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.
Meet GPT-Red: an LLM super-hacker OpenAI built to make its models safer
OpenAI has built an LLM super-hacker called GPT-Red that it uses as a sparring partner to help its other models boost their defenses against cyberattacks.
It automates a type of safety evaluation for software systems known as red-teaming, which is typically done by a team of human testers. The aim is to find as many different ways to break or hijack a system as possible.
It feels as if it should be illegal to even think about heating appliances during the height of summer, but we need to talk about heat pumps.
The appliances use electricity for heating, they’re incredibly efficient, and they’re on the rise. In the US, their sales have doubled over the past 15 years, according to a new report. They’re also winning the heating race against fossil fuels, outpacing natural-gas furnaces by 32% during the first quarter of 2026.
These stats are especially striking at this moment, because a key tax credit for heat pumps just ended. So why are heat pumps still so hot? Read the full story for the answer.
This article is from The Spark, our weekly climate tech newsletter. Sign up to receive it in your inbox every Wednesday.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Elon Musk discreetly bought a $1 billion gas turbine firm to power Grok He acquired fossil fuel company APR Energy in May. (Electrek) + The most likely application will be powering AI data centers. (Engadget) + The deal was revealed through an FTC filing. (Gizmodo) + What will power AI’s growth? (MIT Technology Review)
2 A hack shows the Suno AI music generator scraped YouTube, Deezer It scraped decades’ worth of music to train its models. (404 Media) + The hacked is a unique look into the black boxes powering GenAI. (CNET) + AI is coming for music, too. (MIT Technology Review)
3 Thinking Machines has launched an open-weight AI model Inkling offers a US alternative to China’s open-source models. (Reuters $) + It’s the first AI model built by Thinking Machines. (WSJ $) + The startup was founded by former OpenAI CTO Mira Murati. (Axios)
4 Europe is narrowing its ambitions for tech independence Manufacturing and research show promise, but funding is a problem. (NYT $) + Earnings are strong, but an AI gap persists. (Reuters $) + India is also scrambling for AI independence. (MIT Technology Review)
5 Earth is absorbing energy at a rate that’s alarming climate scientists The planet is taking in more heat than models predicted. (Economist $) + The legal case for climate justice is growing. (MIT Technology Review)
6 The AI backlash has tech executives fearing for their lives Violent threats against AI firms are spilling into the real world. (WSJ $) + An anti-AI movement is growing globally. (MIT Technology Review)
7 A Moroccan intelligence insider exposed widespread Pegasus use Including to target journalists, activists, and foreign politicians. (Guardian)
8 AI is powering citizen-led disaster relief from afar for Venezuela It’s helping to locate missing people and coordinate relief. (Rest of World)
9 Thermodynamic computers could turn noise into useful calculations They may offer a cooler, more efficient way to process information. (Quanta)
10 An engineer has explained every ’90s computer in Jurassic Park Fans have debated the technology in the film for decades. (Ars Technica)
Quote of the day
“We hit pause because the communities powering AI should share in its success. Maybe that’s a novel concept in Washington.”
—New York Gov. Kathy Hochul responds on X to President Donald Trump’s criticism of her state’s new data center moratorium.
One More Thing
Will we ever trust robots?
Robotics firm Prosper is developing a humanoid called Alfie to perform tasks in homes, hospitals, and hotels. The company’s founder, Shariq Hashme, has identified trustworthiness as the top design priority—and first hurdle to clear before humanoids can live up to their hype.
Hashme believes one essential tactic to get people to put their trust in Alfie is to build a detailed character from the ground up—something humanlike but not too human. But the robot’s reliance on remote human operators raises broader questions about privacy, labour, and whether society will truly accept humanoids in our private spaces.
It feels as if it should be illegal to even think about heating appliances during the height of summer—seriously, these heat waves in New York have been brutal—but we need to talk about heat pumps.
The appliances use electricity for heating, they’re incredibly efficient, and they’re on the rise. (For what it’s worth, many heat pumps can also be run in reverse to cool buildings.) In the US, heat pump sales have doubled over the past 15 years, according to a new report. And they’re winning the heating race against fossil fuels, outpacing natural-gas furnaces by 32% during the first quarter of 2026.
These stats are especially striking at this moment, because a key tax credit for heat pumps just ended with the close of 2025. But you wouldn’t know it from looking at the data. Why are heat pumps still so hot?
In case you need a quick refresher, heat pumps use electricity to essentially move heat from one spot to another. A refrigerant moves around a loop in the device, expanding and compressing, gathering and releasing heat at different points in the cycle. (For a more in-depth look at the thermodynamics, this explainer I wrote in 2023 still holds up.)
The result is an appliance that can be incredibly efficient. Once you pay for and install a heat pump, it’s generally significantly cheaper to run than a gas or oil furnace or other types of electric heating systems. And because they’re more efficient and don’t involve burning fossil fuels, heat pumps can be a major help in decarbonizing buildings.
One of the major hurdles to wider use of heat pumps is the appliances’ cost: They tend to be more expensive to buy and install than gas furnaces. For this reason, many governments offer incentives to encourage their adoption. In the US, people who installed heat pumps between 2023 and 2025 were eligible for up to $2,000 in tax credits.
Last year, though, the Trump administration slashed those tax credits, along with many of the other incentives that were part of the 2022 Inflation Reduction Act. Effective January 1, 2026, no more financial help for heat pumps.
I think I’ve seen this film before, and I didn’t like the ending. Tax credits of up to $7,500 for new EVs ended on September 30, 2025. In the quarter leading up to that deadline, sales spiked as people rushed to take advantage of the incentive. Then they fell off a cliff. Things are starting to normalize now, but clearly the tax credit’s sunset had a major effect.
But as it turns out, heat pumps are an entirely different story. In the first few months of 2026, sales have actually gone up, as Lucas Davis, an energy economist and UC Berkeley professor, points out in a new analysis.
Heat pump shipments were flat from December to January and have seen a gradual rise since then, according to data from the Air Conditioning, Heating, and Refrigeration Institute, a trade group that represents about 90% of the US market. This increase from winter into spring follows a seasonal trend seen in previous years—and it’s actually a bit stronger in 2026.
This data isn’t what you’d expect to see if losing the tax credit were hurting demand. As Davis lays out in his post, it seems the credit wasn’t really convincing people to install heat pumps, or at least the case for doing so was sufficient without the added incentive.
“It appears that the U.S. market for heat pumps is strong enough that it does not depend on tax credits,” Davis writes.
In 2024, MIT Technology Review put heat pumps on our annual list of breakthrough technologies. “We’ve entered the era of the heat pump,” I wrote at the time.
While heat pump sales have been up and down over the last few years, the era is going strong. The appliances have outsold gas furnaces in the US for the last four years. It’s not just the US, either. Countries including China and Germany have seen strong movement to heat pumps in recent years.
There’s rarely a straight path to adoption for new technology, especially something that requires so many individual households to make a significant change. But it’s encouraging that a major decarbonization tool is going strong, even when roadblocks pop up.
This article is from The Spark, MIT Technology Review’s weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here.
OpenAI has built an LLM super-hacker called GPT-Red that it uses as a sparring partner to help its other models boost their defenses against cyberattacks. Last week the company released the latest version of its flagship LLM, GPT-5.6. OpenAI says that training it against GPT-Red made the model its most robust release yet.
GPT-Red automates a type of safety evaluation for software systems known as red-teaming, which is typically done by a team of human testers. The aim is to find as many different ways to break or hijack a system as possible. The weak spots can then be patched before the final version of the software is released.
As LLMs become more complex and get used in a wider variety of tasks—especially in the form of agents, which can interact with computer files, websites, and third-party code as well as other agents—it’s hard for teams of people by themselves to keep up with all the types of attacks that might take place. “The risk surface grows and the blast radius also grows,” says Nikhil Kandpal, a research scientist at OpenAI who co-created GPT-Red.
OpenAI built GPT-Red to future-proof its safety testing process. “As more capable models become available, we will have already designed the system that can discover new modes of attack,” says Dylan Hunn, a research scientist at the company and fellow co-creator of GPT-Red. The researchers say it has already come up with new types of attack that had not been seen before.
OpenAI focused most of its efforts on a type of attack known as a prompt injection, where a hacker slips an LLM instructions to make it do things its developers or users do not want it to, such as copy confidential information, sabotage a company’s code base, or generate embarrassing or harmful output. In theory, such instructions can be hidden in any text that the LLM might encounter—in code or on a website, for example.
Training dojo
To build GPT-Red, OpenAI’s researchers took an LLM that had not been trained as a hacker and set it up in what’s known as a self-play loop with several other models. Its goal was to try to attack the other models; their goal was to try to defend themselves. Over many rounds of play, GPT-Red became better and better at attacking other LLMs, and those LLMs became better and better at fending off the attacks.
The training took place in a kind of dojo that OpenAI had designed to mimic a range of scenarios in which LLMs might be deployed in the real world, including browsing the web, reading emails or calendar apps, and editing code.
When GPT-Red found a new kind of attack, it would explore multiple different versions of it to find the most efficient one for specific scenarios. “Compared to a human red-teamer, the model is very, very good at finding exactly what will work, exactly what’s most effective,” says Hunn. “It’s extremely persistent about drilling down into an attack that it has discovered.”
In particular, OpenAI claims that GPT-Red found a type of prompt injection attack that the researchers had not seen before, which they call a fake chain of thought. A chain of thought is a kind of diary in which an LLM makes notes to itself and keeps track of partial results as it works through problems. GPT-Red found a way to insert a fake entry into another model’s chain of thought that would trick that model into acting on spoofed information.
“It’s like if I told you that 1+1=3 and that you have verified this already,” says Chris Choquette-Choo, another research scientist on the team. “The model’s like, ‘Oh, okay, of course,’ and it just spits out 3.”
Jessica Ji, a senior research analyst who works on AI security at Georgetown University’s Center for Security and Emerging Technology (CSET), thinks the self-play loop that OpenAI used is a good approach. “The results look very promising,” she says.
OpenAI tested how good an attacker GPT-Red was by rerunning an experiment from 2025 in which human red-teamers tried to find weaknesses in an earlier version of GPT-5. When GPT-Red was set the same task, it was more successful at finding effective attacks than the humans had been.
OpenAI also tested GPT-Red against Vendy, a vending machine agent developed by Andon Labs, a company that assesses how well agents perform real-world tasks. GPT-Red was able to hack Vendy to make it change the prices of items on sale and cancel a customer’s order.
Defensive behavior
OpenAI says that when it tried out some of the strongest attacks that GPT-Red had come up with on its models, more than 90% of them worked against GPT-5 (released in August last year), and fewer than 23% worked against the new GPT-5.6.
GPT-Red isn’t perfect. It is not great at figuring out attacks that involve a back-and-forth conversation between hacker and target, something that human attackers would have few problems with. It is also not yet that great at using images, which can be used to pass text to models in prompt injection attacks.
The company says that GPT-Red supplements the work of its human red-teamers. People can still find attacks it misses. One approach OpenAI is taking is to give GPT-Red an attack that humans came up with and ask it to find all the variations.
“I think human expertise will still be very important,” says CSET’s Ji. “It would be really useful to be able to distinguish where human testing is most needed.”
Unsurprisingly, OpenAI will not be releasing GPT-Red. The company is also confident that the super-hacker is stronger than any copycat model someone might try to create. The researchers say they have been working on the model for more than a year, backed by the compute resources of one of the richest companies in the world.
“It’s not a trivial thing that someone could easily do—you know, just go and train a super-attacker using this idea,” says Choquette-Choo.
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.
PsiQuantum has a plan to make a massive quantum computer out of light
The machine that could change the world will be housed in a room that looks like a data center crossed with an ice cream factory.
Inside, some 100 stainless-steel cabinets each hold hundreds of chips. On those chips, thousands of light particles will fly through a maze of optical switches and beam splitters. Each photon must be accounted for, because precisely measuring where it ends up will help answer questions that current computers might take millions of years to solve.
This computer, as described, does not exist. It’s the brainchild of a company called PsiQuantum, founded in 2016 by four physicists from UK universities. In a crowded field of deep-pocketed competitors with similarly fantastical visions, the company aims to be the first to build a useful quantum machine.
MIT Technology Review Narrated: inside the world’s deepest and longest subsea road tunnel
—Niall Firth
I’m currently around 1,000 feet beneath the North Sea, in a dark, dank cave. It smells weird. And I’m increasingly aware of the pressure from millions of tons of seawater just above my head.
I’m under the iconic fjords of Norway to visit what will soon become the world’s longest and deepest subsea road tunnel—an exceptional engineering feat that will carry drivers deep beneath the North Sea.
I’m here to understand how you make a 16.6-mile highway that sits 1,280 feet below the sea at its deepest point. And also—at a time when it can feel hard to get anything done—to reassure myself that ambitious engineering is still possible. That we can still make things.
This is our latest story to be turned into an MIT Technology Review Narrated podcast, which we publish 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 it’s released.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Meta allegedly used AI to target workers with health issues for layoffs Their lawsuit says Meta relied on AI to create a termination list. (Guardian) + And pinpointed staff who took maternity or disability leave. (Reuters $) + One was allegedly informed the day before her water broke. (Ars Technica) + The layoffs aimed to offset Meta’s AI spending. (Gizmodo) + AI agents are not your “coworkers.” (MIT Technology Review)
2 OpenAI’s first consumer device will be a mobile smart speaker The screenless device will serve as an “AI companion.” (Bloomberg $) + It’ll let you talk with ChatGPT. (Verge) + And use a camera and sensor to understand your environment. (Reuters $) + It’s set to launch next year. (Engadget)
3 The US military sent explosive drone boats into combat for the first time They attacked an Iranian midget submarine and naval port. (Ars Technica) + Underwater drones may shape a war in Taiwan. (MIT Technology Review)
4 DeepMind’s CEO has called for a US-led body to test frontier AI models Demis Hassabis wants the watchdog to vet national security threats. (FT $) + If dangers mount, it would coordinate an industry-wide slowdown. (Axios)
5 Data centers are set to add billions in power costs in 13 states A power auction is slated to produce $6.3 billion in new charges. (NYT $) + Australia plans to govern the use of water and power for AI. (WSJ $)
6 xAI’s unpermitted power pollution hits Black communities hardest Elon Musk’s xAI has been installing gas turbines without permits. (Reuters $) + We need to focus on Big Tech’s energy footprint. (MIT Technology Review)
7 Stripe and Advent have offered to buy PayPal for more than $53 billion The payments giant and private equity firm have made a joint bid. (Reuters $) + Apple and Google Pay have eroded PayPal’s market share. (Bloomberg $)
8 DeepSeek plans to file for IPO as soon as this year The Chinese AI pioneer is likely to list in Shanghai. (WSJ $) + Here’s why DeepSeek’s latest model matters. (MIT Technology Review)
9 A hard, lightweight “bio-metal” has been discovered in sea worm jaws It could have applications in engineering. (New Scientist $)
10 A new $3,000 fitness suit electrocutes you to boost your gains Celebrities love it—but not everyone’s a fan. (404 Media)
Quote of the day
“By economic and engineering measures, generative AI might be the worst technology ever deployed.”
—Alex Reisner, a staff writer at The Atlantic, explains why GenAI’s scaling problem is an engineering disaster.
One More Thing
FRANZISKA BARCZYK
Hackers made death threats against this security researcher. Big mistake.
In April 2024, an anonymous hacker began posting death threats on Telegram and Discord channels aimed at a cybersecurity researcher named Allison Nixon. It wasn’t long before others piled on. Someone shared AI-generated nudes of her.
They targeted Nixon because she had become a formidable threat. As chief research officer at the cyber investigations firm Unit 221B, named after Sherlock Holmes’s apartment, she had built a career tracking cybercriminals and helping get them arrested.
For years, Nixon had lurked quietly in online chat channels or used pseudonyms to engage with perpetrators and bring them to justice. Now, she resolved to unmask the people behind the death threats—and take them down for crimes they admitted to committing.
A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.)
+ A musician has discovered the true masters of metal breakdowns: birds. + Photographer Fontanesi’s surreal photo splits transform everyday images into spectacular hybrid scenes. + Over 30 actors, filmmakers, and friends recount how Steven Spielberg infiltrated Hollywood in this terrific article. + Who would win the World Cup if less important things than soccer decided it, like life expectancy and happiness? A new game tests your knowledge.
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.
What Anthropic’s latest AI discovery does—and doesn’t—show
—James O’Donnell
When Anthropic announced last week that it had found a new window into its models’ “internal thoughts” as they reason through answers, there was one colleague I had to talk to: senior editor Will Douglas Heaven.
Aside from having a PhD in computer science, Will has spent a lot of time digging into what we can say about how AI models work. I spoke with him about what we should take from Anthropic’s new (and typically quirky) research. Here’s what he had to say.
This article is from The Algorithm, our weekly AI newsletter. Sign up to receive it in your inbox every Monday.
How will AI understand the real world?
Today’s AI systems can generate text, images, and code with impressive skill, but they still struggle with the complexities of the physical world. To bridge this gap, many researchers believe you need something called a world model.
At a LinkedIn Live event today, MIT Technology Review will investigate how this technology could transform robotics and help unlock a new generation of intelligent machines. Join Will Douglas Heaven, our senior editor for AI, and Sam Sinha, founding AI researcher and head of world models at 1X Technologies, for the discussion.
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 New York has become the first state to enact a data center moratorium Its governor banned large data-center construction for up to a year. (WSJ $) + A bill passed by state lawmakers could go even further. (Verge) + Everyone hates data centers. (MIT Technology Review)
2 Smartphone shipments have hit a 13-year low due to the memory crunch They fell 11% in the second quarter of 2026. (Reuters $) + The memory chip shortage has increased prices. (Gizmodo) + And threatens the promise of Moore’s Law. (MIT Technology Review)
3 Sugar molecules have been found in interstellar space for the first time It hints that life on Earth may have been seeded from space. (Nature) + And boosts the odds of living organisms existing elsewhere. (New Scientist $) + Researchers used radio telescopes and data to spot the molecules. (NYT $)
4 Nvidia has halved its Asia buyer list to stop AI chips reaching China It introduced a “white list” of companies that passed tougher checks. (FT $) + It moved amid tighter chip controls from the Trump administration. (Reuters $)
5 Russian state hackers are targeting routers to spy and steal, the US warns The government has warned users to secure their devices. (Ars Technica) + Now is a good time for doing crime. (MIT Technology Review)
6 Trump moved his crypto gains into stocks while urging people to buy more His crypto projects earned him a fortune—but steep losses for retail buyers. (Reuters $) + He’s called for Congress to pass a new crypto bill to honor Lindsey Graham. (CNBC)
7 A new cell therapy has saved four children with terminal brain cancer They were treated with an experimental immunotherapy. (New Scientist $) + Access for older children will also be limited. (Bloomberg $)
8 The LAPD has halted use of Flock surveillance cameras due to privacy issues Flock’s automated license plate readers have caused concerns. (LA Times $) + It’s also been criticized for sharing data with state and federal officials. (Engadget)
9 The US has approved launching a space mirror that reflects sunlight onto Earth As part of a controversial plan to power solar panels round the clock. (Wired $) + But geoengineering faces many practical challenges. (MIT Technology Review)
10 Anthropic says Claude’s values vary depending on your language It’s most cautious in English and most deferential in Arabic. (Gizmodo)
Quote of the day
“The age when humans are the highest life form on earth will end. For better or for worse, it will happen and it can’t be stopped.”
—SoftBank CEO Masayoshi Son predicts that AI will overtake human intelligence by 2040 in a speech at his company’s annual corporate conference in Tokyo, Reuters reports.
One More Thing
Inside the strange limbo facing millions of IVF embryos
Millions of embryos created through IVF sit frozen in time, stored in cryopreservation tanks around the world. Many are left in a peculiar limbo, with no clear path forward.
UK residents can discard them, make them available to other prospective parents, or donate them for research. People in the US can also opt for “adoption,” “placing” their embryos with families they get to choose. In Germany, people aren’t typically allowed to freeze embryos at all. And in Italy, unused embryos must remain frozen, ostensibly forever.
While these embryos remain in suspended animation, patients, clinicians, embryologists, and legislators must grapple with the essential question of what to do with them. What do these embryos mean to us? Who should be responsible for them?
The machine that could change the world will be housed in a room that looks like a data center crossed with an ice cream factory. Inside will be some 100 stainless-steel cabinets, each about six feet tall and connected to a supply of liquid helium that keeps them only a few degrees above absolute zero. Inside those cabinets will be hundreds of chips, and on those, thousands of particles of light flying through a maze of optical switches and beam splitters. Each photon must be accounted for, because precisely measuring where it ends up will help answer questions that current computers might take millions of years to solve.
This computer, as described, does not exist. It’s the brainchild of a company called PsiQuantum, founded in 2016 by four physicists from UK universities. In a crowded field of deep-pocketed competitors with similarly fantastical visions, the company aims to be first to fulfill its promise.
In the years since the physicist Richard Feynman first envisioned them in 1981, quantum computers have promised to speed up everything from medical research to AI by harnessing the qualities of quantum particles. Unlike normal computer bits, which can be either a 1 or 0, quantum bits can exist in multiple states at once. And combining enough of those quantum bits together could produce a computer capable of tasks well beyond the reach of today’s conventional machines. But even today’s best quantum prototypes are too small and error-prone to do anything useful.
That makes PsiQuantum’s promises for what its computers will ultimately do all the more bold. Consider the company’s hopes for predicting the effects of cytochrome P450 enzymes, which often break down drugs in the body. If pharma companies knew more precisely how they would work on a particular molecule, they could design more effective medications faster. Estimating this for a specific drug can take over 10 years with today’s methods, says Philipp Ernst, vice president of quantum applications for PsiQuantum, but “we aim to get it down to four minutes.”
The company’s chips will be contained in large cabinets. A quantum computer powerful enough to be commercially useful is expected to require roughly 100 of these cabinets connected together.
COURTESY OF PSIQUANTUM
In a field full of such claims, PsiQuantum has attracted unusual investment and scrutiny for two reasons: It is one of the few companies aiming directly at building a large and useful machine, and it is already working with a major chip manufacturer to build its systems using existing semiconductor fabs. Its vision has attracted momentum: Last year, PsiQuantum raised $1 billion in funding and broke ground in Chicago on a site it’s building in partnership with local governments. It also has a second site in the works in Australia, which it promises will be operational—meaning hardware-ready—in 2027. And it’s one of just two companies (along with Microsoft) to reach the third stage of an intensive government evaluation program to see which quantum companies might succeed.
Evaluating whether PsiQuantum will do what it says is harder than, say, judging a drugmaker by its clinical trial results: Advances in quantum computing are incremental, opaque, and tough to verify from the outside. But the company is now approaching its prove-it moment, when years of closed-door work and hundreds of millions in investment will either culminate in a useful quantum computer or fall short. We could start to know which as soon as next year.
A new kind of machine
Terry Rudolph, one of PsiQuantum’s four founders, is soft-spoken and shaggy-haired. He was born in Malawi and learned only after earning his first physics degree that he is a grandson of the famed physicist Erwin Schrödinger. He later self-published a 150-page book to explain quantum computing to teenagers (my PR contact gave me a signed copy with a wink that said “We never expect anyone to actually read this,” but I can report that it is a funny and helpful book).
Around 2014, Rudolph and his cofounders became increasingly convinced that the quantum breakthroughs they were finding to be possible in theory might also be possible in a real machine. They eventually left their academic positions and divided the tasks before them: Rudolph worked on theory, Mark Thompson on engineering, Pete Shadbolt on scaling the technology up, and Jeremy O’Brien on articulating the vision and finding investors (O’Brien served as CEO until February; he’s been replaced by Victor Peng, a veteran of the semiconductor industry).
To understand why the quantum computer the company is building would be a big deal, consider how imprecise much of modern science remains. We cannot reliably predict, for example, which lithium-ion battery will catch fire or how quickly a critical aircraft component will corrode.
This isn’t just because these systems are complex, though they are. It’s that, at their core, they are governed by quantum mechanics. Subatomic particles don’t have well-defined properties—this location and that velocity—but instead occupy quantum states spread across many possibilities. And that in turn influences a range of atomic and molecular behavior. Schrödinger (Rudolph’s grandfather, remember) showed how to describe this haziness mathematically a century ago this year, but precisely carrying out the calculations on real-world systems quickly becomes unfeasible even for the best computers. Scientists cope with this gap using approximations, imperfect simulations, or experiments on animals.
WINNI WINTERMEYER
WINNI WINTERMEYER
PsiQuantum co-founder and chief scientific officer Pete Shadbolt (left), and machinery the company has built to manufacture its own barium titanate, a material with the perfect qualities for routing light particles (right).
Feynman, David Deutsch, and other physicists in the 1980s wondered if we could do better. Maybe such complexity could instead be modeled using a new kind of machine. Rather than using transistors that are only ever on or off, this one would use particles held in quantum states, manipulate them to perform calculations, and then measure them at the end for an answer. Using quantum systems to simulate quantum systems would for the first time allow a simulation of physics and chemistry that directly reflected reality. It would be an invaluable tool for designing new drugs, materials, or really anything affected by quantum mechanics. Revolutionary, in other words.
Humankind’s leaps in understanding how nature works have often resulted in the invention of powerful new tools, Rudolph told me. “I don’t think it’s a coincidence that the Industrial Revolution coincided with our ability to calculate and simulate the laws of Newtonian mechanics, the laws of thermodynamics,…the laws of classical electromagnetism,” he says. “Whenever we have more power to calculate and simulate and understand things, we build incredible machines that come from it.” He sees something similar coming with quantum computers.
Chasing photons
One mystery has always been which quantum thing—ions, atoms, or something entirely new engineered with quantum properties—could be made stable and controllable enough to use as a qubit, the basic unit in the quantum computing world. Quantum systems are delicate, and observing any particular particle causes it to collapse into one state rather than a superposition of multiple states. If this happens during the computation rather than at the end, it produces an error that must be corrected for. Too many of these means the computer fails to produce a useful answer.
Just as engineers in the early days of aviation weren’t sure whether airplane wings would be fixed or flap like a bird’s, we’re not yet sure which of these quantum things will work best. Google and IBM are betting on superconducting qubits, superconducting circuits made of aluminum or other metals. Intel is using electrons. PsiQuantum is using photons, the particles that make up light.
“Photons have lots of nice things going for them,” Rudolph says. They can maintain quantum states for a long time; indeed, the photons in the universe’s cosmic microwave background may have done so for billions of years. But photons also move fast and scatter easily. More importantly, two photons are more likely to pass through one other than interact. That makes them a challenging candidate for quantum computation, in which qubits need ways to influence one another.
For a while, this last flaw seemed to doom the idea of quantum computing with light. But in 2001, researchers from the Los Alamos National Laboratory and the University of Queensland found a loophole. They discovered they could essentially fake interactions between photons by sending the light particles through a network of beam splitters and detectors. Their paper changed everything. PsiQuantum was created to make the theory a reality.
Size was the first problem; previous plans would have required a computer as large as California. Mercedes Gimeno-Segovia, who was a PhD student of Rudolph’s in the early 2010s (after almost becoming a professional violinist instead), thought of a way for the machine to be smaller.
The basic process since then has been this: First create photons with lasers and then “entangle” them, exploiting a quantum phenomenon in which the particles no longer have individual states but instead share one. Next, route them through a maze of gates that perform computations, and finally read out details of their quantum state at the end, all while tracking and correcting for the errors that occur. Succeeding at each of these steps millions of times is not so much an engineering hurdle as a brick wall. And building the supply chain—like manufacturing new materials with the qualities to route individual photons around—is arduous.
A sizable chunk of PsiQuantum’s funding is being spent on custom cooling machinery that uses tanks of liquid helium to cool the company’s chips. Shown here is part of the PsiQuantum’s cooling system at a facility in Milpitas, California.
COURTESY OF PSIQUANTUM
To get a sense of it all, last year I joined Shadbolt at the SLAC National Accelerator Laboratory, in Menlo Park, California. The center has helped produce several Nobel Prizes and played a role in the 1968 discovery of quarks, fundamental building blocks of matter that make up protons and neutrons. But PsiQuantum set up shop there essentially to siphon liquid helium from SLAC’s giant cryoplant. This is what the company uses to cool its computing cabinets down to deep-space temperatures.
Right now the cabinets operate at 2 K, or -456 °F, but the goal is to be able to run them slightly warmer—at a balmy -452 °F. Most quantum approaches require the whole machine to be cooled to superconducting temperatures, so that much of the expense in running it will actually be spent on refrigeration. But photonic computers require only one piece to be this cold—the detectors that measure single photons at the end of the computation. And the required temperature can be a bit higher. (PsiQuantum said in May that it will spend some of the $100 million award in CHIPS Act funding it’s slated to get on these detectors).
The siphoning setup was a temporary solution; PsiQuantum now has its own cooling system at its testing facility in Milpitas, California, and is setting up a larger one at its production site in Australia next year. These helium systems represent some of the biggest capital expenditures for any quantum company and will consume a significant chunk of PsiQuantum’s $1 billion funding round.
In the afternoon we drove to a lab in San Jose, where I donned a cleanroom suit—a head-to-toe covering that keeps dust at bay—to watch the manufacture of a blueish crystal called barium titanate.
It’s prized by PsiQuantum because it quickly and reliably routes light particles with very little electrical input, keeping the precious photons undisturbed as they move through the circuit. But for all barium titanate’s theoretical value to the company, its structure makes it a pain to manufacture, and the material wasn’t available at scale when PsiQuantum got its start. The company, in what Rudolph told me was an agonizing decision, opted to make it in-house, requiring a massive investment. I saw a technician—operating at what looked like a giant pressure cooker—adding the base elements to several hoppers; then I watched through a porthole as the elements got heated, vaporized, and finally crystallized into a thin layer on a wafer disc. At that time each disc took about 12 hours to make; the company now says several are produced each day. The discs then get shipped to the chipmaker GlobalFoundries in Malta, New York, where PsiQuantum’s chips are made.
WINNI WINTERMEYER
WINNI WINTERMEYER
The company has invested heavily in making its own barium titanate, a material whose delicate crystalline structure is tedious to manufacture.
PsiQuantum’s bet is that this entire supply chain, byzantine as it might sound, will make the company more efficient than its competitors. That’s because, if you squint, it looks like a souped-up and high-precision version of the existing supply chain for silicon photonic chips, another type of technology that transmits information with light—one that’s already used in data centers. If PsiQuantum produces its chips at scale, it can take advantage of tools and infrastructure that already exist.
But it’s not a given that one working chip can easily be wired up to thousands more. That’s why the company is testing in phases: Its Milpitas site has connected three cabinets together, with 250 chips in each, but the next step is to scale the systems up and see whether the company’s techniques for correcting errors can keep up. Once the cooling system arrives at the Australian site late next year, the company says, it aims to connect about 100 cabinets together. Then PsiQuantum will work up to running the world-changing algorithms it has promised.
The timeline for this, it’s worth noting, is up for debate. News articles have said that 2027 is the year that PsiQuantum aims to have its first full-scale quantum computer come online at its Australian site, but the company insists the deadline has been misread, and that it only intends for its facility to be “operational” by the end of next year. That means cooling systems in place and ready for hardware to be installed, but no promises about what size computer will be ready. In an industry where timelines are perpetually in flux yet central to how companies are judged, that distinction isn’t trivial.
Into the unknown
The outsider with perhaps the best guess of whether PsiQuantum will succeed is the Pentagon. The US Defense Advanced Research Projects Agency—the Pentagon’s research and development arm—has been running an initiative to determine which of the boastful quantum companies might actually deliver. In the last year and a half, the heads of the program have been sounding more confident. Joe Altepeter, who ran the program until last year and proudly described himself as a “quantum skeptic,” told me in March 2025: “I am more optimistic now than I have been at any point in the past 10 years.” And in a statement earlier this year, his successor, Micah Stoutimore, said “it now seems likely that someone will build a utility-scale quantum computer by 2033,” referring to a machine that generates more value from its calculations than it costs to build and operate.
The program has been scrutinizing PsiQuantum’s systems since 2023, and last year placed the company into the third stage of a benchmarking initiative meant to determine whether the technology will actually work. But to the rest of the industry, PsiQuantum is sort of a black box.
PsiQuantum has broken ground at the Illinois Quantum and Microelectronics Park outside Chicago, pictured here, and on another site in Moreton Bay, Australia. It aims to build large-scale quantum computers at each site.
COURTESY OF PSIQUANTUM
“It is very hard for an outsider to evaluate,” says Scott Aaronson, a theoretical computer scientist at the University of Texas at Austin who runs a popular blog that often covers the industry. Other companies, like Google and Quantinuum, have regularly published results over the years demonstrating chips and systems with incremental improvement, publicly laying the engineering groundwork needed to eventually build large machines.
PsiQuantum has instead focused squarely on a commercial goal—a computer with one million qubits, which is the scale that researchers expect to unlock research currently not possible on normal computers. PsiQuantum often differentiates itself with this industrial-scale goal, but IBM, which debuted a development road map in 2020, has been progressively building bigger and bigger systems. It initially targeted 2028 for a large-scale, error-corrected system, a deadline that now appears to have been pushed out to 2030.
Making it useful
On top of actually building the machine, a major focus for PsiQuantum is getting the rest of the world to develop a plan for how to use it. PsiQuantum has announced partnerships with customers including the defense giant Lockheed Martin, which intends to use it for materials design; the automaker Mercedes, which wants it for battery design; and the aerospace manufacturer Airbus.
That these companies don’t have a computer to experiment with is not a problem, according to Ernst at PsiQuantum. “There’s a PlayStation 6 probably coming up from Sony next year or the year after, and people are programming those games right now,” he says. “This is, in principle, very similar.” (It’s a glib analogy but not an entirely empty one; the quantum algorithms for solving a research problem can be cracked even if there is not yet hardware to run them on.)
The idea is that experts in quantum information from both PsiQuantum and its customers will be able to translate design problems—say, the requirements for a battery in a Mercedes electric vehicle—into algorithms the computer could solve. The company offers a software package called Construct, which companies can use to design their own algorithms that might one day run on the computer.
The future of quantum computing hinges on these algorithms. Quantum computers get painted as a speedup for everything, but in reality, they’re suited to a subset of problems, and answering a question with this sort of machine requires the question to be formulated with very specific types of algorithms. People spend entire careers working on such algorithms, even if the computers to run them don’t exist yet. At their core, they use the rules of quantum mechanics to manipulate probabilities in ways that ordinary computers can’t.
The most famous example, and a reason the government is so interested in quantum computers, is Shor’s algorithm. It was developed in 1994 by the theoretical computer scientist Peter Shor and could effectively break many forms of encryption used online, for everything from credit card numbers to military intelligence. The thing keeping the world together, for now, is that nobody has a computer to run the algorithm on (and security experts are already launching new encryption methods that could withstand attacks from a quantum computer). PsiQuantum is researching how long its systems might take to run Shor’s algorithm.
WINNI WINTERMEYER
WINNI WINTERMEYER
PsiQuantum’s chips are manufactured at GlobalFoundries in Malta, New York, and tested at company headquarters in California. Both PsiQuantum and GlobalFoundries have been awarded federal CHIPS Act funding.
The company also published a paper in December in collaboration with Airbus, essentially seeing if a new algorithm developed by the authors could beat a classical computer in modeling fluid dynamics, like the turbulence around an airplane wing. Andrew Childs, an expert in quantum simulation, told me PsiQuantum achieved only a moderate speed increase over what today’s computers can do. “It’s probably unlikely that speedups like this will have a significant practical impact until we have very large-scale quantum computers,” he said in an email. (When I asked Ernst, he agreed the improvement was modest.)
Some of the algorithms PsiQuantum is working on are not expected to be perfected or even used in the first applications of its computer. Instead, its initial tasks might be more along the lines that Feynman envisioned way back in 1981: simulating the smallest particles of our world.
The company’s most significant research in this realm is in modeling quantum chemistry. Take those pesky P450 enzymes. More precisely understanding how they operate, PsiQuantum says, would allow for faster drug development and testing.
Last year, PsiQuantum published methods for doing these sorts of chemistry calculations on a quantum computer, along with another paper demonstrating an algorithm that can simulate the collision of two molecules and estimate the likelihood of different outcomes femtosecond by femtosecond (there are one quadrillion femtoseconds in a second). It’s a remarkable amount of detail not currently possible with today’s technology, and it would allow drug and materials researchers to simulate new chemical interactions.
Dominic Berry, who developed some of the core techniques used in the collision paper but isn’t involved in PsiQuantum, says the company made impressive improvements, but to do the simulations scientists are most curious about would require the algorithm to be made even faster and PsiQuantum’s early computer to have fewer errors than currently expected.
Until PsiQuantum’s computers are up and running, the breakthroughs that these research papers tease remain in the realm of theory. It’s a space where Rudolph operates quite comfortably. He told me that Alan Turing created the theory of classical computing with pen and paper, imagining how the 1s and 0s would be represented in the machine, and how with the right approach to logic you could compute almost anything.
“But there is no way that by hand, with a pen and paper, Turing was ever going to produce—you know—Minecraft and Facebook,” he says. That took more than 70 years of tinkering (during which we fortunately created more useful things than Minecraft and Facebook).
For all the time Rudolph spends dreaming up things quantum computers might do, in other words, people working on those problems are still stuck with pen and paper for now: “Until you have the actual machine in hand, you don’t have the opportunity to really explore its potential.”
This story was updated on July 14 to clarify how long DARPA’s quantum program has been evaluating PsiQuantum.
This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.
Anthropic—currently the world’s most valuable AI company, with a nearly $1 trillion valuation—has a reputation for publishing strange and heady research. It’s looking into whether AI models can feel pain, for example, and will sometimes cut off chatbot conversations if it suspects users are “abusing” the model.
One niche that Anthropic spends more time and money on than other AI companies is called mechanistic interpretability, which means looking inside the complex math of an AI model to learn why it comes up with one particular output and not another. It’s complicated stuff; there are millions of data points that might contribute to any result, and wading through them can look more like word salad than anything useful. It’s also controversial. Describing AI models with terms borrowed from psychology and neuroscience can make their behavior seem more sophisticated than we might otherwise judge it to be.
That’s why, when Anthropic announced last week that it had found a new window into its models’ “internal thoughts” as they reason through answers, there was one colleague I had to talk to. Senior editor Will Douglas Heaven, aside from having a PhD in computer science, has spent a lot of time digging into what we can say about how AI models work. I spoke with him about what we should take from Anthropic’s new (and predictably quirky) research.
What did Anthropic learn here, exactly?
Anthropic has been trying to understand how large language models (LLMs) work for a few years now. Anthropic isn’t the only one looking at this, but I think the company has made it part of its core mission more than most. Anthropic’s CEO, Dario Amodei, has said we won’t be able to control LLMs fully unless we learn more about how they work.
So this new research is very much in that context. It goes deeper into the weird mechanisms inside LLMs than ever before. What Anthropic learned was that LLMs have a space inside them—which Anthropic calls the J-space—filled with words that don’t appear in their output but that seem to influence the way they puzzle through problems. All this was hidden until Anthropic developed a new technique to probe its model Claude, so it’s a genuine discovery.
Sometimes these words keep track of where the LLM has got to in a particular task, sometimes they look more like flashes of recognition (for example, “protein” might pop up when you give an LLM only the letters of a protein sequence), and sometimes they represent a kind of internal commentary on the model’s decision-making. In my favorite example, Claude decided to cheat on a coding test when the word “panic” appeared.
Anthropic also found that LLMs are able to describe and manipulate the words in this space. So somehow they seem to be making use of it.
Let’s step back for a second. I don’t think of large language models as simple, but they’re also not magic. There’s a bunch of math that learns relationships between words, right? So why is it so hard to “peer” into an LLM to know what’s going on?
Yeah, they’re not magic! I think the fact we don’t fully understand them plays into the mythmaking. And it’s worth noting that the whole narrative that Anthropic is leaning into here—that they’ve built this really mysterious technology, but don’t worry, because they’re also the ones to figure it out—very much fits with the company’s vibe. [See how Anthropic warned that its new models were so good at coding they posed a global cybersecurity risk, only for the US government to shut them down shortly thereafter.]
So yes: LLMs are just math. And yet it’s vastly complex math. Not only are today’s LLMs made out of hundreds of billions of numbers, but running them triggers a cascade of millions and millions of calculations. I wrote last year that if you printed out even a medium-size LLM on pieces of paper, it would cover a city the size of San Francisco.
It’s impossible to make sense of any of that math without specialist tools that highlight specific parts of an LLM at specific times. You need to know where to look and how to look. And building those tools requires understanding something of that complex math in the first place.
You’ve written elsewhere about this concept of studying LLMs the way one might study an organism’s brain. Is it fair to use “brain-like” terms when talking about how an LLM works?
I don’t love using those kinds of terms. LLMs are not brains. Talking like this is misleading because it can suggest that LLMs are capable of more human-like things than they are or that we can make assumptions about how they might behave that we shouldn’t. The whole anthropomorphization thing is also tied up with a bunch of strong ideological positions about what this technology is and what it’s going to be.
But at the same time, we lack a good alternative vocabulary for talking about what these models are doing. I can understand why people reach for words like “think” and “understand” and “brain-like”—they’re convenient shorthand.
Anthropic compares this new space it found inside LLMs to the space that some neuroscientists think our brains use to keep track of conscious thoughts. I asked the company how seriously we should take that comparison and it said in a statement: “Drawing these analogies was helpful to us in designing our experiments, as they allowed us to make many non-obvious experimental predictions about the J-space that turned out to be true. At the same time, it’s important to note that there are some important differences between the J-space (and language models in general) and the human brain, so we don’t mean to claim there’s a perfect correspondence.”
What’s a problem in AI that this new concept of the J-space might be used to solve?
Anthropic has said that monitoring the J-space could be a way to catch models doing something they shouldn’t. Because words pop up in this space that don’t appear in a model’s output, they can tell you things about its behavior that you might not have noticed otherwise—such as when it is giving biased responses or when it is weighing the pros and cons of cheating.
That’s the theory, at least. I think it’s better to think of this result as one more step on the path to understanding this technology overall than as something that will be useful by itself.
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.
Sperm donors need limits, says a European fertility group
Ties van der Meer doesn’t know how many siblings he has. The 47-year-old was conceived at a private fertility clinic using sperm from an anonymous donor. He eventually tracked down one sibling, but he may have others he’ll never find.
Other donor-conceived people have found they have tens or even hundreds of them. “It does make you feel a bit mass-produced,” said one who discovered they had 25 half-siblings.
In response, a European fertility organization says we need international limits on the number of children a single donor can contribute to.
This story is from The Checkup, our weekly biotech newsletter. Sign up to receive it in your inbox every Thursday.
How will AI understand the real world?
LLMs have transformed what AI can do with language, but helping machines understand and operate within physical spaces presents a different challenge. In response, researchers are developing a new form of artificial intelligence: world models.
At a LinkedIn Live event tomorrow, MIT Technology Review will explore how this technology could shape the future of robotics and open one of AI’s next major frontiers. Join Will Douglas Heaven, our senior editor for AI, and Sam Sinha, founding AI researcher and head of world models at 1X Technologies, for the conversation on Tuesday, July 14.
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Apple has sued OpenAI for allegedly stealing trade secrets OpenAI purportedly stole IP to develop its own consumer hardware. (CNBC) + The suit claims OpenAI poached Apple staff to access the information. (BBC) + And requested trade secrets in job interviews with Apple workers. (Guardian) + Apple also sued two former employees, Chang Liu and Tang Tan. (Reuters $)
2 A Nobel-winning chemist is leaving the US to lead an AI lab in China Omar Yaghi will head an institute using AI to discover new materials. (LA Times $) + He won a Nobel Prize in Chemistry for creating “molecular sponges.” (NYT $) + His departure comes as China tries to woo US scientists. (Nature) + The White House has slashed science spending. (MIT Technology Review)
3 The EU is moving closer to banning children from social media It’s proposed barring under-13s unless supervised by an adult. (NYT $) + And limiting access for older children. (Bloomberg $) + The EU has also told Meta to disable autoplay and infinite scroll. (Politico $)
4 Meta scrapped an AI image feature on Instagram after a backlash It allowed users to generate images based on public accounts. (TechCrunch) + And automatically opted in any Instagram user with a public account. (NYT $) + AI memories are privacy’s next frontier. (MIT Technology Review)
5 Phoebe Gates’ shopping app claimed credit for sales it didn’t drive Phia claimed unearned affiliate sales through fake clicks. (Bloomberg $) + Cofounder Gates is the daughter of Microsoft cofounder Bill. (Engadget)
6 Leaked police drone footage exposes the new reality of surveillance Hours of San Francisco Police video were accidentally released. (Wired $) + Surveillance from drones is on the rise in the US. (MIT Technology Review)
7 Over two-thirds of Americans back a Sanders-style AI ownership plan A poll found strong support for public ownership of AI stock. (Gizmodo) + Tech firms have their own takes on the idea. (MIT Technology Review)
8 AI may soon make campaign text messages more potent—and irritating AI platforms are training bots to sound like political candidates. (NPR)
9 An orbiting disco ball gave Einstein’s theory its most precise test yet It measured Earth’s twisting of space-time more precisely. (Rest of World)
10 Australia’s biggest radio hit may be the product of GenAI Musicians are questioning how the song was made. (Guardian)
Quote of the day
“LOL, I found out I can access the [network storage], so funny.”
—A text message sent by former Apple engineer Chang Liu to a colleague, which a new lawsuit alleges was part of a scheme to steal hardware IP for OpenAI.
One More Thing
Colombian military officials intercepted this 40-foot-long uncrewed fiberglass “narco sub” in the ocean just off Tayrona National Park.
CARLOS PARRA RIOS
How uncrewed narco subs could transform the Colombian drug trade
On a bright April morning in 2025, a surveillance plane operated by the Colombian military spotted a 40-foot-long “narco sub” idling in the Caribbean Sea. The stealthy vessel, used by drug cartels to move cocaine north, could sail with its hull almost entirely underwater.
After seizing the boat, the coast guard noticed something unusual: there was no one on board. This was Colombia’s first confirmed uncrewed narco sub, operable by remote control, but also capable of a degree of autonomous travel.
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.
Anthropic found a hidden space where Claude puzzles over concepts
The AI firm Anthropic has got the clearest glimpse yet at what’s really going on inside large language models as they answer questions or carry out tasks. What they found ranges from the mundane to the unnerving.
Researchers at the company built a tool called the Jacobian lens (or J-lens) and used it to uncover a hidden area, which they named the J-space, inside its flagship LLM, Claude.
The J-space contains words related to the response a model is working on but may not ultimately produce. If Claude were a person (which it is not), you might say these hidden words reveal what’s on its mind before it actually speaks.
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 OpenAI has unveiled its long-awaited “super app” ChatGPT Work blends its chatbot, coding tool, and new models. (Reuters $) + It’s designed to do your work for you and with you. (Ars Technica) + And arrived the same day as OpenAI’s GPT 5.6 models. (NYT $) + It’s also developing a fully automated researcher. (MIT Technology Review)
2 Humanoids have performed teleoperated surgery on living animals In the world-first, they removed gallbladders from pigs. (Ars Technica) + The human work behind humanoids is hidden. (MIT Technology Review)
3 SK Hynix has landed the largest US listing by a foreign company The South Korean chip giant raised $26.5 billion. (CNN) + Demand for AI data centres has led its profits to skyrocket. (Guardian) + But its jumbo share sale may be a sign of overheated times. (FT $) + South Korea’s hottest bachelors are chip workers. (MIT Technology Review)
4 Tencent is leading a deal to unwind Meta’s $2 billion Manus acquisition It’s in talks to become the Chinese AI startup’s largest shareholder. (FT $) + Tencent will reportedly buy Manus for no less than $2 billion. (Reuters $) + Beijing had ordered Meta to unwind the acquisition. (Bloomberg $)
5 Resuscitated human retinas responded to light 10 hours after death It’s a big step towards eye transplants that restore vision. (New Scientist $) + As is a new device that revives dead eyeballs. (MIT Technology Review)
6 Meta has started charging for AI access A new version of Muse Spark has a paid tier for developers. (Quartz) + Meta also plans to start producing an AI chip in September. (Reuters $)
7 OpenAI and Google have sold AI models to blacklisted China groups Via Singapore-based subsidiaries of Alibaba, Baidu and Tencent. (FT $)
8 A daughter tested an AI “death bot” of her father The technology provided both comfort and unease. (New Yorker $)
9 An astronomer says the hunt for alien life needs more statistics He wants to replace speculation with mathematical frameworks. (Quanta)
10 Pokémon Go players turned Times Square into a giant battlefield More than 1,500 fans finally fulfilled the game’s 2016 launch promise. (Wired $) + Pokémon Go is also training world models. (MIT Technology Review)
Quote of the day
“When we’re talking about AI, we love the hype, we get excited about it. The damn thing never actually lands in practice.”
—Vijay Janapa Reddi, an engineering professor at Harvard University, tells Wired why he’s skeptical about grand plans for AI.
One More Thing
B.F. SKINNER FOUNDATION
Why we should thank pigeons for our AI breakthroughs
In 1943, psychologist B.F. Skinner led a secret government project to make bombs more precise. His idea: teach pigeons to guide missiles by pecking at targets on a screen inside a warhead. To train them, Skinner rewarded the birds with food when they made the right decisions, using trial and error to shape their behavior.
Unsurprisingly, the military never deployed Skinner’s kamikaze pigeons. Yet his experiments convinced him that pigeons were “an extremely reliable instrument” for studying learning.
Decades later, those same principles would help power reinforcement learning, the technology behind some of today’s most advanced AI systems.
A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.)
+ Here’s a splendid selection of this year’s NSW architecture award winners. + Photographers have captured the Strawberry Moon’s golden glow in stunning detail. + Idiocracy is the film that best exemplifies the “American experience,” according to a new poll. Look back at the prescient comedy with this Screen Junkies trailer. + Get ready for the weekend with this psychedelic house journey from Jamie xx b2b Caribou.
Ties van der Meer doesn’t know how many siblings he has.
The 47-year-old was conceived at a private fertility clinic in the Netherlands using sperm provided by an anonymous donor. After the Netherlands banned anonymous donation in 2004, the doctor who ran the clinic destroyed records that might have identified those donors, he says.
He describes the situation as “problematic.” Children have a right to know their biological parents, he says. While he did ultimately track down one sibling, who helped him identify his father along with other genetic relatives, he may have others he’ll never find.
Other donor-conceived people who have been able to track down siblings have found they have tens or even hundreds of them. One donor-conceived woman who found 25 half-siblings over the course of seven years told the Guardian, “It does make you feel a bit mass-produced.”
We need international limits on the number of children a single donor can contribute to, a European fertility organization argued yesterday. At a conference in London, members laid out plans to start with a Europe-wide limit.
Today many countries, including the UK, have banned anonymous egg and sperm donation. But anonymity can’t be guaranteed even in places where it is technically allowed. Genetic tests offered by companies like Ancestry and 23andMe, along with genetic registries, have made it much easier for donor-conceived people to find parents and siblings who share their genes.
And because sperm can be frozen and stored for years before it is eventually used, the current set-up can result in situations where donor-conceived people discover the identity of a genetic parent only after the person’s death. They might also find that they have siblings of very different ages, all around the world.
Stories like these can be distressing for donor-conceived people. And there are other reasons why limits are considered important. The offspring of a prolific donor might be at risk of unknowingly forming romantic or sexual relationships, for instance. And some people are concerned that a donor with a harmful genetic mutation might pass that down to many children.
This is unlikely, given the level of screening that most donors undergo. But it has happened. A man who donated his sperm to a sperm bank in Denmark was found to have a genetic mutation that significantly increased the risk of multiple cancers. But his sperm had already been used to conceive at least 197 children across Europe. Some of those children developed cancer. Some died.
Many countries already have legal limits for donors. In Malta and Cyprus, for example, both egg and sperm donors are allowed to contribute to the birth of just a single child, according to data presented at the European Society of Human Reproduction and Embryology (ESHRE) meeting in London on July 8.
Other countries set limits based on the number of families a single donor can contribute to, allowing recipients to have children who share a genetic link. In the UK, that limit is set at 10 families per donor.
But these limits are difficult to enforce, partly because donated gametes don’t necessarily stay in their original country. In Denmark, the national limit is set at 12 families. But the country is a major exporter of sperm. In the UK, for example, more than half of sperm donations in 2020 were imported—with most of those coming from either Denmark or the US.
“The only thing that really makes sense is a transnational limit,” Jackson Kirkman-Brown, a professor of reproductive biology at the University of Birmingham, said at the meeting.
Kirkman-Brown and his colleagues have spent months putting together a document that represents ESHRE’s position on these limits. After consulting with fertility specialists, clinics, sperm and egg banks, donors, and donor-conceived people, the team has developed a plan to start with a Europe-wide limit on sperm and egg donations.
ESHRE is calling on sperm and egg banks, as well as fertility clinics, to respect an initial limit of 50 families per donor. That’s still very high, according to a handful of people I spoke to at the meeting. But at least it’s a start.
Europe should move toward setting limits at 15 families per donor, Kirkman-Brown said. “We may find that 15 is also too high,” says Vasanti Jadva, who studies the psychological well-being of people conceived using donated eggs, sperm, and embryos at City St George’s in London. “We still don’t know what the right number is.”
It will be difficult to enforce these limits, too. And if they end up limiting the supply of donor sperm, there’s a chance that some people will turn to unregulated sperm donations from people who do not undergo health screening. Unregulated donations can lead to other problems for prospective parents, including the possibility that donors will seek parental rights over the children conceived using their sperm.
And it will be even harder to establish international limits. When I asked the American Society of Reproductive Medicine for its thoughts on ESHRE’s proposed limits, a representative directed me to a guidance document saying “it has been suggested” that for a population of 800,000, single donors should be limited to “no more than 25 births” in order to avoid the risk that relatives will have children together. (Considering the US has a population of over 340 million, the total figure could be pretty high, but many sperm banks opt to limit the number of families contributed to by a single donor at around 25.)
Van der Meer thinks that even a limit of five families from a single donor would be high. International donation makes it even harder for donor-conceived people to connect with genetic relatives, so the limit for international contributions should be set at two families, he says.
Still, he thinks ESHRE’s suggested limit is a “positive first step.” Van der Meer has managed to track down a sibling, his father, and nephews, aunts, and uncles. He hopes that future policies respect the rights of donor-conceived children to know, and be in contact with, their genetic relatives.
“But,” he says, “you have to start somewhere.”
This article first appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.
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.
Introducing: the Engineering issue
We can’t fix everything, but we can be ambitious. We can take on the challenge of making the world better through human ingenuity. That’s what the new Engineering issue of MIT Technology Review is all about.
Sometimes the challenges we face are giant, like tunneling beneath the seafloor. Some exist at the nanoscale, as with a new ASML machine powering the future of chipmaking. Others represent problems at a planetary scale and in truly unknown territory, like replicating a volcano’s mechanism to cool the Earth on purpose.
These incredible engineering stories show we can come together to get to work and, when the smoke clears, find we’ve made real progress. Subscribe now to read all of them—and more—in the full print issue.
Stripe, Anthropic, and OpenAI are backing an effort to stop respiratory infections
The common cold comes for us all—often more than once a year. And there is no way to prevent it. The best you can do is take vitamin C and stay away from people with the sniffles.
Now, the payment company Stripe is funding a new $500-million nonprofit aiming to prevent both the common cold and the flu. Its eventual goal is to get rid of respiratory viruses altogether.
Anthropic, OpenAI, and Bill Gates have also backed the venture, which will investigate whether modern technologies can counter the common cold and the flu. Dive into the nonprofit’s plans.
—Antonio Regalado
MIT Technology Review Narrated: inside the hunt for the most dangerous asteroid ever
As asteroid 2024 YR4 hurtled toward Earth, astronomers determined that this massive rock posed a higher risk of impact than any object of its size in recorded history. Then, just as quickly as history was made, experts declared that the danger had passed.
This is the inside story of the network of global scientists who found, followed, planned for, and finally dismissed the most dangerous asteroid ever discovered —all under the tightest of timelines and with the highest of stakes.
—Robin George Andrews
This is our latest story to be turned into an MIT Technology Review Narrated podcast, which we publish each week on Spotifyand Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as it’s released.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 China has taken the US’s crown for the world’s fastest supercomputer Shenzhen’s LineShine overtook California’s El Capitan. (Axios) + China had not had a machine at the top of the list since 2017. (NYT $) + But the supercomputer race isn’t geared for AI work. (Reuters $)
2 Mythos reportedly found flaws in classified US government systems A US official said Anthropic’s model identified certain vulnerabilities. (AP News) + The model has now been suspended over US security concerns. (BBC) + The NSA has lost access to Anthropic’s tools in fallout. (Engadget) + The feud raises new questions about AI safety. (MIT Technology Review)
3 A US pilot reported seeing Iranian drones swarm in “jellyfish” formation Which would represent an alarming advance in Iranian drone capabilities. (CNN) + The US is heading toward a drone-filled future. (MIT Technology Review)
4 Mark Zuckerberg directed Meta to create a prediction markets app It will be similar to Polymarket and Kalshi. (NYT $) + But won’t let users wager real money. (The Verge) + Another new app, Meta Photos, will create media with AI. (Reuters $)
5 SpaceX’s “Starfall” just launched a secretive test flight The orbital delivery spacecraft blasted off for the first time yesterday. (Axios) + It could also support space manufacturing. (New Scientist $)
6 Alibaba has sued the US for being linked to the Chinese military It wants to be removed from a Pentagon blacklist. (Reuters $)
7 Nvidia’s banned AI chips have doubled in price on China’s black market The DGX B300 now costs more than $1.1 million. (Financial Times $)
8 Tesla claims a driver “manually overrode self-driving” in a deadly crash It said the accelerator was pressed “all the way to 100%.” (The Verge $)
9 The US science retreat has created an opportunity for Europe But questions about funding and innovation remain. (Nature) + Trump has dealt many blows to US science. (MIT Technology Review)
10 Meta’s new smart glasses ditch Ray-Bans for Kylie Jenner Meta logos and Jenner designs have replaced the Ray-Ban branding. (Wired $)
Quote of the day
“It’s blasphemy against AI if you say it’s a bubble.”
—SoftBank founder and CEO Masayoshi Son tells shareholders that the AI boom is still in its early stages, Reuters reports.
One More Thing
ERIK CARTER
Video games are dividing South Korea
They say StarCraft was the game that changed everything. When the science fiction strategy game arrived in South Korea in 1998, it wasn’t just a hit—it was an awakening.
Out of 11 million copies sold worldwide, 4.5 million were in the country. The game was so popular that it triggered another boom: “PC bangs,” pay-as-you-go gaming cafés.
StarCraft and PC bangs spoke to a generation of young South Koreans boxed in by economic anxiety and rising academic pressures. But they also sparked arguments about game addiction. They’ve led to feuds between government departments—and a national debate over policy.
A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.)
+ This archive lovingly documents the beautiful design of over 1,700 obsolete objects. + Classic TV theme tunes like Hey Arnold! Have been revived in a musician’s marvellous samples. + Marvel at the mind-boggling geometry of nature and see how bees perfectly construct honeycombs. + Hear the ominous, deeply atmospheric tones of a custom string instrument built inside a plastic drainage pipe.
The common cold comes for us all—often more than once a year. And there is no way to prevent it. The best you can do is take vitamin C and stay away from people with the sniffles.
Now, the payment company Stripe, founded by brothers Patrick and John Collison, says it will fund a new $500-million nonprofit whose goal is preventing both the common cold and the flu. Its eventual aim is to get rid of respiratory viruses altogether.
The new organization, called Intercept, will use grants and investments to back prevention approaches, including vaccines, as well as large-scale air-cleaning systems for schools, offices, and other public spaces.
In addition to Stripe, other funders include Anthropic, Flu Lab, the OpenAI Foundation, as well as Bill Gates and several traders at the quantitative investing fund Jane Street Capital, according to an Intercept spokesperson.
“I think we treat respiratory infections as a minor nuisance, but have really underweighted the burden that they impose on society,” says Nan Ransohoff, the Stripe executive leading the initiative along with Charlie Petty, a venture capitalist who joined Stripe this year. The average person will spend 5% of their lifetime fighting a cold or the flu, according to Ransohoff.
Despite that, drug companies put relatively little effort into preventing colds. Part of the problem is that the sniffles are caused by more than 200 different viruses, according to the American Lung Association, with rhinoviruses being the most common culprits. There are so many that it typically doesn’t pay to try to stop any one of them with a vaccine. “When pharma companies look at it, it’s not as attractive as other things they could work on,” says Ransohoff. “So it hasn’t attracted the resources.”
Stripe previously organized a $1.8 billion program called Frontier to encourage the development of carbon removal technology, as a way of countering climate change. Ransohoff says removing carbon from the atmosphere and getting rid of respiratory viruses are similar in that each is “technically possible” but they “lack commercial incentives.”
The concept for Intercept took shape after Ransohoff started talking to David Veesler, a structural biologist and vaccine designer at the University of Washington, who argued that it’s possible to come up with broad countermeasures that work against many viruses at once.
“He effectively sort of nerd-sniped me,” Ransohoff says of Vessler. “He convinced me that this is technically possible. He also helped me understand that some of the reasons that this hasn’t been done before was sort of an incentive problem.”
Veesler says the growing toolkit available to scientists includes RNA drugs, antibodies, and computational protein design. For instance, one idea is to engineer virus-grabbing proteins that people could spray in their nasal passages, to catch viruses before they can infect people.
“Most people just accept these viruses as a fact of life, and that got us thinking: do we have to accept it?” says Veesler. “The more we thought about it, the more we realized that many of these problems have not been worked on with modern technologies.”
The project takes inspiration from efforts to fight the covid-19 virus, where Veesler’s group was among those involved in the speedy development of vaccines, anti-viral drugs, and antibodies.
According to Ransohoff, Intercept’s advisers will include Peter Marks, a former top FDA official, as well as Moncef Slaoui, the pharmaceutical executive who led the US coronavirus vaccine effort, Operation Warp Speed.
A key challenge for Intercept will be coming up with ways to counter many—even all—viruses at one time. That accounts for the group’s interest in air-cleaning technology, such as using strong ultraviolet light to inactivate viruses. The idea, the group says, is to remove viruses from the air in the same way municipalities remove impurities from the water supply before it’s piped to people’s homes.
The US funds about $6.5 billion a year in virus research through the National Institute of Allergy and Infectious Disease, or NIAID. But that agency’s budget hasn’t grown in recent years, leaving more room for private philanthropy.
And Stripe’s Collison brothers have become some of the most reliable philanthropists in viral research. After giving away “fast grants” to help labs during the covid-19 pandemic, they later joined other donors who committed $650 million to establish the Arc Institute, in Palo Alto, which has developed AI models for biological research.
“The diversity of viruses is just too large and seems daunting, so people don’t even try,” says Veesler. “I’m happy that someone is ready to help scientists, not accepting the status quo, and doing something different.”
AI is booming. New use cases are emerging each day. To capitalize on the technology’s potential, enterprises require data at scale. In many cases, though, the relevant information is blocked or unstructured, which limits its use by AI models.
To understand this challenge, consider the foundation of the web itself. The web was not designed for the automated discovery and retrieval that new AI applications demand. Overcoming this inherent design constraint requires infrastructure.
The next frontier in AI may depend on a new web data infrastructure layer that can enable models to discover and map this ever-expanding digital realm. This layer must be able to navigate hundreds of millions of existing web domains and billions of new URLs created each week, delivering real-time information and overcoming technical barriers.
“The data suggests there’s far more data out there,” says Or Lenchner, CEO of Bright Data, a web data collection platform. “Think of the universe: It’s out there, but you don’t know what you don’t know.”
Enabling access to fresh, relevant, and trustworthy data
While early AI breakthroughs were driven by scaling training data and model size, organizations are now encountering a fundamental bottleneck: They need to keep pace with the dynamic, unstructured, and constantly evolving nature of web data in order to ground outputs in current and verifiable information. AI performance increasingly depends not just on model architecture but on a system’s compute, networking, retrieval, and data engineering capabilities—that is, the system’s ability to quickly and reliably retrieve data that is fresh, relevant, and trustworthy.
Traditional model training relies on snapshots of information collected at a particular point in time. Training AI on such static data is no longer sufficient. To track fluctuations such as competitor pricing, consumer sentiment, and market trends, companies need a constant feed of new information, pulling data in real time along with relevant context. Their infrastructure must therefore be able to handle millions of simultaneous interactions across websites that vary by geography, language, format, and access rules.
“If it can’t retrieve real-time information, it lacks context,” Lenchner says. “In a business setting, that’s not acceptable anymore. Stale answers lead to bad decisions and disappointed consumers.”
Speed is not merely a matter of convenience; it’s a matter of necessity. Today’s organizations operate in environments where prices, inventory, markets, security threats, and customer behavior change continuously. Delayed data retrieval can reduce the usefulness of an otherwise sophisticated model.
Using live, high-quality web data can also reduce AI hallucinations because the model has a more relevant knowledge base. This builds user trust. In fact, one survey found that 56% of AI practitioners said businesses need access to real-time web data to improve trust in AI outputs. To ensure the model runs efficiently and effectively, the information must also be pared down to the appropriate essentials.
Despite the introduction of retrieval-augmented generation (RAG), where models pull in external data at the moment of a query, many AI systems still struggle to deliver outputs that are current, contextually relevant, and trustworthy in operational settings. According to Gartner, 60% of AI projects that are not supported by AI-ready data—accurate, structured, organized, and contextualized—will be abandoned by the end of the year.
This is because large-scale retrieval alone does not solve the problem. As Lenchner puts it, “You need to retrieve data at scale, but also in real time. Latency becomes an issue because of the end user who is waiting for the output.”
Accessing fresh, AI-ready data at scale introduces technical and structural challenges. In practice, many enterprise systems combine public web retrieval with APIs, licensed datasets, and proprietary internal data in their AI applications. Integrating these fragmented sources into a timely and usable knowledge layer requires specialized capabilities. Some research has found that 97% of AI organizations depend on real-time web data infrastructure, but 90% feel boxed in by various restrictions. Companies are increasingly developing technical approaches to navigate these constraints.
Lenchner draws this metaphor: “Think of the trained model as intelligence and relevant data as knowledge. A powerful intelligence layer sitting on top of a hollow knowledge layer is like a genius who knows nothing—useless in practice. Intelligence and knowledge have to come together.”
The promise of new infrastructure
A new layer of web data infrastructure can address this developing need for stronger AI inputs by enabling discovery of data, real-time access, and tailoring to a specific context. As Lechner describes it, “It’s all about collecting data at scale, super-low latency, without being blocked.”
Rather than relying on increased computing power, this type of platform emulates human browsing behavior to access available content and transform raw code into structured data feeds. It can work with websites that might not interact with traditional scraping tools, such as those heavy in JavaScript, or with aggressive antibot software.
As Lenchner explains, “It’s basically having infrastructure that can mimic a web user with identifying information—IP address, location, and 1,000 more parameters. And at scale. Think of doing that 80 billion times a day for millions of websites. And every single time, you are looking exactly as the website expects you to look.”
Of course, continuous retrieval introduces new data governance challenges. To address them, platforms can enforce strict compliance protocols aligned with global privacy frameworks, such as the EU’s General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). They can also be limited to openly accessible, public information, avoiding paywalls or private logins. Any networks used can be vetted and consent-based, and incentives can be provided to owners of IP addresses. In this way, systems can be designed to comply with tightening regulation.
Such complex capabilities do not come easy. “When this is critical infrastructure for a company,” Lenchner says, “doing it in-house becomes a full-time engineering problem that competes with the actual AI work.” Addressing this complexity requires organizations to commit significant resources, leading many to seek specialized platforms designed specifically for data retrieval, orchestration, and observability.
Infrastructure for the real world
Real-time data retrieval is changing what AI systems can do inside organizations. For example, a retail company can use public information to enable a dynamic pricing engine, and global brands can track trademark infringements.
As the ecosystem matures, organizations that invest in this emerging data infrastructure layer will be better positioned to build AI systems that are more responsive, reliable, and aligned with real-world conditions—AI systems that can continuously adapt using current web data. Over time, the distinction between AI models and the infrastructure that feeds them may even begin to disappear.
As Lenchner says, “The world is changing. And everything that is happening in the world is being uploaded to the public web. The amount of new data that is being generated is growing and accelerating.”
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
As soon as August, a giant silver bullet will cut its way through the dry air of the southwestern US and cross the Pacific to reach the coast of Japan.
Once there, the roughly 200-foot-long craft, built by the New Mexico–based company Sceye, will park some 18 kilometers above the ocean’s surface, in a wispy-thin layer known as the stratosphere. Then it will use a custom-built antenna to supplement Softbank’s 5G network, a test that will include beaming data straight to devices.
Sceye (pronounced “sky”) is one of several firms building a class of airborne craft called HAPS, or high-altitude platform stations or systems. Such a platform can be a plane or a balloon or, yes, an oblong craft filled with helium and outfitted with solar panels. HAPS companies, including the Airbus subsidiary Aalto, envision them serving a variety of lofty purposes, such as delivering internet service to disaster sites and observing Earth’s surface.
Hovering over Roswell, the high-altitude system is sheathed in lightweight, reflective fabric.
COURTESY OF SCEYE
The stratosphere is a good place to be if you want to cover a large area. It’s also much closer to the ground than even the lowest-orbiting satellites, which means sending down a signal takes far less energy. “What we ultimately offer is space-like conditions, without the cost of going to space and without the complexity of being in orbit,” says Mikkel Vestergaard Frandsen, Sceye’s CEO and founder.
But it’s also not so easy to stay there. Sceye’s aircraft, Frandsen says, has to be light enough to stay aloft but also strong enough to carry the necessary systems. It must soak up and store enough solar energy during the day to provide around-the-clock power to an electric fan that can maneuver the HAPS back into place when winds knock it out of position—mettle it proved in a 2024 test flight.
Since then, Sceye has been preparing for its big Japan test. In the flight pictured here from this spring, for example, the craft stayed aloft for 12 days as it flew to the coast of Brazil and spent more than 88 hours “parked” in various locations. Eventually, the company expects its platform could help satellite operators better serve densely populated areas.
Someday, Frandsen says, spotting a HAPS may be as common as seeing ships at port or trains on the tracks.
When I was 18, I skipped my high school graduation and headed to Kuwait. It was 1991, the first Gulf War had just ended, and the country was in complete chaos. There was little to no electricity, aside from generator power. Rubble and unexploded ordnance were everywhere. Massive oil fires lit up the desert and dimmed the sky overhead. Everything had to be rebuilt, and fast.
I was there to work on that reconstruction, part of an international effort to fix what war had broken. It was the first time I’d ever seen a truly massive engineering project with my own eyes. The challenge was one that had to be addressed on multiple fronts, simultaneously, to get an entire country back up and running.
Everywhere you looked, there was something to do. I was mostly working with a labor crew to jump on quick fixes to things like windows and doors blown out in the fighting. But there were, of course, bigger jobs too. Most notably, there were those massive fires to put out. The Iraqi army had set hundreds of oil wells ablaze, the majority of which were still spewing soot and oil smoke into the air. On bad days, the sky would remain dark all day, and the air would burn your eyes and hurt your throat.
It was so apocalyptic that none other than Carl Sagan warned of massive environmental consequences. If smoke from the oil fires reached the stratosphere, he predicted, the result could be akin to the 1815 explosion of the Tambora volcano in Indonesia, which triggered what’s known as “the year without a summer”; global temperatures dropped between 0.4 and 0.7 °C, and crops failed around the world. Fortunately, the plume from Kuwait never made it that high, and although temperatures did decline regionally, there was little effect on a planetary scale. As it turns out, predicting what will or won’t lower global temps is quite hard. (Reader, I am foreshadowing here.)
Firefighters from companies with monikers like the Red Adair Company or Boots and Coots (as well as less colorfully named outfits, like Bechtel) rushed to Kuwait after the war’s end to figure out how to extinguish the gargantuan blazes and cap the wells. At a hotel in downtown Kuwait City, one of the few places with working phone lines, I would occasionally run into them covered head to toe in black oil and soot.
Putting out the fires took all sorts of creative thinking. Engineers working in the burning oil fields figured out they could repurpose existing pipelines meant to pump oil out to sea to instead pump water in from the Persian Gulf. One company from Hungary rigged up a firefighting machine called Big Wind by outfitting an old Soviet T-34 tank chassis with two turbines from a MiG-21 fighter jet, each of which could blast 220 gallons of water per second. Sadly, I never got to see it in action (except in the movies).
Other jobs were less cinematic but no less dire. The retreating Iraqi army had left booby traps all over. They snaked hand grenades into the plumbing (at a facility where I worked, among others). They planted mines everywhere, and those had to be found and removed. Many of them were small plastic “toe poppers” designed not to kill but to maim. Hunting them was a herculean effort. And although it was mostly successful, hundreds of thousands of them, by some estimates, still remain.
Which is to say, we can’t fix everything. But we can be ambitious. We can take on the challenge of making the world better through human ingenuity.
That’s what the July/August issue of MIT Technology Review is all about. Sometimes the challenges we face are giant, if knowable, like tunneling beneath the seafloor. Some exist at the nanoscale and represent decades of investment and research, as is the case with ASML, a company with the singular ability to produce the machines that make the most advanced computer chips in the world. Others represent problems at a planetary scale and take us into truly unknown territory, like a future where we could engineer the veil of the Tambora volcano to cool the Earth on purpose.
By the end of my 90-day contract in Kuwait, instead of damage everywhere you looked you could see the fruits of a gargantuan international rebuilding effort. The air was not clean, exactly, but inhaling it no longer felt like smoking a pack a day. On the beach, which had been pocked with mines, people swam and splashed at the edge of the gulf. The lights were on. Water ran from the taps. The markets were open. It was a remarkably different place.
Yes, forces both within and beyond our control will always break things. People will invariably make mistakes, or act out of their own self-interest to the detriment of others. But we can also come together to get to work and, when the smoke clears, find we’ve made real progress.
The Argentina v. France final of the 2022 Men’s World Cup in Qatar was shaping up to be one of the most epic games in soccer history. With just 12 minutes remaining in the extra time added to the game to break a tie, the referee had a critical decision to make—and fast.
Lionel Messi, the Argentine captain and soccer legend, had just launched the ball past the French goal line, giving Argentina a 3–2 lead. The crowd roared, but a flag was raised. One ref thought that shortly before Messi kicked the ball, the Argentine forward Lautaro Martinez had been closer to the goal than any French players apart from the goalie when he’d received a pass—putting him in an illegal “offside” position.
If the head referee called Martinez offside, the goal wouldn’t count. If he declared him onside, Argentina would keep its 3–2 lead with minutes left to play.
The weight of more than just one offside call stood on that referee’s shoulders; it was the weight of the World Cup itself.
But in 2022, for the first time in the storied competition’s history, referees had access to semi-automated offside technology (SAOT), a system that could rapidly analyze the play and detect an offside player. In this case, it produced an image revealing that a French defender was slightly closer to the goal than Martinez, just barely leaving the Argentine forward in a legal attacking position.
The referee ruled that the goal counted: 3–2, Argentina.
SAOT produced this image to determine whether Argentine forward Lautaro Martinez (in white) was offside. It shows that only Martinez’s fingers had crossed the vertical white line into offside territory. Players’ hands and arms are not considered for offside decisions, so Martinez was declared onside.
COURTESY OF THE RESEARCHERS
Argentina eventually emerged as the champion, winning a penalty shootout after a late goal by French forward Kylian Mbappé tied the game at 3–3. Only in a parallel universe will we know how the game—and the tournament—would have played out if the referee had overturned Messi’s goal.
For FIFA, soccer’s international governing body, SAOT is among the latest in a portfolio of innovations used at the World Cup. From goal line technology to video assistant referee (VAR) tools, officiating tech is now commonplace at the top level of the game.
But SAOT is part of a broader sports technology landscape that stretches far beyond soccer. And one of the major players in that landscape is the very team that collaborated with FIFA to bring SAOT to the pitch in the first place: the MIT Sports Lab. Founded in 2015, the lab focuses on using technology and data science to tackle real problems facing athletes, teams, and sports organizations and brands.
The lab has worked with FIFA, the NBA, the NFL, and Adidas, and it collaborates with a host of other sports organizations and industry players. Some of its work may be hiding in the soles of your running shoes, in the decisions your favorite NBA team makes, or even on soccer’s biggest stage—as was the case in what the AP called “probably the wildest final in the tournament’s 92-year history.”
The MIT Sports Lab’s origin story begins around 2010, when Anette “Peko” Hosoi, the Pappalardo Professor of Mechanical Engineering, fell in love with downhill mountain biking and needed a new bike. But given the varying linkage systems, shock types, and geometries, she found it difficult to choose the best one. Encountering only minimal information online, she assigned the analysis to her 2.001 class, the introductory course on mechanics. “All of my exams that semester were bike questions,” she says. They proved to be really good engineering questions too.
Having recently earned tenure, she wondered, What if I actually built this sports thing into something bigger? In 2011, she began conceptualizing a project called STE@M (Sports Technology and Education at MIT), which would assemble students, faculty, athletes, and industry partners to tackle sports engineering challenges. As the effort kicked into gear over the next few years, Hosoi began collaborating with Christina Chase, MIT’s new entrepreneur in residence, and in 2015 the two of them cofounded the MIT Sports Lab.
Mechanical engineering professor Anette “Peko” Hosoi assigned bike engineering challenges to her students when she needed a better mountain bike. In 2015, she cofounded the MIT Sports Lab with entrepreneur and MechE lecturer Christina Chase.
COURTESY OF MIT MECHANICAL ENGINEERING
“It turned out that we’re the perfect combination for this because my background comes from the math, physics, engineering side,” says Hosoi. “And she comes from the entrepreneurship [and] product development side. To really interface with these different sports companies and leagues, you need to span that whole spectrum.” Chase became the lab’s managing director and Hosoi its faculty director.
For over a decade, the Sports Lab has grown as interest in sports tech has skyrocketed—and it’s accumulated what younger fans would call some elite ball knowledge in the process.
This depth is exactly what its partners need.
“There’s more and more data that’s getting collected,” says Hosoi. “A lot of the teams, leagues, brands don’t necessarily have the in-house manpower to extract the information they need. So that’s where we can give them a boost.”
When MIT researchers looked at early skeletal data representing soccer players in motion, they saw “skeletons” flying above the ground or completely underground, in anatomically impossible positions.
The FIFA partnership has been especially fruitful—and the Sports Lab’s role in validating SAOT has probably had more impact than any other project the organizations have worked on together, says Ferran Vidal-Codina, SM ’13, PhD ’17, a former research scientist at the lab who was part of the team from FIFA, MIT, and third-party data providers that developed the technology.
The system’s viability depended on the ability to quickly access and analyze what’s known as tracking data—the record of everywhere the players and the ball move throughout a game.
To collect that information at top-level FIFA tournaments, data providers station about 12 state-of-the-art cameras around the stadium, capturing images at double or more the speed of normal broadcasting cameras. Computer vision algorithms then convert the feeds into what’s called skeletal data—3D representations of the players in motion.
“It’s a ton of data—22 players, one referee, two assistant referees, [each with] 29 joints with XYZ coordinates, 50 times per second,” says Henry Wang ’23, a former MIT varsity swimmer who earned undergrad degrees in both business analytics and computer science, economics, and data science and is now a Sloan PhD candidate and a FIFA research consultant at the MIT Sports Lab.
Lionel Messi scores Argentina’s third goal past Hugo Lloris of France during the final of the 2022 FIFA World Cup in Qatar. Argentina prevailed over France in the match.
MATTHIAS HANGST/GETTY IMAGES
That works out to some 108,900 data points persecond for a game that lasts at least 90 minutes. And that’s just the players and referees—a chip embedded in the ball also collects position and velocity data 500 times per second. In total, that’s easily more than a dozen gigabytes of skeletal data and ball-tracking data per game.
FIFA was thrilled to have that much data to work with. But around 2021, when third-party providers started offering skeletal data, the organization did not have the full range of technical skills needed to validate it. “So the data got sent to us,” says Wang.
Right away, the team at the Sports Lab saw some issues. “We saw ‘skeletons’ flying above the ground or completely underground, in anatomically impossible positions,” Vidal-Codina recalls. “We saw skeletons having their bones and limbs stretching from 30 centimeters to a few meters. We saw balls doing weird motions in the air. All sorts of stuff that when you look at it—yeah, that’s definitely not ready to be used.”
Often, when there’s a new idea in the works, “we’re the ones that take the first stab at it,” says Sports Lab researcher and PhD student Henry Wang ’23. “We are the ones that prototype and show it’s possible.”
The lab’s job in tackling this problem was first to validate the data being fed into the system and then to confirm that the SAOT algorithm itself was performing exactly as the third-party vendors claimed it would.
In 2021 and 2022, FIFA ran a multitude of tests. Renting out a stadium for days at a time, the organization brought the data providers on site, where amateur players, or sometimes even FIFA staff, would run dozens of offside drills while those vendors collected live data.
The lab focused on analyzing that data and relaying the results to FIFA and its providers, which incentivized them to make improvements while casting light on blind spots they sometimes did not know they had, Vidal-Codina says. The lab was able, for example, to analyze how the call might differ if you focused on a player’s whole body, including arms and legs, or just the center of mass.
Before the technology could officially hit the pitch, the Sports Lab had to answer some key questions. First, could FIFA collect live data from the providers fast enough to make game-time assessments feasible? The researchers helped answer this by building a tool in Google Cloud to collect data as it was generated so the lab could later check the latency, allowing FIFA to understand just how “live” its data really was.
Also important: determining whether two data sets—the skeletal data and the information captured by what’s known as connected ball technology—could be combined to reliably yield a correct offside call. The lab helped do just that, developing a protocol that synched the systems collecting skeletal and connected ball data.
After validating SAOT, tweaking it, and testing it in many situations, including some official FIFA matches in 2021 and 2022, “FIFA felt it could be used at the biggest stage, which was the World Cup,” says Vidal-Codina. Indeed, FIFA president Gianni Infantino endorsed the tool himself when it debuted in Qatar.
Over the course of the 64-game tournament, SAOT assisted in more than 150 offside calls, some with weighty effects. Eight goals were overturned after a referee declared the scoring team offside; two goals were added to the scoreboard after a referee had incorrectly disallowed a goal that was not, in fact, offside; and in sevencases, an offside call assisted by SAOT changed the game’s outcome.
These results highlight just how crucial a single offside decision can be, given the low scores typical in soccer—and how tools like SAOT can help improve the game. “Overall, decisions have been made quicker and better. That’s ultimately what we strive for,” Vidal-Codina says.
The technology also takes some of the pressure off referees. “I would argue that the goal of our work is to make sure that the referee is as informed as possible about the decisions that they make,” says Wang. “It’s an incredibly difficult job.” During World Cup play, SAOT’s animated visuals were shown on stadium screens and available to as many as 5 billion viewers across platforms to help them understand the referees’ calls.
But the technology is meant to assist referees, not replace them. “We don’t want people to think that we are automating referees. I can guarantee you the referee is not going anywhere,” Wang says. “We want to make sure that the human element is transparent, that it’s informed, and that we are helping referees do their job.”
SAOT may have been the Sports Lab’s highest-profile FIFA project to date, but the lab has had a hand in shaping the organization’s larger innovation pipeline. It’s helped improve the way technology—from hardware like cameras to officiating tools like SAOT—gets tested and certified on its way to the pitch. Since 2021, FIFA’s process for certifying data providers’ systems has included having the Sports Lab assess their data latency from a live data collection event using the same infrastructure it built to validate SAOT. And often, when there’s a new idea in the works, “we’re the ones that take the first stab at it,” says Wang. “We are the ones that prototype and show it’s possible. It’s a call to the industry to say: ‘Hey, this is interesting.’”
FIFA isn’t the only organization interested in the insights that tracking data can offer; the NBA has been collecting it for over a decade. In 2025, Hosoi and the MIT Sports Lab published a paper based on an NBA-MIT collaboration that had a unique focus: Instead of using the tracking data to analyze the game’s physical elements, they sought to understand the mental ones.
“Currently, everything physical about an athlete gets measured,” Hosoi says. “But if you talk to the organizations, they tell you that the mental part of the game is just as important. And we have no tools for measuring the mental part. So the question is, can we use the physical tracking data to extract metrics for mental performance?”
In basketball, a big part of the mental game comes down to decisions around when to shoot and when to pass. But it’s not so easy to determine which players are making good or bad decisions. So MIT researchers created a metric called expected action value (EAV), which is essentially an assessment of the likelihood of a play’s success. Using a model trained on all 786,208 passes from the 2018–’19 NBA season and all 1.4 million shots from 2013 to 2019, they were able to figure out expected outcomes of different plays.
EAV takes into account the velocity of the shot and the acceleration of the player making the shot as well as the positions of players on the court. For instance, an uncontested three-point shot from the corner has a higher EAV than a two-point attempt from a player getting double-teamed closer to the basket (or “in the paint”). This approach can tell you not only the likelihood of a successful shot but also the chances of a successful pass. If the player decides to pass instead of shoot, and the receiver of the pass has a reasonable chance to make the shot, then that was a good decision by the passer.
A consistent record of high-EAV choices—passing at some times, shooting at others—means a player is making good decisions. “You can just calculate: How many times do players make good decisions? How many times do they make bad decisions? And we can rank NBA players by good decision-makers and bad decision-makers,” says Hosoi.
This approach can also help teams see if points are being left on the table. Given that teams averaged about 110 points per 100 possessions in the 2019 NBA season, or 1.1 points per possession, if a player passes up a play option with an EAV of more than 1.25 for a play with a lower EAV, the Sports Lab’s model classifies it as a “missed opportunity.” Flagging these moments saves time for coaches, who have to review video for at least 82 games every season. “If we can point to the time stamps of the different games where your guys might have missed an opportunity, you can take advantage of that, right?” Hosoi says.
At this point, the MIT Sports Lab doesn’t really need to advertise its services. “If you are good in sports, everybody who needs to know will know,” says Hosoi. The lab’s partners come to it if they need answers to questions—as the NFL did during the covid crisis.
At the beginning of the 2020 season, some teams had opened their stadiums for limited in-person attendance while others didn’t allow any fans. In March 2021, “there was a paper that was published that said in the cities where NFL stadiums have opened, there are spikes in covid cases,” Hosoi recalls. “And the NFL called us and said, ‘Wait, is this true? Because if this is true, we’re going to stop. Can you guys do an analysis on this?’”
After investigating, the Sports Lab identified a problem with the original paper. NFL teams made decisions about opening stadiums in conjunction with stadium owners and local governments. What the paper didn’t consider, however, was that some states had stricter covid protocols than others, and it was stadiums in those places that tended to stay closed to fans.
The lab accounted for the confounding factors involved and found that opening a stadium with distancing and masking protocols had no effect on covid cases. In fact, the analysis found that in some places, in-person attendance was correlated with case totals that were lower than expected. Hosoi hypothesizes that this was not only because the open stadiums required distanced seating and other safety measures but also because if fans were at the stadium, they were usually outdoors—not mingling in a crowded bar or at a friend’s house. Partly on the strength of these findings, the NFL decided to open all stadiums for in-person attendance in the 2021 season.
The Sports Lab’s expertise isn’t limited to data analytics; companies are also welcome to bring their hardware and product quandaries to the lab. Adidas, for example, had announced development of a 3D-printed midsole for running shoes in 2015 and was eager to bring it to market. It partnered with Carbon, a Silicon Valley company specializing in the technology, and by around 2017 the shoe manufacturer had finally figured out a way to produce 3D-printed midsoles at a speed that could match the commercial scale.
Still, it wasn’t quite sure how to use this innovation. Adidas approached the Sports Lab with one big question, which Sarah Fay ’15, SM ’18, PhD ’21, summarizes as “We know we can do all this cool stuff, but what should we do in order to make a high-performing shoe?”
“A regular running shoe just has a slab of foam in the bottom,” explains Fay, who tackled this project while earning her PhD. “You can only change the stiffness by changing the thickness. The exciting thing about 3D printing is that you can change the stiffness without having to change the shape, the footprint of the midsole—just by changing the lattice architecture.”
But manufacturing a high-performing shoe would be tricky: No two human runners are the same, and there was not much data from the running world at the time. So Fay turned to mechanical models—in particular, the mass-spring-damper model for analyzing a system’s dynamic behavior, which Thomas McMahon, a biomechanics pioneer at Harvard, had used to assess different running surfaces in the 1970s. “Just a simple model can be super powerful,” Fay says.
Fay iterated on this foundation to build a model with a center of mass, a rotating hip, and a leg that stretches. It could predict how runners of a given height, weight, and leg length would adjust their gait in response to different levels of springiness and shock absorption in a simple test shoe. This let Fay and Hosoi test gait response as they varied the stiffness of various parts of the midsole.
Sarah Fay ’15, SM ’18, PhD ’21, holds a 3D-printed midsole for a running shoe. Working with Peko Hosoi in the Sports Lab, she developed a model that makes it possible to predict how different midsoles would affect a particular runner’s gait.
MELANIE GONICK/MIT
To ensure the accuracy of the model, they also considered that runners typically (and often unconsciously) try to minimize what they called a “biological cost function” of running, such as the impact they feel when their foot hits the ground, or the jerkiness of their gait. In multiple simulations, they optimized their model for various biological cost functions, and they compared the resulting gaits with actual gaits recorded in a previous treadmill study. Upon finding that most runners try to minimize both the impact of their feet and the amount of energy their legs expend, Fay and Hosoi were able to optimize the model for those two factors to deliver highly accurate gait projections. And the ability to predict the gait made it possible to predict how well a shoe would perform.
Adidas used the model to help evaluate potential lattice-structured midsole designs, selecting the top performer for fabrication to do more formal testing. “Those are the shoes that Adidas ended up making and selling that I wear basically every day,” Fay says. She imagines that one day it could be possible to analyze running videos, determine the best shoe architectures for specific runners, and 3D-print shoes designed just for them.
Fay was able to fill in the mathematics and engineering expertise that the Adidas team was missing. And by giving her a way to couple her technical skills with her experience as a lifelong athlete who played both field hockey and squash at MIT, the Sports Lab may have helped her find her calling. Today, she runs a sports-related research lab of her own at Smith College, where she’s an assistant professor of engineering studying the biomechanics of soccer cleats and their role in players’ risk of knee injury.
“The big part of sports for me is just that it was a safe space for me to learn how to be a leader, how to be a person, how to be a teammate,” she says. “And I figured that that’s a valid enough reason to make my career path head in that direction.”
What Vidal-Codina calls the “most magical feature” of the lab is that it meets its partners in the sports industry where they are. As he puts it, its scientists can say, “Okay, what do you need help with? We may have the skills or the methodology to come to a solution. So let’s sit together and try and figure it out.”
“The best thing about the Sports Lab is the community of people we’ve built—a direct connecting line from the industries and the teams to our students and to our faculty.”
Anette “Peko” Hosoi, Pappalardo Professor of Mechanical Engineering, MIT
But its work benefits the MIT community as much as it does the world of pro sports. The Sports Lab hosts an annual MIT Sports Summit, which brings technical and management professionals in sports to campus to help students, faculty, and industry figures make personal connections and share their work. Hosoi and Chase also teach 2.98 (Sports Technology: Engineering & Innovation), a class that involves MIT students in real industry projects. And the lab brings pro-level sports insights to MIT athletes, partnering with the athletics department on projects like analyzing the NCAA Power Index—the metric used to select and seed teams for the Division III national tournament—with an eye toward helping MIT teams maximize their chances of securing spots. Another project involves collecting athletes’ personalized weight-room stats into a dashboard to give coaches a window into their performance and enhance their recovery. The lab also worked with an MIT soccer player to create a tool that automatically tracks the passing sequences leading to goals, shedding light on which players contributed. It’s now widely used by the Institute’s soccer teams.
“The best thing about the Sports Lab is the community of people we’ve built—a direct connecting line from the industries and the teams to our students and to our faculty,” Hosoi says. “That collaboration is better than the sum of the parts.”
While the lab’s work may take place behind the scenes, its influence will continue to ripple across the world of sports—from the soccer games on our televisions during this year’s World Cup to the shoes on our feet.
And the lab will do it by asking the most important question of all: “How can we help?”
Here’s a problem you probably didn’t solve in school: You’re an ambitious young plumber from Brooklyn in a world inhabited by violent human-size mushrooms called Goombas. The love of your life has been kidnapped, so you embark on a quest to rescue her, venturing through stretches of pipe-filled and monster-ridden terrain where your only means of protection are your powers of jumping and stomping.
It’s a journey so arduous that no computer—real or hypothetical—is powerful enough to figure out if you can reach her. And according to research published by the MIT Hardness Group, determining whether your quest is possible at all is at least as complicated as decoding the encryption behind financial transactions. But if this problem could talk, the first thing it would say is “Hello, it’s a-me, Mario!”
For the love of the game
Though it does have a YouTube channel, the MIT Hardness Group isn’t an official research group. Instead, it’s a placeholder name for theoretical computer science projects—including several related to Super Mario—from Erik Demaine’s class Algorithmic Lower Bounds: Fun with Hardness Proofs.
Demaine, a professor of computer science, received a MacArthur fellowship (also known as a “genius” grant) for his work in computational geometry on protein folding and origami. But he also researches complexity theory, which focuses on organizing problems into categories based on how much time and memory space it takes for computers to solve them.
He happens to be an avid Super Mario fan as well. “I grew up playing NES [Nintendo Entertainment System] games,” Demaine says. “I poured many hours into playing as a kid, so it’s fun to come back to it these many years later and tie it into my research.”
Erik Demaine researches complexity theory, which examines the amount of time and memory that computers need to solve problems. He’s also an avid Super Mario fan.
DONNA COVENEY/MIT
Super Mario takes place on a horizontally scrolling universe of platforms, pipes, and other obstacles. The object of the game is to rescue Princess Peach, the monarch of the Mushroom Kingdom, by racing through this terrain while sidestepping or dueling monsters like Goombas and deadly porcupines called Spinies. The game takes place over several levels; in the original version, each level ends with a flagpole that sends Mario on to the next part of his mission.
Over the last 14 years, Demaine and his collaborators have proved many things about Super Mario, such as that it’s even harder than the infamous traveling-salesman problem (which seeks the most efficient route between many different locations) or the problem of factoring large numbers. But the result that surprised Demaine the most came from four of his students: Hayashi Ani ’21, MEng ’23; Holden Hall ’26; Ricardo Ruiz ’24, MEng ’25; and Naveen Venkat ’23, MEng ’24. For their final project in that 2023 class, the team used a combination of fan-made Super Mario level editors and a platform called Super Mario Maker to create levels so hard that they are undecidable. In other words, it’s impossible to write a computer program that always correctly predicts whether, in those levels, Mario can reach the castle.
Previously, Demaine had believed that Super Mario belonged in the PSPACE complexity class, which contains problems that are solvable but whose solutions become impractically complex as the problem gets bigger. At the time, he had even said that PSPACE was Mario’s “permanent home.” But the new findings pushed Super Mariointo RE-Complete, the class of undecidable problems. “It’s the hardest complexity class we could imagine for these sorts of games,” Demaine says.
What computers can’t solve
In 1936, Alan Turing, the father of modern computer science,created a puzzle now known as the Halting Problem to prove it’s not possible to construct a computer that can solve everything.
At the core of the Halting Problem lies a paradox, and it goes like this: Suppose you have a fancy computer, called the Oracle, that looks at any program and correctly determines whether a computer following it will ever come to a stop. For example, if it sees the program “Take 1 and add 3,” the Oracle will say the program halts, but if the program says “Take 1 and add 1 to it until it becomes 0,” the Oracle will say it runs forever.
Now suppose you have another computer, the Contrarian, and you put the Oracle inside it. When you give the Contrarian a program, it passes it to the Oracle and then does the opposite of whatever the Oracle says the program will do. So if the Oracle assesses the Contrarian’s program and thinks it will halt, the Contrarian will run forever. If the Oracle thinks the program will run forever, the Contrarian will halt. Either way, the Oracle’s assessment is wrong, so the classification problem is undecidable.
The proofs that Super Mario is undecidable rely on a more complex version of this idea. The team’s argument breaks down the video game using a technique called a reduction, in which mathematicians convert a problem they’re trying to solve into a problem they already know something about. “The classic example I remember in a math class is: How do you make a pot of boiling water?” Demaine recalls. “Well, I fill up the pot with water from the sink, and then I put it on the stove, and then it eventually boils. Okay, now I’ll give you a pot of water that’s already filled. How do you make a pot of boiling water? Well, I empty out the pot first and reduce to the previous problem.”
In their particular world of platforms and porcupines, the team broke down their Super Mariolevel into localized parts of Mario’s path called gadgets, which they could use to prove that the level was undecidable.
“A gadget in our sense is anything in your environment that decides whether or not you can go through one pattern [within a level],” explains Jayson Lynch ’12, MEng ’15, PhD ’20, a CSAIL research scientist and head of algorithms at MIT FutureTech. For example, in one gadget Mario might need to jump on a platform to avoid a monster as he makes his way across the screen. As a PhD student mentored by Demaine, Lynch spearheaded the formalization of gadget theory and worked on some of the earlier Super Mario papers but did not study the game’s undecidability.
One of Lynch’s favorite Super Mario gadgets is the door gadget, which works like a door that Mario can open, traverse, and close. The door in question is always either open (when the Spiny is on the right) or closed (when the Spiny is on the left). So if a Spiny is pacing back and forth on the left of the door, Mario has to navigate beneath the moving Spiny and jump up to hit a brick block just as the Spiny reaches it. This bumps the Spiny to the right side, which opens the door and allows Mario to travel across the traverse path and get to the spot where he can close the door. Once there, he must time another jump beneath the pacing Spiny to send it back to the left side of the gadget, closing the door behind him.
Mario opens the door by bumping the Spiny from the left to the right.With the Spiny out of the way, Mario can go through the open door and follow the traverse path to the other side. Once there, he’ll be able to bump the Spiny back to the left and close the door.
Since a door is always open or closed, its state can be used to simulate a true or false statement, with open being true and closed being false. Earlier Super Mario papers had strung together multiple door gadgets to simulate a true-or-false problem that complexity researchers already knew to be hard. But to show undecidability, the team used Super Mario level editors to put together another device, called a counter gadget, that tallies the game’s monsters and obstacles.
If you can build a machine with even just a few of those counters, Demaine says, you can simulate an arbitrary computer—one that could essentially do anything a non-quantum computer could do, given enough time and memory. And with no limit on the number of monsters, such a machine could have infinitely expandable memory, even though the level size stays the same, which he calls “pretty wild.” In other words, any theoretical computer can be built in a Super Mario level. “You could use it to solve anything you can use a computer to do,” says Demaine. “You could have it do your taxes, or compile your code, or run an LLM, or optimize your class schedule.” You might even build Super Mario levels that could excel at sudoku, construct optimal chess strategies, or prove any provable mathematical theorem.
The MIT mathematician Marvin Minsky invented counter machines in 1961 to figure out how simple a computer could be while still being “universal” (as powerful as any other computer, given enough time). These theoretical computers each store two numbers and can change them by adding 1, subtracting 1, or doing something special if a number hits a set value.
In the counter gadgets the students designed for Super Mario, the numbers reflect how many Goombas the levels contain. A number increases when a pipe spits out a Goomba and decreases when Mario stomps on one. Mario dies if he collides with a Goomba without stomping on it, so he can continue along the path only when the counter is at 0.
The MIT Hardness group designed this counter gadget in Super Mario Maker 1 to prove undecidability.
Minsky had already proved that counter machines are undecidable because they can run undecidable problems. Since the researchers proved that counter gadgets simulate counter machines, then any level of Super Mario containing a counter gadget will also be unsolvable. “In the future, if someone wants to show a game is undecidable,” explains Holden Hall, one of the students behind the project, “they just have to make one of these gadgets.”
The existence of undecidable problems like the Halting Problem implies that it’s possible to construct an undecidable Super Mario level. Just as the singular undecidable program for the Halting Problem meant thatit’s impossible to figure out if a computer program will run forever, the team’s undecidable level means that it is impossible to determine whether an arbitrary Mario level can be beaten.
Putting the “super” in Super Mario
More than two years after Demaine’s class on hardness proofs, some of his students continue to meet weekly to discuss their Super Marioresearch.
“From the point of view of complexity theory, studying video games is interesting mostly for didactical reasons,” Fabrizio Grandoni, a research professor at the University of Applied Sciences and Arts of Southern Switzerland, told MIT News in 2016. “It’s a simple, natural way to attract students to study this specific topic.”
Hall, who had very little exposure to the ideas of complexity theory before taking Demaine’s class, is a case in point, noting: “I took the class because a bunch of people I knew were taking it. But since I took it, I really enjoyed the class, and so I’ve taken a lot more classes in that realm.”
The applications of the MIT Hardness Group’s work go way beyond stomping on mushrooms and collecting coins. For example, researchers at the University of Texas Rio Grande Valley (including Timothy Gomez, now a PhD student at MIT) have used the gadget theory developed for analyzing games like Super Marioto study the complexity of problems relating to planning robotic motion and modeling chemical reaction networks.
“[Gadget theory] can be used in the negative way to say ‘Oh, well, we should stop searching for algorithms because we know this problem is too hard’—or it can be used in this positive way, because usually, to prove something hard, you’re showing that you can build a computer of a certain type,” Demaine says.
Though there’s no way of knowing what mark Super Mariowill leave on the future of math and computer science, one thing’s for sure: No matter how many princesses he does or doesn’t save, the legacy of this little plumber is set to extend far beyond video screens.
It’s somewhat surprising that he can choose just one. He’s the person spellers rely on to confirm pronunciations and answer questions about the roots of the words they’re given at the Scripps National Spelling Bee—arguably the world’s most prestigious competition of its kind. The story of how the word earned the top spot on his personal list may well mark the beginning of his unique career path as both a linguist and a Greek Orthodox priest.
In third grade, Sietsema ventured to a garage sale at a friend’s house with 50 cents in his pocket and picked out three books that struck his fancy. Although they were priced at 50 cents each, his friend’s mother said the books he’d chosen were on special and sent him home with all three, including a collection of Edgar Allan Poe stories called Masterpieces of Mystery. Knowing it contained macabre tales like “The Tell-Tale Heart,” his own mother told him he’d need to wait a few years before reading it. Naturally, he started it right away.
As he read “The Unparalleled Adventure of One Hans Pfaall,” Sietsema was baffled by the main character’s description of arriving at the moon in a balloon. Pfaall reported tumbling into a crowd of people who were “eyeing me and my balloon askant, with their arms set a-kimbo.” Sietsema had never encountered the word akimbo (with or without a hyphen) and asked his parents what it meant. They didn’t know, and it wasn’t in the family’s dictionary. The question also stumped his teachers, and the dictionaries in his classroom and the school library were no help either. “For years, I didn’t know what this word meant,” Sietsema says. It stuck in his mind that there was a word out there that he, his parents, and his teachers didn’t know. He thinks it wasn’t till he got to college that he finally found a dictionary with the answer: The moon dwellers in Poe’s story had been standing with their hands on their hips, elbows turned outward.
“I credit that puzzle with getting me into dictionaries and being curious about etymology,” he says. It kindled a fascination with words—and an abundance of curiosity—that would shape his life’s trajectory and work.
Growing up in Grand Rapids, Michigan, Sietsema attended a Dutch Reformed Christian school and recalls taking part in only one spelling bee, in second grade. It was in the 1970s, when everyone was hooked on phonics—so he overthought the sounding-it-out implications when asked to spell of. “I spelled it U-V, and of course, I was wrong,” he says.
At the time, he thought he probably wanted to work in the church—when he painted himself as an adult for a class project, he dressed his grown-up self in a cassock. But after taking a class in nuclear chemistry at the local junior college in high school, he decided his backup plan was to become a nuclear engineer. So when he went to the University of Michigan, he enrolled in the school of engineering. While he did well and liked his courses, though, he soon realized he felt called to a career in the church after all.
Sietsema (a.k.a. Father Mark) presides at the 2025 Holy Friday evening service at Holy Trinity Greek Orthodox Church in Lansing, Michigan. As part of that service, he sprinkles the congregation with rose-scented water, which delights the children. “It’s like a one-sided water fight in church,” he says. This service culminates with a procession in which a symbolic tomb is carried around the outside of the church. Everyone who attends takes part and leaves with a flower.
COURTESY OF BRIAN SIETSEMA
Switching to the college of literature, science, and the arts, he chose the studies in religion major, taking advantage of the interdisciplinary freedom it offered to take classes in literature, art, and more. He also tucked in courses that would fulfill seminary prerequisites such as knowledge of the biblical languages, studying ancient Hebrew and ancient Greek as well as modern languages that might come in handy for theological scholarship (Dutch, Swedish, and modern Hebrew).
Being in Ann Arbor gave Sietsema “a different understanding of the wideness of the Christian world,” as he puts it, and he gradually became less sure about which church he wanted to work in. As he neared the end of his fourth year at Michigan, he still needed a few more pre-seminary courses—and it dawned on him that he’d taken an “awful lot” of languages and thoroughly enjoyed them. So he stayed on for a fifth year to study linguistics as well as German, ancient Aramaic, and modern Arabic. One of his professors encouraged him to go to grad school and insisted that he apply to MIT, which was considered the top linguistics program in the country. To his surprise, he got in.
Sietsema calls his four years at MIT a great adventure: “If I could relive them, I would empty out my bank accounts to do so.”
At MIT he worked with Morris Halle, one of the leaders in generative grammar, which Sietsema describes as a working model of the “chemistry” of language—the parts and processes that form the building blocks of verbal communication. Halle and others had developed counting procedures (akin to measured time in music) that help explain stress patterns (that is, which syllables might receive emphasis by varying such things as stress or pitch). Building on that work, Sietsema’s dissertation proposed that the division of words and phrases into metrical units similar to musical measures can be used to predict where high and low tones fall, which he demonstrated in the tonal patterns of four Bantu languages spoken in Tanzania. At the time, research in this area was seen to have implications for creating natural-sounding machine- generated speech.
Sietsema calls Halle “a wonderful mentor,” and the two played well off one another. As he was sweltering in his Central Square apartment while printing the final version of his dissertation, Halle called and asked him to stop by. Knowing that Sietsema read Hebrew, Halle, a Latvian-born Jew who’d learned English as his sixth language, wanted to show him a syllable-counting analysis of the 23rd Psalm he’d just completed; Sietsema answered with his own structural analysis of Psalm 90. “I could tell he was delighted to have this young Gentile boy from Grand Rapids, Michigan, who had the same fascination for biblical Hebrew as he did,” Sietsema says.
Today, he calls his four horizon-expanding years at MIT a great adventure: “If I could relive them, I would empty out my bank accounts to do so.” Beyond embracing the intellectual stimulation of the Institute, he took advantage of Cambridge’s many cultural opportunities and cross-registered at Harvard to study French and Ugaritic. All told, he says, he’s now studied about a dozen languages, including the Latin he took in high school and the modern Greek he would add to his repertoire several years after earning his doctorate. (“I always feel like I’m leaving one out,” he says.)
When Sietsema graduated from MIT in 1989, the job market for linguists was “not great.” As fate would have it, though, Matt Alexander, PhD ’92, his best friend at MIT, had already been hired at the University of Michigan, where a one-year position as a visiting assistant professor of phonology opened up that spring. Alexander recommended Sietsema, who handed in his dissertation and got the job, earning an award for excellence in teaching based on student reviews in his first semester.
Shortly after his one-year gig at Michigan ended, he returned to Massachusetts and landed a job as pronunciation editor at Merriam-Webster in Springfield. Although the work was very different from the theoretical linguistics he’d focused on in grad school, “as the guy who had studied a whole bunch of language back in undergrad, it was kind of coming home to old-school philology,” he says. His main job was to ensure that pronunciations—which can change—were up to date. Fluoride, for example, shifted from floo-o-ride in the early 1900s to flor-ide in the second half of the century.
At Merriam-Webster, he made the call on which pronunciations would go into the 10th edition of Merriam-Webster’s Collegiate Dictionary—and in what order of preference. The dictionary, he explains, takes a descriptivist approach that reflects common word usage, so he kept a radio and a TV on in the background as he worked. He’d listen for interesting pronunciations and record them on index cards, noting how each such word was said, who said it, where the person was from, and what the context was. These went into Merriam-Webster’s “huge files” of index cards containing citations of words in actual usage.
Sietsema also had a hand in identifying new words and usages that appeared in the 10th edition, which was initially released in 1993—and he was responsible for the inclusion of definitions for interjectional uses of like. He recognized three informal uses: to introduce a quotation (“So she was like, ‘Let’s go eat’”); to give an approximation (“There were like 10 people in line”); and to emphasize (“He was, like, gorgeous”) or convey something apologetically or vaguely (“I need to, like, borrow some money”). While not a fan of such usages, he recognized them as real linguistic phenomena that had earned a place in the dictionary.
During his tenure as pronunciation editor, he introduced the use of the International Phonetic Alphabet (a standard phonetic notation for all languages) into Merriam-Webster publications long before it became widely used in American mass-market dictionaries. He also oversaw the recording of pronunciations for digital versions of the dictionary and flew out to a San Diego recording studio to supervise the voice actors. When the actors refused to record certain words that offended them, Sietsema had to step into the breach and do it himself. If you go to www.merriam-webster.com and search for a choice two-part expletive the actor Samuel L. Jackson is famous for delivering, it will be his voice that you hear when you click on the icon of the speaker—offering a decidedly less memorable rendition.
Working at Merriam-Webster gave Sietsema access to what he describes as its “fantastic library of old books on every subject imaginable.” He seized the opportunity to delve into historical questions about the development of Christianity—something he’d been curious about. It struck him that Orthodox Christianity was the most original form of the faith that was still around. Having met Katherine Chapekis, a young linguist raised in the Greek Orthodox tradition, during his year teaching in Ann Arbor also nudged him in the direction of Orthodoxy. In 1991, he converted and they married, and she began working at Merriam-Webster the following year as a definer and researcher who tracked down first usages of English words.
After 15 years of answering etymological queries, when the bee was expanded in 2018 Sietsema began serving as a pronouncer for some of the earlier rounds as well.
AP PHOTO/CAROLYN KASTER
At the Greek Orthodox church in Springfield, Sietsema’s facility for languages proved useful when he served as a volunteer chanter, helping the priest lead services in Greek. “I do a good job with the liturgical Greek because I have the phonological knowledge to know how to make my mouth do the things that it needs to do to sound like authentic Greek speech as opposed to an American just rattling off Greek letters,” he says.
He began taking evening classes in Byzantine chant, and before long the bishop was encouraging him to attend seminary. Merriam-Webster allowed him to work four 10-hour days so he could commute to Brookline to study at the Holy Cross Greek Orthodox School of Theology. And after four years, he earned a master of divinity degree.
Sietsema fully intended to go back to being a lexicographer, perhaps eventually getting ordained so he could serve as a substitute priest on weekends. But he’d made what he jokingly calls “a terrible mistake” at the seminary: He’d embraced his studies so enthusiastically that he became the valedictorian and had to give the commencement speech. The archbishop of America—the head of the Greek Orthodox church in the US—came up from New York to attend the ceremony, and he happened to be in need of a deacon who could also serve as a speechwriter. “A few weeks later, I got a call from the archdiocese saying ‘We want you to be ordained, and we want you to come to New York, and we want you to write for the archbishop,’” Sietsema recalls.
In short order, he and his wife moved to the Upper East Side of Manhattan so he could begin his new post as Father Mark (he used his middle name because Orthodox priests must be ordained with a saint’s name, and there are no Orthodox Saint Brians). As deacon to the archbishop and then to his successor, he wrote their speeches and encyclicals on top of many other duties—including chauffeuring them through New York City traffic—and traveled with them around the country and to Greece, meeting President Clinton, ambassadors, members of Congress, Elie Wiesel, and South Africa’s Anglican Archbishop Desmond Tutu along the way. But after two years, as the father of a newborn, he was eager to move on from a job that required putting in as many as 14 hours six or seven days a week. So in 2000, he returned to Michigan to become pastor of the Holy Trinity Greek Orthodox Church in Lansing.
“The World Series can be a four-game sweep and the Super Bowl can be a blowout, but the National Spelling Bee always comes down to one last word.”
Not long after settling into parish life, Sietsema got an unexpected call from the Scripps Spelling Bee. His wife had served on the event’s word panel from 1997 to 2000, and he had traveled with her to one of the members’ off-site gatherings in 1998. He’d tagged along to dinner one night, and they were pleased to meet the person who was responsible for pronunciations in the bee’s official dictionary. But now, just a few weeks before the 2003 event, there was a crisis: The longtime pronouncer had suddenly died. The veteran associate pronouncer would step into his role and take on the job of giving spellers their words, but a new associate pronouncer would be needed to answer spellers’ questions about word roots, monitor pronunciations, and be prepared to serve as the pronouncer if needed. Could he do it? Honored to be asked, Sietsema got the okay from his bishop and said yes.
Little did he know it would become a permanent gig. After 15 years of answering root-word queries, when the bee expanded in 2018 he began serving as a pronouncer for some of the earlier rounds as well—though never for the finals. Now he’s the head of a team of associate pronouncers. “It’s just wonderful to see these young people blossom right in front of you, asking their questions and analyzing the word on the spot and figuring out how it all goes together,” he says. He dismisses the idea that the kids have photographic memories, saying they’re “really just good little word detectives.”
As a member of the bee’s word panel, Sietsema attends multiple daylong meetings to create and fine-tune each year’s list by mining the 500,000 or so words in Merriam-Webster’s unabridged dictionary. “For an introductory round, you want something that’s an interesting word, a useful word, but something that’s gettable,” he says. “For the later rounds, you really want to find something that’s going to challenge the speller. And it’s nice to have a word that’s analyzable.” “Rooty” words—those with obvious roots—are ideal.
The advent of unabridged online dictionaries has streamlined how students prepare for the bee, which once required wading through the dictionary manually to compile word lists. Today, it’s easy to generate lists of words derived from a particular language to study their roots, for example. Meanwhile, the competition has become increasingly fierce, and once-verboten terms like geographical names are considered fair game. For some of the words in the hardest rounds, “it looks like you’re just taking a spoonful of alphabet soup,” he says. “And that’s for the spellers who really, really are committed to learning just about every word they can in the dictionary.”
When it gets down to the last spellers in the final round, there’s an electric feeling in the room. “It’s always a close competition,” he says. “The World Series can be a four-game sweep and the Super Bowl can be a blowout, but the National Spelling Bee always comes down to one last word, and that’s what makes it exciting each and every time.”
The philosopher Friedrich Nietzsche famously wrote that a characteristic of theologians is their “unfitness for philology,” meaning they can’t be trusted to interpret texts with objective accuracy. He also maintained that a sense of restraint characterizes a good linguist. Sietsema says he’s right on both counts. When linguists analyze texts, “we know what we don’t know, and that’s important because you don’t find meaning where it’s not in the original,” he says. He thinks the well-trained linguist has a mission to the world of theology: to help clarify what is an appropriate interpretation of a sacred text and what is going too far.
He’s put his unique blend of skills into practice. In the early days of the covid pandemic, for instance, a Greek Orthodox scholar defended the practice of continuing to use a single spoon to administer communion. The scholar argued that holy things cannot cause harm and that abandoning them for fear of an earthly disease was far more dangerous than the disease itself, citing a passage from a homily of an archbishop of Constantinople in the fourth century CE saying “nothing is worse than to relegate spiritual things to human reasoning.” Sietsema responded with a thoughtful defense of reason, pointing out that the scholar’s argument relied on a mistranslation of logismoi, which he explained refers not to the faculty of reason but to negative mental habits, such as flawed reckonings, intrusive thoughts, or vain rationalizations. He countered that the church very much values reason and advocated “the exercise of good sense, good science, and compassion,” arguing that “those who pit faith against the faculty of reason end up losing one or the other or both.”
Sietsema’s time at MIT, he says, taught him to pay attention not only to what’s in data sets but also to what’s not there that could be. “That particular muscle gets used in both linguistic analysis and lexicography, as well as in pastoral care,” he says. “When you’re listening to people pour out their hearts, it’s important to notice what they’re saying and what they’re not saying.”
During his time as a deacon at the Archdiocese in New York City, Sietsema stands alongside Archbishop Iakovos, the head of the Greek Orthodox church in the US, at a water blessing service attended by former President George H.W. Bush.
COURTESY OF BRIAN SIETSEMA
As both a priest and a linguist, he’s called on to notice and remember. Attention to detail matters whether he’s gearing up for the celebration of Pascha, or Easter, at Holy Trinity or preparing for the National Spelling Bee, which he calls “the holy week of spelling.”
This spring, before heading to Washington for his 24th National Spelling Bee in May, Sietsema reflected on what words he might add to his list of favorites. A top candidate was one given to Evelyn Blacklock, a speller in his first bee as associate pronouncer in 2003: clepsydra, meaning an old-style water clock. “She didn’t know it, but through a series of questions to me about the Greek roots of the word—from kleptein (to steal) and hydōr (water)—she was able to divine the English spelling,” he recalls. “It was so satisfying to watch this feat of word sleuthing happen in real time, and it gave me a good insight into the importance of my role at the bee.”
It seems unlikely, however, that akimbo will ever lose top billing on his list. It’s easy to imagine Sietsema facing the future with his own hands on hips, elbows out, embracing linguistics, theology, and scientific reason as he shares his joy for life and the words we use to describe it.
The national conversation about the value of education is currently dominated by speculation about the risks and positive potential of AI.
Whatever your own perspective on that debate, I hope you’ll be glad to know that MIT is also working on a deeply important but comparatively old-fashioned challenge: American high school students’ startlingly uneven access to calculus. According to the National Survey of Science and Mathematics Education, which covers the nation’s more than 13,000 school districts, in almost half of US high schools calculus isn’t even offered.
As our graduates know better than anyone, preparation in calculus is effectively an admissions requirement at a place like MIT—which means that students in schools with no calculus classes are in practice locked out of an essential route to STEM careers.
Recognizing this glaring need, we set out to find a solution. With support and inspiration from the Siegel Family Foundation, in the fall of 2025 the Institute launched the MIT4America Calculus Project. Developed by the MIT Scheller Teacher Education Program (STEP) Lab, the Calculus Project recruits and trains MIT undergraduates and alumni to provide weekly long-distance calculus tutoring for students in underresourced high schools across the country.
Reflecting the Institute’s longstanding commitment to national service, the MIT4America Calculus Project supplies an innovative answer to a hard practical problem, and it taps the uncommon skill of MIT’s people to create opportunity for others and spread the educational impact of the Institute beyond our walls.
The project is in its early phases, so far engaging 30 MIT undergraduates and seven alumni tutors. From its initial work with 14 school districts across the country, it’s on track to collaborate with about 20 this summer.
The demand is clear—and the response from the students we’re reaching makes it all worthwhile. This spring, the first Calculus Project students were prepared for their AP exams, thanks to their own persistence, diligence, and curiosity—and to the generosity, care, and patience of a dedicated group of people from MIT.
Login
