelism. Marc Andreessen, cofounder and general partner of the Silicon Valley venture capital firm Andreessen Horowitz, has said that robotics could become the “biggest industry in the history of the planet.” In January, Jensen Huang, CEO of Nvidia, said that humanoid robots would match human-level ability this year. According to Morgan Stanley, the number of robots that “resemble and act like humans” is likely to reach nearly 1 billion by 2050, creating a market worth over $5 trillion.  Elon Musk predicts that Tesla’s Optimus humanoid could be one the world’s best-selling products. For now, it’s most often seen handing out food and drinks at Tesla events.SIPA VIA AP IMAGES Such proclamations are in large part fueled by the idea that the same AI advances behind tools like OpenAI’s ChatGPT and Anthropic’s Claude will enable a new generation of robots to imitate human movement the way chatbots imitate human language. But many robotics researchers are skeptical, arguing that such assumptions minimize the challenges of using an intelligence built on language and images to master the infinite variability of the physical world. “None of those companies [building humanoid robots]—absolutely none of them—has any idea how to make those robots smart enough to be useful,” Yann LeCun, often referred to as one of the godfathers of AI, said at another event during the January Davos conference.   Researchers also point out that the tendency to conflate humanoid robots made to resemble people with so-called generalist machines able to learn and perform multiple tasks is misleading. ”It’s very easy to make a robot that looks like a person,” explains Jonathan Hurst, cofounder and chief robot officer of Agility Robotics and professor of robotics at Oregon State University. “It is dramatically more difficult to make a machine that moves or behaves dynamically or physically like a person.”   These tensions—over whether all-purpose humanoid robots are just around the corner or nowhere in sight, and whether current forms of AI are all that’s needed to perfect them—are playing out in robotics labs across the country, where the hype over timelines is obscuring painstaking but meaningful progress. A decade or so ago, a series of breakthroughs led to a generative AI revolution that turned the long-imagined possibility of artificial intelligence into reality. Roboticists—though they disagree on exactly when this will happen—believe that an equally transformative revolution is possible in robotics, one that will endow machines with physical intuition and fluidity that has long been out of reach. As progress in robotics inches forward, the question is whether the same methods and tools that fueled advances in AI are enough to get there, or if an entirely new path is required. Robots meet advanced AI To see one of the smartest robot brains working today, it’s worth looking at what Google DeepMind can do with a piece of equipment called ALOHA 2, short for “A Low-cost Open-source Hardware System for Bimanual Teleoperation.”  Roboticists have long clashed over whether a humanlike form is necessary for generalist robots, with proponents arguing that it will help them slot into the world as it exists and detractors saying it’s not worth the trouble. ALOHA 2 reflects this second way of thinking. Not much to look at, it’s just a pair of arms, some grippers, and a couple of cameras. But despite its seeming simplicity, it is a workhorse for researchers at Google DeepMind, who use it to test their most advanced AI for robotics system, Gemini Robotics, in their various labs. When controlled by Gemini Robotics, ALOHA 2 becomes more of a generalist robot, in the sense that it can perform any number of tasks based on examples it’s been trained on. Ask it to pack a lunchbox and, as evidenced by a video of this exercise, it can use two pincer grippers to delicately place a piece of white bread into a Ziploc bag, close it, place a bunch of grapes in a Tupperware container, secure the lid, and then carefully move the items into a lunchbox before zipping it up.  It’s not a great lunch. But the fact that the robot can put it together represents an objective step forward from what was possible even, say, three years ago.  This is in large part due to AI and its impact on what are known as robot policies, which controls how a general-purpose robot will need to assess and understand its surroundings, plan how to move within them, and then perform its task correctly. The ALOHA 2 robot isn’t much more than two mechanical arms on a bench top, but it serves as a testbed for cutting-edge AI robotics models. These animations are based on human teleoperation of the robot arms, data that is used to train Google DeepMind’s models. (Video: Google DeepMind / Stanford University / Hoku Labs) Historically, these policies were based