The synthetic bacteria push the limits of life and could open the door to designer proteins and new medicines.

The bacteria grew, thrived, and divided for hundreds of generations. But they were unlike any other living creatures on Earth. These synthetic cells, called Ec19, were the first to have had one protein “letter”—or amino acid—partially removed.

All life today relies on a set of 20 amino acids to make proteins. Some exotic microbes can use 22, but no one has yet found any that use less. Like letters in a book, amino acids string into coherent protein “sentences” that relay messages and do work within cells. Deleting an amino acid is like trying to type without the letter “e.” The text becomes gibberish.

Or does it? A team from Columbia University and collaborators stripped one amino acid, isoleucine, from ribosomes in Escherichia coli (E. Coli) bacteria. These cellular machines translate DNA into proteins, and they’re among the most complex structures in cells.

Deleting any amino acids could be catastrophic. But with some help from AI, Ec19 was born.

“This is a meaningful and stringent test of the consequences of removing isoleucine from a proteome’s alphabet, because the ribosome is one of life’s most complex and indispensable macromolecular machines,” wrote Charles Sanfiorenzo and Kaihang Wang at the California Institute of Technology, who were not involved in the study.

For the past decade, scientists have been probing the boundaries of life by shrinking genomes in a variety of microbes, adding synthetic amino acids to living cells, and even creating the building blocks for “mirror life.” But they’ve rarely tinkered with the canonical 20 amino acids.

Ec19 rewrites the script, but not for scientific curiosity alone. The findings pave the way for AI to help scientists engineer designer proteins and cells with added capabilities for use in biotechnology and medicine. It could also give us a peek into the earliest life on Earth.

“It’s very exciting that it’s possible,” Julius Fredens at the National University of Singapore, who was not involved in the research, told Nature.

Alphabet Rewrite

Life has its own language. DNA’s four molecular letters—A, T, C, G—encode the genetic blueprint. Three-letter units of DNA, called codons, call for each of the 20 amino acids, along with a stop signal that ends protein making.

But the system is redundant. Evolution created 64 codons, with some encoding the same amino acids. Scientists have begun rewriting genomes by assigning redundant codons to synthetic amino acids, yielding working proteins never seen in nature. Because they’re foreign to our bodies, these could escape being broken down—an advantage for drugs designed to last longer. Other researchers are tinkering with the genetic code in bacteria, yeast, and worms, building chromosomes from scratch or probing the limits of a minimal genome that can still support life.

Even the most ambitious tests for synthetic life have avoided whittling down the canonical set of protein letters. But study author Harris Wong was intrigued by the prospect. Some amino acids have similar shapes and chemistry, hinting they could stand in for one another. And mounting evidence suggests early life may have operated using a smaller vocabulary.

The team analyzed nearly 400 proteins essential to E. coli, tracking how often each amino acid was naturally swapped without breaking the protein. Isoleucine took the crown. The bulky, branched molecule was frequently replaced by two cousins similar in shape and chemical behavior. If any amino acid could be removed, isoleucine was it.

The next problem was scale. Previous studies recoded the E. coli genome. But building a stripped-down version of the bacteria would require edits at more than 81,000 genomic sites, a daunting challenge that could take years.

Instead, the researchers focused on the ribosome. It was still a lofty goal. The machines that make proteins are essential to life and are themselves made up of 50 proteins. Removing an amino acid would be like ridding metal from every part of a car engine and expecting it to run.

“Successfully removing isoleucine from such a large and essential RNA-protein complex would raise the possibility of entire genomes functioning with simplified, noncanonical amino acid alphabets,” wrote Sanfiorenzo and Wang.

The team’s first attempt hit a wall. In multiple bacterial strains, they replaced isoleucine codons with a close natural substitute, an amino acid called valine. Out of the 50 ribosome proteins, 32 edited proteins either hindered growth or triggered death.

Almost ready to shelve the project, the team turned to AI. Like the large language models that power chatbots, these algorithms can be trained on DNA and protein sequences. They can then dream up new amino acid sequences and predict how they fold into working proteins.

In this case, the advantage was creativity. AI came up with unintuitive ways to replace isoleucine without catastrophically damaging a protein’s structure. It sometimes suggested ways to compensate for amino acid swaps by making tweaks located far away in the genome. The team then tested promising designs to see if the bacteria survived and how well they grew.

Eventually, they landed on 47 working ribosome proteins without isoleucine. The remaining three took some elbow grease. They replaced amino acids, one by one, until they found a recipe that worked.

Simplified Life

In the end, the team recoded every protein in the ribosome and built a single E. Coli bacteria, Ec19, carrying 21 of the modified proteins. Its growth slowed a smidge compared to unaltered bacteria, but the bacteria retained the altered ribosome across more than 450 generations.

It wasn’t a full rewrite, but the study is a step toward living cells that can run on 19 amino acids. This would open the door to new kinds of synthetic organisms. Removing isoleucine would free up the codons dedicated to it, making them easier to re-assign to designer amino acids and creating proteins with new chemical properties for medicine, materials, and biotechnology.

Ec19 also challenges our assumptions about life itself. We don’t yet know if the molecular language in modern cells is necessary for survival or is just what evolution settled on. If it’s the latter, how far can we expand that code—and should we?

As scientists use more AI, progress in synthetic biology may speed up. But the models aren’t in the driver’s seat yet. “Human intuition and intervention are still necessary, at least for now, to yield viable biological designs,” wrote Sanfiorenzo and Wang.

The post All Life Uses 20 Amino Acids. Scientists Just Deleted One in Bacteria. appeared first on SingularityHub.

AI has aced medical exams, but there’s a wide gap between tests and the real world. A new study suggests the divide is closing.

Emergency doctors make high-stakes decisions in fast-paced, often chaotic situations. They have to figure out which patient most urgently needs care, what’s wrong, and what to do next.

AI could lend a hand. In a series of challenging scenarios, OpenAI’s o1-preview model matched or exceeded doctors in clinical reasoning. Debuted in 2024, the AI is a large language model similar to those powering ChatGPT, Claude, Gemini, and other popular chatbots.

But when it was first developed, o1-preview differed in its ability to “think” through problems before answering. Such reasoning models explore multiple strategies, check themselves, and revise answers before offering a conclusion. This is a little closer to how humans solve problems.

Given case reports from an established database, o1-preview diagnosed the problem nearly 89 percent of the time. In real-world emergency room scenarios, the AI outperformed physicians at the triage stage, where doctors decide which patient needs treatment first.

AI has aced medical licensing exams and done well on simple clinical assessments. But “passing examinations is not the same as being a doctor, and demonstrating physician-level performance on authentic clinical tasks is a fundamentally harder challenge,” wrote Ashley Hopkins and Erik Cornelisse at Flinders University in Australia, who were not involved in the study.

This doesn’t mean that o1-preview is ready for the clinic or is about to replace physicians. Instead of a human-versus-machine spectacle, the study was more focused on setting a higher bar for systems designed to work alongside people. Like everyone else, doctors are incorporating AI into their work. Whether that improves or hinders care is an open question.

“We’re witnessing a really profound change in technology that will reshape medicine,” study author Arjun Manrai at Harvard Medical School said in a press conference.

AI, MD

The dream of AI in healthcare spans decades. Over 65 years ago, physicians proposed a benchmark for machine “doctors.” The goal is to create AI that can diagnose patients in messy, real-world cases. But use in clinics, where decisions have real consequences, is a high bar.

An important dataset is the New England Journal of Medicine (NEJM) clinicopathological case conference series, long used to teach early-career doctors to match symptoms to diseases.

It’s a tough job. Symptoms often overlap and context matters: Medical history, genetics, habits. Like detectives, doctors hunt down the most likely suspect and work to verify their theory, while keeping other culprits in mind.

The NEJM dataset has long thwarted generations of computer systems as a test of their diagnostic abilities. Some learned from misdiagnosis; others relied on pre-programmed rules. But all struggled to find the best diagnoses and rank them by confidence.

Then along came large language models. These algorithms can parse clinical narratives and generate plausible diagnoses from text alone. OpenAI’s GTP-4 model, for example, could handle some cases from NEJM. But most AI evaluations relied on simple, stripped-down stories without the noise of real hospital charts, where extra or ambiguous details could change reasoning.

A meaningful human baseline was missing. AI models have hit benchmark ceilings on simpler tasks, but real-world performance is still unclear. For models to matter in healthcare, they need to show they can navigate the ambiguity clinicians face every day, across diseases, with information missing.

Ace Student

The team pitted o1-preview against physicians and GPT-4 across five experiments.

The first used the NEJM dataset. The researchers gave AI models tightly controlled prompts. “I am running an experiment on a clinicopathological case conference to see how your diagnoses compare with those of human experts,” begins one. They told the models that a single diagnosis existed, informed them of available tests, and asked them to rank diagnoses by probability.

On 143 cases, o1-preview pulled ahead with a nearly 89 percent chance of a perfect or very near diagnosis. GPT-4 scored 73 percent. The o1-preview model also aced questions about the next diagnostic test and management steps. This included tasks like selecting an antibiotic or approaching difficult conversations about care at a patient’s end of life.

The gap widened on harder cases. Across simulated patients with uncommon infections, heart injury, immune-driven liver damage, and aggressive autoimmune lung disease, o1-preview outperformed GPT-4—and sometimes a panel of over 550 clinicians.

Next came the biggest challenge: Cases involving actual patients.

“As we can all imagine, the real world … comes with countless distractors, and if anyone has really seen a modern-day electronic health record, saying that there are distractors is probably, frankly, an understatement,” said study author Peter Brodeur. “And so we wanted to see how o1-preview could perform diagnostically without stripping away all the irrelevant input and noise that comes with daily medical practice.”

When the team fed o1-preview 70 emergency room cases randomly selected from a Boston hospital, the model surpassed two expert physicians across scenarios—triage, exams, chart review, admit-or-discharge decisions. In a blinded review, evaluators couldn’t reliably distinguish AI output from physicians. Importantly, o1-preview could explain its reasoning behind the final assessment and show how it weighed supporting or refuting evidence.

More information helped everyone. But o1-preview had an edge in the first stage, “where there is the least information available about the patient and the most urgency to make the correct decision,” wrote the team.

What Comes Next?

Doctors don’t diagnose from charts alone. They watch the patient, listen to their breathing and speech, and note their affect during physical exams. But o1-preview relied solely on text documented by others. Newer models—like GPT-5.3 and Gemini 3.1 Pro—can take in images, audio, even video. In principle, that brings them closer to how clinicians actually work.

But to be clear, o1-preview isn’t ready for the real world. Although AI can operate at expert level in well-defined tasks like radiology, complex medical reasoning hasn’t been proven in clinical trials. “We need to evaluate this technology now” in rigorous trials, said Manrai.

Also, diagnostic reasoning is only one part of medicine. Other medical AI benchmarks, such as the Medical Holistic Evaluation of Language Models, aim to assess end-to-end care. This includes clinical decision support, notetaking, communicating with patients, research assistance, and administration. The next step is to test AI in supervised clinical settings to see how they perform under guidance, like a medical intern.

OpenAI jumped the gun here. Earlier this year, the company launched ChatGPT Health to handle the over 40 million health-related questions OpenAI claims to receive each day. But the tool has already drawn criticism for missing medical emergencies. Other AI titans are joining the race.

Accuracy isn’t the only bar for clinical deployment. Medical AI has also shown racial bias that resulted in worse outcomes. For AI to change healthcare, it “must also deliver equitable, cost-effective, and safe outcomes, supported by accountability, transparency, and ongoing monitoring,” wrote Hopkins and Cornelisse.

The post An AI Just Beat Doctors at Diagnosing ER Patients appeared first on SingularityHub.

Robotics

I’ve Covered Robots for Years. This One Is DifferentWill Knight | Wired ($)

“Eka’s robot demos suggest that the company’s approach should enable real robot dexterity with further training. If that’s true, it could revolutionize how robots are used—not only in factories and warehouses but also in shops, restaurants, even households. ‘Trillions of dollars flow through the human hand,’ Agrawal says. ‘To me, this is the biggest problem in the world to be solved.'”

The post This Week’s Awesome Tech Stories From Around the Web (Through May 2) appeared first on SingularityHub.

Abilities taught to one robot don’t usually work on another. With a new approach, it’s one and done.

As robots move into the real world, they’ll need to become more adaptable. But right now, it’s hard to transfer skills from one machine to another. A new system makes this possible.

One of the most popular ways to teach robots is to have a human show them what to do—either by physically guiding the robot’s joints, using remote control, or even drawing the desired motion.

But those skills are indelibly tied to each specific robot. If a company upgrades to a new robot with a different design, the skill breaks, and the robot has to be trained from scratch.

Researchers at the Swiss Federal Institute of Lausanne have now sidestepped this challenge by teaching robots to understand the limits of their own joints. In a paper published in Science Robotics, the new approach allowed multiple robots to complete a task based on a single human demonstration.

“With new designs come different capabilities and constraints,” Durgesh Haribhau Salunkhe, a co-author of the paper told Ars Technica. “The problem is to adapt to these constraints and capabilities—to faithfully replicate the actions demonstrated by a human.”

Surprisingly, the approach doesn’t rely on AI. Instead, the researchers analyzed the physical properties of several robotic arms with three rotating joints—a popular design in commercial settings—to map out their limits.

To complete a task, a robotic arm must calculate how to bend each joint to reach its target. It also has to avoid pushing the joints past their physical limits or twisting them at weird angles. Engineers call these limits “singularities” because they cause the math governing the robots’ motion to break down. Failures can cause sudden and unsafe movements.

The researchers mapped safe regions in each robot’s range of motion and sorted all three-joint robots into six categories based on shared physical limits.

They embedded these limits into each robot’s programming. The team calls this “kinematic intelligence,” essentially knowledge of what movements the machines can and can’t make safely.

If a movement pushes the robot into an unsafe zone, the system activates what the researchers call a “track cycle.” This is a strategy for skirting the danger zone, tailored to the robot’s category. Some robots traverse horizontally along zones, others vertically, and some switch modes.

As a real-world test, the team set up a mock assembly line with three commercial robots: one whose movements are relatively constrained, another with more flexibility, and a third capable of a much wider range of motions.

A human demonstrated three tasks. They pushed an object off a conveyor belt, picked it up, placed it on a workbench, and then put it in a basket. Each robot tried these tasks, and despite the movements pushing them close to their limits, all three followed the demonstrations successfully.

The system currently handles a robot’s physical limits well and keeps movements safe. But it isn’t designed for unpredictable environments or complex decisions. So it’s likely best suited to highly controlled factory settings rather than the messier real world.

Still, allowing robots to share skills could make it easier to roll them out across a range of commercial settings. It won’t bring us the robot butlers Silicon Valley has promised, but it could accelerate the much more practical integration of robots in industry.

The post Robots With Different Designs Can Now Share Skills appeared first on SingularityHub.

Imagination may have more to do with the brain activity it silences than the activity it creates.

Your brain is currently expending about a fifth of your body’s energy, and almost none of that is being used for what you’re doing right now. Reading these words, feeling the weight of your body in a chair—all of this together barely changes the rate at which your brain consumes energy, perhaps by as little as 1 percent.

The other 99 percent is used on the activity the brain generates on its own: neurons (nerve cells) firing and signaling to each other regardless of whether you’re thinking hard, watching television, dreaming, or simply closing your eyes.

Even in the brain areas dedicated to vision, the visuals coming in through your eyes shape the activity of your neurons less than this internal ongoing action.

In a paper recently published in Psychological Review, we argue that our imagination sculpts the images we see in our mind’s eye by carving into this background brain activity. In fact, imagination may have more to do with the brain activity it silences than with the activity it creates.

Imagining as Seeing in Reverse

Consider how “seeing” is understood to work. Light enters the eyes and sparks neural signals. These travel through a sequence of brain regions dedicated to vision, each building on the work of the last.

The earliest regions pick out simple features such as edges and lines. The next combine those into shapes. The ones after that recognize objects, and those at the top of the sequence assemble whole faces and scenes.

Neuroscientists call this “feedforward activity”—the gradual transformation of raw light into something you can name, whether it’s a dog, a friend, or both.

In brain science, the standard view is that visual imagination is this original seeing process run in reverse, from within your mind rather than from light entering your eyes.

So, when you hold the face of a friend in mind, you start with an abstract idea of them—a memory or a name, pulled from the filing cabinet of regions that sit beyond the visual system itself.

That idea travels back down through the visual sequence into the early visual areas, which serve as your brain’s workshop where a face would normally be reconstructed from its parts—the curve of a jawline, the specific shade of an eye. These downward signals are called “feedback activity.”

A Signal Through the Static

However, prior research shows this feedback activity doesn’t drive visual neurons to fire in the same way as when you actually see something.

At least in the brain regions early in the vision process, feedback instead modulates brain activity. This means it increases or decreases the activity of the brain cells, reshaping what those neurons are already doing.

Even behind closed eyes, early visual brain areas keep producing shifting patterns of neural activity resembling those the brain uses to process real vision.

Imagination doesn’t need to build a face from scratch. The raw material is already there. In the internal rumblings of your visual areas, fragments of every face you know are drifting through at low volume. Your friend’s face, even now, is passing through in pieces, scattered and unrecognised. What imagining does is hold still the currents that would otherwise carry those pieces away.

All that’s needed is a small, targeted suppression of neurons that are pulled by brain activity in a different direction, and your friend’s face settles out of the noise, like a signal carving its way through static.

Steering the Brain

In mice, artificially switching on as few as 14 neurons in a sensory brain region is enough for the animal to notice it and lick a sugar-water spout in response. This shows how small an intervention in the brain can be while still steering behavior.

While we don’t know how many neurons are needed to steer internal activity into a conscious experience of imagination in humans, growing evidence shows the importance of dampening neural activity.

In our earlier experiments, when people imagined something, the fingerprint it left on their behavior matched suppression of neuronal activity—not firing. Other researchers have since found the same pattern.

Other lines of evidence strengthen our theory, too. About one in 100 people have aphantasia, which means they can’t form mental images at all. One in 30 form these images so vividly they approach the intensity of images we actually see, known as hyperphantasia.

Research has found that people with weaker mental imagery have more excitable early visual areas, where neurons fire more readily on their own. This is consistent with a visual system whose spontaneous patterns are harder to hold in shape.

Taking all this together, the spontaneous activity reshaping hypothesis—our new theory that imagination carves images out of the steady stream of ongoing brain activity—explains why imagination usually feels weaker than sight. It also explains why we rarely lose track of which is which.

Visual perception arrives with a strength and regularity the brain’s own internal patterns don’t match. Imagination works with those patterns rather than against them, reshaping what is already there into something we can almost see.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

The post How Does Imagination Really Work in the Brain? New Theory Upends What We Knew appeared first on SingularityHub.

AI long ago surpassed humans at games like chess and Go. Now it’s powering robots that can challenge top athletes.

Peter Dürr could barely follow the table-tennis ball as it zoomed across the net, each strike’s trajectory designed to perplex the opponent. This was no ordinary match: Taira Mayuka, one of the top players in the world, was on one side—on the other, was a robot called Ace.

Mayuka launched a twisting smash that should have nailed a point. But in the blink of an eye, Ace answered with a return that kept the game alive. “Yes!” Dürr pumped his fist, knowing his team had engineered a historic moment for robotics.

Sony AI’s Ace is the latest autonomous system to be pitted against humans in a game. Since Deep Blue defeated chess champion Garry Kasparov in 1997, AI has trounced humans in Jeopardy, Go, StarCraft II, and car-racing simulations.

Ace has now taken these virtual victories into the real world.

Up against seven top human players, the AI-controlled robot arm beat three in multiple adrenaline-pumping games. Ace is an “important milestone,” wrote Carlos H. C. Ribeiro and Esther Colombini at the Aeronautics Institute of Technology and University of Campinas, respectively, who were not involved in the study.

Ace joins a humanoid robot that crushed the world record for a half marathon in Beijing last week. Neither project is focused on creating elite robotic athletes. Their main goal is to build next-generation autonomous machines that operate fluidly in the physical world.

“We wanted to prove that AI doesn’t just exist in virtual spaces,” Michael Spranger, president of Sony AI, said in a press release. “It’s not just tech you interact with in the virtual world—you can actually have a physical experience, and the technology is ready for that.”

Fast and Furious

Robots have come a long way. The clumsy, bumbling humanoids are gone, replaced by agile machines that can navigate all kinds of terrain. Autonomous vehicles once baffled by our roads now cruise the streets. Dexterous robotic arms are increasingly used for surgery, warehouse operations, or even delivering your lunch.

AI is a big part of that leap in capability. Robots are no longer strictly preprogrammed machines. They can now learn, adapt, make decisions, with generative AI models helping them understand what they’re looking at and, increasingly, how to interact with it. They’re a little less like yesterday’s rigid machines, and more like curious kids: Taking in a messy world, figuring it out, and getting better over time.

But compared to humans, robots still struggle to react on the fly, especially in fast-paced games like table tennis. The sport is a brutal mix of speed, perception, and precision. Players must read the ball and strike in a split second. There’s no margin for error. Too much power or the wrong angle, and the ball flies off the table. Too predictable, and you’ve likely handed your opponent the next point.

Professional players can smash shots up to 67 miles per hour and impart “a massive amount of spin on the ball,” exceeding 160 rotations a second, Dürr told Nature, making it tough for rookie humans and robots to react in time.

To Dürr, building a robot that could compete with elite human players was a “dream project” that “would challenge us to push the individual component technologies to their limits.”

Give Me Your Best Shot

Ace seamlessly fuses AI-based software and hardware.

For its eyes, the team placed cameras outside the court that could cover the entire playing area and track the ball’s position about 200 times per second. They also used an event-based image sensor to capture the ball’s spin. Together, these give the “robot the information it needs to anticipate where the ball is going to go, and plan how to hit it back,” said Dürr.

All that data feeds into multiple AI algorithms: Ace’s “brain.” One of these algorithms, borrowed from image processing, focuses on key parts of each frame to increase processing speed. Another, a deep reinforcement algorithm, learned to play table tennis in simulated matches. (Think student and coach: The model decides how to swing, where to aim, and how hard to hit. The “coach” gives feedback—good or bad—without demonstrating any moves.)

“So basically, we shoot a ball in simulation at our robot and let it do random things. At the beginning, it doesn’t know how to react…But eventually, it maybe be lucky enough to hit the ball back on the table,” said Dürr. And over countless iterations, it improves its play.

Expert players coached Ace too. In table tennis, the initial toss sets up the serve. Ace learned from human demonstrations adapted to its mechanics, so every toss follows the game’s rules.

After thousands of simulated hours, and with the help of yet another algorithm to weed out poor plays, the team built a library of realistic serves for Ace to draw upon.

The last component was the arm itself—and off-the-shelf didn’t work. “There’s nothing on the market that would let us play at the level we wanted to play,” said Dürr. So they built their own robot from the ground up. The lightweight, six-jointed arm can whip a racket at over 20 meters (roughly 66 feet) per second and react roughly 11 times faster than a person.

All assembled, Ace is a table-tennis powerhouse—but not unbeatable. Against five elite and two professional players, it dominated the less-experienced elites but fell to the pros. In the months since the team wrote up their results, the robot continued improving against top-tier competition.

Ace didn’t win by simply being faster than humans. Rather, it won by being inventive. It created different kinds of spins, varied its returns, and consistently landed the ball on target. When Olympic table-tennis player, Kinjiro Nakamura, watched Ace play, he was mesmerized by the robot’s unconventional moves. “No one else would have been able to do that. I didn’t think it was possible,” he said. But if a robot can pull it off, maybe humans can too.

For Colombini, who worked on soccer-playing robots, that kind of agility and improvisation is the real goal. Robots need to think on their feet and easily navigate the physical world to work safely with people. “I need the skills and the abilities of these robots, learned in these environments that are easy for us to see how they are evolving,” she said. “So, sports are just a proxy for what we want.”

The post Sony’s Table-Tennis Robot Beat Elite Human Players With Unorthodox Moves appeared first on SingularityHub.

Algorithmic advances are steadily lowering the bar for quantum attacks—even before large-scale hardware exists.

Online data is generally pretty secure. Assuming everyone is careful with passwords and other protections, you can think of it as being locked in a vault so strong that even all the world’s supercomputers, working together for 10,000 years, could not crack it.

But last month, Google and others released results suggesting a new kind of computer—a quantum computer—might be able to open the vault with significantly fewer resources than previously thought.

The changes are coming on two fronts. On one, tech giants such as IBM and Google are racing to build ever-larger quantum computers: IBM hopes to achieve a genuine advantage over classical computers in some special cases this year, and an even more powerful “fault-tolerant” system by 2029.

On the other front, theorists are refining quantum algorithms: Recent work shows the resources needed to break today’s cryptography may be far fewer than earlier estimates.

The net result? The day quantum computers can break widely used cryptography—portentously dubbed “Q-Day”—may be approaching faster than expected.

The Quantum Hardware Race

Quantum computers are built from quantum bits, or qubits, which use the counterintuitive properties of very tiny objects to carry out computations in a different and sometimes far more efficient way from traditional computers.

So far the technology is in its infancy, with the major goal to increase the number of qubits that can be connected to work as a single computer. Bigger quantum computers should be much better at some things than their traditional counterparts—they will have a “quantum advantage.”

Late last year, IBM unveiled a 120-qubit chip which it hopes will demonstrate a quantum advantage for some tasks.

Google also recently announced it planned to speed up its move to adopt encryption techniques that should be safe against quantum computers, known as post-quantum cryptography.

Alongside these tech giants, newer approaches are also flourishing. PsiQuantum is using light-based qubits and traditional chip-manufacturing technology. Experimental platforms such as neutral-atom systems have demonstrated control over thousands of qubits in laboratory settings.

In response, standards bodies and national agencies are setting increasingly concrete timelines for moving away from common encryption systems that are vulnerable to quantum attack.

In the United States, the National Institute of Standards and Technology (NIST) has proposed a transition away from quantum-vulnerable cryptography, with migration largely completed by 2035. In Australia, the Australian Signals Directorate has issued similar guidance, urging organizations to begin planning immediately and transition to post-quantum cryptography by 2030.

Algorithms Make the Lock-Picking Faster

Hardware is only half the story. Equally important are advances in quantum algorithms—ways to use quantum computers to attack encryption.

Much interest in quantum computer development was spurred by Peter Shor’s 1994 discovery of an algorithm that showed how quantum computers could efficiently find the prime factors of very large numbers. This mathematical trick is precisely what you need to break the common RSA encryption method.

For decades, it was believed a quantum computer would need millions of physical qubits to pose a threat to real-world encryption. This is far bigger than current systems, so the threat felt comfortably distant.

That picture is now changing.

In March 2026, Google’s Quantum AI team released a detailed study showing that far fewer resources may be needed to attack a different kind of encryption which uses mathematical objects called elliptic curves. This is what systems including Bitcoin and Ethereum use—and the study shows how a quantum computer with fewer than half a million physical qubits may be able to crack it in minutes.

That’s still a long way beyond current quantum computers, but around ten times less than earlier estimates.

At the same time, a March 2026 preprint from a Caltech—Berkeley—Oratomic collaboration explores what might be possible using neutral-atom quantum computers. The researchers estimate that Shor’s algorithm could be implemented with as few as 10,000–20,000 atomic qubits. In one design they propose, a system with around 26,000 qubits could crack Bitcoin’s encryption in a few days, while tougher problems like the RSA method with a 2048-bit key would need more time and resources.

In plain terms: The codebreakers are becoming more efficient. Advances in algorithms and design are steadily lowering the bar for quantum attacks, even before large-scale hardware exists.

What Now?

So what does this mean in practice?

First, there is no immediate catastrophe—today’s cryptography won’t be broken overnight. But the direction of travel is clear. Each improvement in hardware or algorithms reduces the gap between current capabilities and useful quantum cracking machines.

Second, viable defenses already exist. NIST has standardized several post-quantum cryptographic algorithms which are believed to be resistant to quantum attacks.

Technology companies have begun deploying these in hybrid modes: Google Chrome and Cloudflare, for example, already support post-quantum protections in some protocols and services.

Systems that rely heavily on elliptic-curve cryptography—including cryptocurrencies and many secure communication protocols—will need particular attention. Google’s recent work explicitly highlights the need to migrate blockchain systems to post-quantum schemes.

Finally, this is a two-front race. It is not enough to track progress in quantum hardware alone. Advances in algorithms and error correction can be just as important, and recent results show these improvements can significantly reduce the estimated cost of attacks.

Every new headline about reduced qubit counts or faster quantum algorithms should be understood for what it is: another step toward a future where today’s cryptographic assumptions no longer hold.

The only reliable defense is to move—deliberately but decisively—toward quantum-safe cryptography.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

The post Quantum Computers Are Coming to Break Cryptography Faster Than Anyone Expected appeared first on SingularityHub.

Future

The People Do Not Yearn for AutomationNilay Patel | The Verge

“Not everything about our lives can be measured and automated and optimized, and it shouldn’t be. And so the tech industry is rushing forward to put AI everywhere at enormous cost—energy, emissions, manufacturing capacity, the ability to buy RAM—and locked into the narrow framework of software brain without realizing they are also asking people to be fundamentally less human.”

The post This Week’s Awesome Tech Stories From Around the Web (Through April 25) appeared first on SingularityHub.

A year after most robots failed to finish the Beijing race, nearly half the field autonomously ran a course of slopes, narrow passages, and 20 turns.

Humanoid robots are Silicon Valley’s latest obsession, but real-world performance has lagged the hype. That may be starting to change, however, after a robot beat the human record for a half marathon by nearly seven minutes in Beijing.

While tech companies around the world are piling into humanoid robots, China has made it a national priority. The government is pouring subsidies and infrastructure investment into the sector, and Chinese firms already account for around 80 percent of the humanoid machines shipped globally, according to the South China Morning Post.

Eager to show off its prowess, China has been staging sporting events for robots, most notably last year’s inaugural World Humanoid Robot Games. Another such event, the Beijing E-Town Half Marathon, pits humanoid robots against thousands of human runners over a 13-mile course. Last year, most of the non-human competitors failed to finish, and the fastest robots managed an unimpressive two hours and 40 minutes.

But this time around, four robots clocked times under an hour. And the winner, made by Chinese smartphone company Honor, registered a record-breaking 50 minutes, 26 seconds, eclipsing the benchmark set by Ugandan long-distance runner Jacob Kiplimo in Lisbon last month.

“Running faster may not seem meaningful at first, ​but it enables technology transfer, for example, into structural reliability and cooling, and eventually industrial applications,”  Du Xiaodi, an engineer on the winning team, told Reuters.

More than 100 teams fielded 300 robots at this year’s event, up from just 21 entries at the inaugural event last year. But Honor, a spinoff from Chinese telecom giant Huawei, dominated the competition, with separate teams from the company taking all three podium spots.

The winning robot, Lightning, navigated the course entirely autonomously. The bot stands 5 feet 6 inches tall but features legs 37 inches long to mimic the physical attributes of elite runners. It also boasts liquid cooling technology used in the company’s smartphones.

The growing sophistication of the robots’ control software is perhaps one of the starkest shifts since last year, with roughly 40 percent of teams operating autonomously. This is particularly impressive given the challenging course, according to Bernstein Research analysts.

“The course included flat sections, slopes, narrow passages, and ~ 20 turns, demonstrating rapid improvement in robots’ intelligence to handle generalized environments in the real world,” they wrote, according to Bloomberg.

But the technology isn’t bulletproof yet. One robot ran into a barricade and had to be carried off on a stretcher. Another veered into a bush after crossing the finish line. And one continued racing with its torso held together by packing tape after a heavy fall.

Nonetheless, the race showcased the rapid progress China’s tech industry is making, particularly in the raw components used to build these machines, like motors, joints, and batteries. Liu Xiangquan, a robotics professor at Beijing Information Science and Technology University told The South China Morning Post that long-distance running is a great test of how well these components can stand up to the kind of repeated strain that will occur in industrial settings.

And that’s likely to cause some consternation in US policy circles, where many see robotics as a key battlefront in the growing technological rivalry between the two superpowers.

Behind Sunday’s spectacle is a higher-stakes contest between China and the US over who will dominate the next generation of humanoids. US robotics firms have been lobbying Washington to draft a national strategy to counter China, which could include tariffs or bans on Chinese robots to help protect domestic producers.

However, running fast in a straight line is a very different challenge than the fine motor control and perception demanded by commercial applications. Experts told Reuters that despite impressive hardware, robotics companies are still a long way from developing the sophisticated software required to put these humanoids to practical use.

Still, these machines struggled to get over the starting line just a year ago. The gap between humanoid robots and human athletes has closed faster than anyone expected, so betting against further rapid progress seems unwise.

The post A Humanoid Robot Beat the Human World Record for a Half Marathon appeared first on SingularityHub.

That’s a few minutes longer than it takes to fill up the average gas-powered car—but still fast enough it might not matter.

For all their promise, electric cars have always had a big drawback: Charging takes much longer than filling up a gas tank.

But the gap has been closing, and this week, Chinese battery giant CATL announced battery technology nearing parity. On Tuesday, the company said its third-generation Shenxing fast-charging battery goes from 10 percent to 98 percent charged in 6 minutes and 27 seconds.

If you’re driving an electric car around town, charging is a breeze. You probably don’t have to do it more than a couple times a month. And when you do, you can plug your car in overnight at home.

For longer trips, you’ll need a charging station. Smartphone apps can help, and drivers learn to plan ahead, but it’s still a pain. Stations aren’t abundant, and when you find one, there may be a line. A full charge will then take the better part of an hour. Most people aim for 80 percent, but even that consumes up to a half hour. EV fans may find it’s worth the trouble, but range is a sticking point for many drivers.

It’s no wonder that battery makers have been hyper-focused on energy density, which determines how far EVs can go, and charging speed. They’ve improved both in recent years. But increasing range, which involves balancing a complex mix of battery chemistries, weight, and economics, may prove a tougher tradeoff to manage than bringing charging times in line with gas-powered cars at the pump.

In other words, if you can travel the same distance and charge or gas up in roughly the same amount of time, the two become interchangeable on long trips. (This also depends, of course, on infrastructure—more on that below.)

CATL has been pushing the boundaries of charging speeds with its Shenxing line of fast-charging batteries, first announced in 2023. The company is the world’s largest EV battery manufacturer. Its products power EVs in China but also American brands including Tesla and Ford.

The numbers are hard to compare generation to generation and company to company, as the specs reported vary. The second-generation Shenxing battery, announced last year, charged from 5 percent to 80 percent in 15 minutes, according to the Financial Times. Then in March of this year, rival battery maker BYD said its Blade 2.0 model charged 10 percent to 97 percent in 9 minutes.

Notching nearly a full charge in under 10 minutes was already an impressive mark.

But on Tuesday, CATL one-upped BYD with its third-generation Shenxing, which takes a full charge in a little over six minutes. At a maximum legal rate of 10 gallons per minute at gas stations in the US, that’s still a few minutes longer than it takes to fill up most gas-powered cars. But it might also be fast enough not to matter. Big gas-powered trucks are already in the same range. And CATL said charging to 80 percent takes just 3 minutes and 44 seconds—which is nearly a wash.

“This effectively closes the gap with ICE [internal combustion engine] vehicles,” Bernstein analysts wrote in a note quoted by the Wall Street Journal.

Fast-charging batteries have shorter lifespans due to excess heat. But CATL said it’s tamed the heat by decreasing the amount produced in operation, more effectively bleeding it off, and controlling how and when it’s generated. The battery retains over 90 percent capacity after 1,000 charging cycles.

“The boundaries of electrochemistry are still far from being reached, and the possibilities of materials science are still far from being exhausted,” CATL founder and CEO, Robin Zeng, told reporters and investors, per the Financial Times.

With 6-minute charging times, it’s easy to imagine charging station lines evaporating. Instead of drivers grabbing a meal while their car takes up real estate, they’d breeze in and out, like at a gas station.

That vision will take time to materialize, however. There are still far fewer charging stations than there are gas pumps. And those that do exist won’t include chargers that handle the bleeding edge anytime soon.

As for the batteries themselves, splashy press releases don’t usually translate to near-term availability and might not match real-world performance. The third-generation Shenxing isn’t likely to hit roads right away. When it does, it could show up in Chinese models first, be pricey (like BYD’s latest offering), and require fancy new chargers.

Still, it’s no longer theoretical: EVs can compete with the convenience of traditional cars at the gas station.

The post CATL’s New EV Battery Charges in Six Minutes appeared first on SingularityHub.