Singularity HUBModify

2025-11-22 23:00:00

The post This Week’s Awesome Tech Stories From Around the Web (Through November 22) appeared first on SingularityHub.

2025-11-22 02:24:34

The global investment frenzy around AI has seen companies valued at trillions of dollars and eye-watering projections of how it will boost economic productivity.

But in recent weeks the mood has begun to shift. Investors and CEOs are now openly questioning whether the enormous costs of building and running AI systems can really be justified by future revenues.

Google’s CEO, Sundar Pichai, has spoken of “irrationality” in AI’s growth, while others have said some projects are proving to be more complex and expensive than expected.

Meanwhile, global stock markets have declined, with tech shares taking a particular hit, and the value of cryptocurrencies has dipped as investors appear increasingly nervous.

So how should we view the health of the AI sector?

Well, bubbles in technology are not new. There have been great rises and great falls in the dot-com world, and surges in popularity for certain tech platforms (during Covid for example) which have then flattened out.

Each of these technological shifts was real, but they became bubbles when excitement about their potential ran far ahead of companies’ ability to turn popularity into lasting profits.

The surge in AI enthusiasm has a similar feel to it. Today’s systems are genuinely impressive, and it’s easy to imagine them generating significant economic value. The bigger challenge comes with how much of that value companies can actually keep hold of.

Investors are assuming rapid and widespread AI adoption along with high-margin revenue. Yet the business models needed to deliver that outcome are still uncertain and often very expensive to operate.

This creates a familiar gap between what the technology could do in theory, and what firms can profitably deliver in practice. Previous booms show how quickly things wobble when those ideas don’t work out as planned.

AI may well reshape entire sectors, but if the dazzling potential doesn’t translate quickly into steady, profitable demand, the excitement can slip away surprisingly fast.

Fit to Burst?

Investment bubbles rarely deflate on their own. They are usually popped by outside forces, which often involve the US Federal Reserve (the US’s central bank) making moves to slow the economy by raising interest rates or limiting the supply of money, or a wider economic downturn suddenly draining confidence.

For much of the 20th century, these were the classic triggers that ended long stretches of rising markets.

But financial markets today are larger, more complex, and less tightly tied to any single lever such as interest rates. The current AI boom has unfolded despite the US keeping rates at their highest level in decades, suggesting that external pressures alone may not be enough to halt it.

Instead, this cycle is more likely to end from within. A disappointment at one of the big AI players—such as weaker than expected earnings at Nvidia or Intel—could puncture the sense that growth is guaranteed.

Alternatively, a mismatch between chip supply and demand could lead to falling prices. Or investors’ expectations could quickly shift if progress in training ever larger models begins to slow, or if new AI models offer only modest improvements.

Overall then, perhaps the most plausible end to this bubble is not a traditional external shock, but a realization that the underlying economics are no longer keeping up with the hype, prompting a sharp revaluation across related stocks.

Artificial Maturity

If the bubble did burst, the most visible shift would be a sharp correction in the valuations of chipmakers and the large cloud companies driving the current boom.

These firms have been priced as if AI demand will rise almost without limit. So any sign that the market is smaller or slower than expected would hit financial markets hard.

This kind of correction wouldn’t mean AI disappears, but it would almost certainly push the industry into a more cautious, less speculative phase.

The deepest consequence would be on investment. Goldman Sachs estimates that global spending on AI-related infrastructure could reach $4 trillion by 2030. In 2025 alone, Microsoft, Amazon, Meta, and Google’s owner Alphabet have poured almost $350 billion into data centers, hardware, and model development. If confidence faltered, much of this planned expansion could be scaled back or delayed.

That would ripple through the wider economy, slowing construction, dampening demand for specialized equipment, and dragging on growth at a time when inflation remains high.

But a bursting AI bubble would not erase the technology’s long-term importance. Instead, it would force a shift away from the “build it now, profits will follow” mindset which is driving much of the current exuberance.

Companies would focus more on practical uses that genuinely save money or raise productivity, rather than speculative bets on transformative breakthroughs. The sector would mature. But it would probably do so only after a painful period of adjustment for investors, suppliers and governments who have tied their growth expectations to an uninterrupted AI boom.

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

The post Is the AI Bubble About to Burst? What to Watch for as the Markets Wobble appeared first on SingularityHub.

2025-11-21 06:37:16

A host of startups are racing to achieve commercial fusion power. Zap Energy just drew a line in the sand after announcing its latest device recorded the highest-ever plasma pressures for its particular class of reactors.

Getting atoms to fuse typically requires you to subject an ionized gas, known as a plasma, to extreme heat and pressure. This normally involves massive rings of powerful magnets or enormous laser arrays. But Zap Energy is pursuing a novel approach known as a sheared-flow-stabilized Z-pinch configuration, which uses electrical currents to compress and heat the fuel.

The company says this should make its devices much smaller and cheaper than competitors, but the technology is considerably less mature than other leading fusion reactor designs. Now, though, the company has achieved plasma pressures of 1.6 gigapascals—roughly 10,000 times atmospheric pressure at sea level—in a machine only 12 feet long, a record for a sheared-flow Z-pinch system and a major step toward commercialization, according to the company.

“This was a major effort by the team that was successful because of a tightly coupled cycle of theoretical predictions, computational modeling, rapid build and test engineering, experimental validation, and measurement expertise,” Ben Levitt, vice president of research and development at Zap, said in a press release.

“With a smaller system we have the benefit of being able to move quickly, and achieving these results in systems that are a fraction of the size and cost of fusion devices of comparable performance is a big part of what makes this such a significant accomplishment.”

The idea behind Zap’s reactor design is surprisingly simple. Like most fusion systems, it uses special hydrogen isotopes as fuel. These are contained as gas in a thin tube at the reactor’s heart. The machine fires a massive electrical current through this gas, superheating and turning it into a plasma.

The electrical current also creates a powerful magnetic field that squeezes the plasma—a phenomenon known as a Z-pinch—and generates extremely high pressures in a small area. In theory, with careful design and high enough currents, this should generate the conditions for fusion.

The process is actually a little more complicated in reality. Zap’s reactor first uses electrical current to accelerate the plasma along the length of the tube, which helps stabilize it. When the plasma reaches the cone-shaped end of the tube, the magnetic field squeezes it into a Z-pinch.

Zap’s recent record-breaking pressure was thanks to a new design separating the processes of accelerating and compressing the plasma. Earlier devices used two electrodes to deliver current to the reactor. These achieved good levels of heating but didn’t allow the team to hit the high pressures they were targeting.

The new FuZE-3 system incorporates a third electrode, which makes it possible to deliver two power pulses rather than just one, spokesperson Andy Freeborn told TechCrunch. The company says this new setup allows them to independently control plasma acceleration and compression—key to achieving the latest record-breaking results.

Generating useful power from fusion reactions requires a careful balance between plasma density, temperature, and confinement time. Zap says their approach represents a middle-ground, aiming for relatively high pressures and reasonably long confinement times.

However, TechCrunch notes that Zap believes it will have to boost plasma pressures at least tenfold before it hits scientific breakeven—the point where energy created by the reaction outweighs the amount required to kickstart it.

Scientific breakeven doesn’t account for the energy use of supporting infrastructure or the ability to extract energy from the reaction, so even this is only a stepping stone towards commercial viability.

Nonetheless, Zap is powering ahead with work on a next-generation device, due to come online this winter even as FuZE-3 tests are ongoing. Given the huge uncertainties around the feasibility and timelines of different fusion approaches, the more people driving progress in this field the better.

The post Startup Zap Energy Just Set a Fusion Power Record With Its Latest Reactor appeared first on SingularityHub.

2025-11-19 07:48:32

Before the age of one, KJ Muldoon had already made medical history. He was the first person to receive a gene editing therapy specifically designed for him. KJ was born with a deadly gene mutation. His body couldn’t remove ammonia, a byproduct of eating protein. The illness eventually leads to serious brain injury. Roughly half of infants with the disease don’t survive, and those who do suffer severe debilitation and often require liver transplants.

The disease stems from a single mutated DNA letter that prevents the body from making a working enzyme. The clock ticking, teams of scientists developed a gene editor to replace the mutated letter with a normal version. Just weeks after three infusions, KJ was tolerating more protein in his diet and meeting developmental milestones.

Gene editors usually require years to test and perfect. KJ’s treatment took only six months.

Now, his doctors are looking to bring the “transformative” technology to others with rare inherited diseases. In an ambitious clinical trial, they will use base editing—an offshoot of CRISPR gene editing—to correct DNA mutations in rare metabolic diseases. After months of negotiation with the FDA, they have streamlined the convoluted and time-consuming process of gene therapy approval, saving precious time that many young patients don’t have.

A trial could start as early as 2026. At least five kids will receive customized gene editors to test each treatment’s safety and efficacy.

More than 30 million people in the US suffer from rare genetic diseases. Most are so unique that drug companies aren’t willing to invest years to develop gene therapies that only benefit a few, leaving these patients in limbo.

If successful, the trial could launch “a future of ‘interventional genetics’ in which such therapies are the standard of care,” wrote Drs. Rebecca Ahrens-Nicklas and Kiran Musunuru at the Children’s Hospital of Philadelphia in a recently published roadmap of the approach.

A Single Miracle

KJ’s mutation was in the CPS1 gene. A single swapped DNA letter shuts down the liver’s ability to make an enzyme that rids the body of ammonia. Symptoms include vomiting, lethargy, and brain damage. The condition is called urea cycle disorder, or UCD.

Scientists have long known about UCD. While there is a drug to manage symptoms, patients must adhere to a very low protein diet, which limits a baby’s normal development. Viral infections, common in young infants, can also spike ammonia to dangerous levels.

Before treatment, KJ was sequestered in a hospital room, unable to go home and meet his siblings. His symptoms were so severe that at one point his physician discussed palliative care with his heartbroken parents.

Mutations in seven known genes can cause UCD, making a one-size-fits-all gene therapy impossible. But doctors already knew KJ’s mutation—a single letter swap—making him a perfect candidate for base editing.

A version of CRISPR gene editing, base editing is especially good at swapping single DNA letters. Flipping one DNA letter out of the roughly three billion in the human genome seems inconsequential, but the change often alters the final form and function of a protein. In KJ’s case, it saved his life.

Base editing is already in clinical trials for people genetically prone to dangerously high cholesterol levels, with promising initial results. One trial is being led by Verve Therapeutics, which Musunuru co-founded. Other studies are using the tool to correct genetic faults in stem cells that lead to sickle cell disease.

These attempts all target a known mutation in a disease-causing gene shared by people with the same illness. KJ’s genetic typo was unique to him. Any life-saving base editor had to be made from scratch.

Over the next six months, a remarkable collaboration between doctors, academics, and biotech companies crafted KJ’s treatment. Base editors require two components: A guide RNA “bloodhound” that scans the genome for the defect and a protein that swaps out the faulty DNA letter. The team wrapped instructions for both inside tiny bubbles of fat, which once injected, made their way to the liver, the target organ for the therapy.

Within weeks KJ started feeling better. By roughly 10 months of age, he was discharged from the hospital and is now learning to take his firsts steps at home.

The treatment was tailored to KJ, but base editing is plug-and-play. Guide RNA can easily be reprogrammed to hunt down other single-letter DNA mutations that lead to rare diseases. At least in theory. The cost of development can be prohibitive, partly because of the time it takes to test each individual treatment. Regulatory hurdles further draw out the process.

One for All

KJ’s doctors are now pushing for an even faster timeline to treat kids with his condition.

In their proposed trial, five kids with genetic mutations across seven genes will receive a custom treatment similar to KJ’s. The only difference between the treatments will be the guide RNA, which will be tailored to each child’s particular mutation. Doctors will then follow the children’s health for 15 years.

The FDA usually requires safety data for each new gene therapy formulation. Here, however, they agreed on a single safety trial that covers all formulations based on the same principle. KJ’s safety data will also be taken into consideration. This “regulatory innovation” could massively accelerate development time, wrote the team.

KJ’s success story has brought others on board. In July, the Center for Pediatric CRISPR Cures launched at the University of California, Berkeley to pursue technologies for life-saving custom gene therapies in children.

Meanwhile, the Advanced Research Projects Agency for Health, a US government agency, launched two new programs in mid-September to make custom gene therapies for people with rare genetic disorders a reality.

One of these, called THRIVE, is focused on building a platform to rapidly develop personalized gene editing tools. Another, GIVE, aims to bring high-quality cell and gene therapy manufacturing technologies to local clinics, slashing transportation costs. Both initiatives are now welcoming proposals.

“Our vision is to rapidly produce multiple kinds of genetic medicines so that breakthrough treatments are accessible, affordable, and ready to dose within a week of diagnosis,” GIVE program manager Dr. John Schiel said a press release.

Ahrens-Nicklas and Musunuru are confident personalized gene therapy can play a role in future healthcare. “With full-throated support from funding bodies…and from regulatory agencies such as the FDA, we are optimistic that in the coming years, our team and other teams will be able to take tangible steps toward making interventional genetics the standard of care for many diseases,” they wrote.

The post Scientists Race to Deliver Custom Gene Therapies for Incurable Diseases in Weeks—Not Years appeared first on SingularityHub.

2025-11-18 08:04:13

As its whiskers flitter, the mouse’s brain sparks with activity. A tiny implant records the electrical chatter and beams it to a nearby computer.

Smaller than a grain of salt, the implant is powered by and transmits data with light. Unlike most implants, it moves with the brain to reduce scarring. Dubbed MOTE, the device reliably captured electrical signals for a year in mice—about half their lifespan—without obvious damage.

“The long-term recording of neural activity could be used to understand complex behaviors and disorders,” Sunwoo Lee at Nanyang Technological University, Alyosha Molnar at Cornell University, and team wrote in a recent paper describing the implant.

Penny for Your Thoughts

Brain implants are helping decode—and restore—the neural signals behind our thoughts, memories, movements, and behavior.

Most devices rely on arrays of microelectrodes inserted into the brain, though some sit on the brain’s surface to minimize damage. From translating neural activity into computerized speech to restoring movement in people with paralysis, these devices have already transformed lives.

But there’s a major drawback. Most implants use wires plugged into a port embedded in a person’s skull to transfer signals, requiring extensive surgery. Implanted electrodes, although small, are like fixed pins inside a wobbly Jell-o block. They can’t move with brain tissue. Over time, scarring reduces the implants’ efficiency, and the hardware triggers inflammation.

Scientists have been tackling these roadblocks with clever ideas like wireless implants that transmit data on radio frequencies, a bit like walkie-talkies. “Neurograin” implants, for example, record and stimulate the brain wirelessly and transmit data to a thin electrical patch on the scalp. Other devices use ultrasound for power and to send signals to a controller.

But most wireless implants are still bulky, “equivalent to a sizable fraction of the mouse brain,” wrote the team.

Then there’s the ultimate enemy: Time. The brain is bathed in fluid for nourishment and waste removal, but this soupy concoction eats away at electrical components. Although some methods can capture neural activity over many months with a microscope and implanted light probes, they only work in genetically engineered mice with glow-in-the-dark neurons.

A durable wireless implant for living brains has so far escaped scientists’ grasp.

“Our goal was to make the device small enough to minimize that disruption while still capturing brain activity faster than imaging systems, and without the need to genetically modify the neurons for imaging,” Molnar said in a press release.

Power of Light

The new MOTE device, smaller than a grain of salt, combines electronics and LEDs for wireless recording and communication.

Red and infrared light penetrate the scalp, skull, and brain with minimal distortion, making them useful energy sources. The device has a diode that turns those wavelengths of light into electrical energy—a bit like those inside solar panels—to power the device. Once the implant captures electrical signals from the brain, it sends them to a computer on short pulses of light.

Like morse code, the exact timing and duration of the pulses reflect neural activity. This technology is widely used in satellite communication, wrote the team, and requires very little power to operate.

MOTE’s onboard electronics are like computer chips. Each packs 186 transistors, which form the basis of three main circuits. One circuit boosts recorded brain signals, another recodes them into light pulses, and the third drives LEDs for transmission to a computer.

These components are protected by a custom sheath made by coating the implant one atomic layer at a time. The ultra-thin sheath protects MOTE from the brain’s corrosive environment. Each fabrication step can be done in parallel, making nearly 100 devices at the same time.

“As far as we know, this is the smallest neural implant that will measure electrical activity in the brain and then report it out wirelessly,” said Molnar.

Live Long and Prosper

In a first test, the devices reliably captured electrical activity from heart muscle cells in petri dishes, suggesting they worked as intended.

The team next implanted the device into a unique part of the mouse brain. Mice heavily depend on their whiskers to navigate the world. These signals are processed in the barrel cortex. A range of electrical patterns capture sensations and generate twitches in each whisker.

Some mice received the implant on top of their brains, instead of penetrating into the delicate tissue. But most had the device implanted using a nanoinjector. Over the next year, the device faithfully transmitted data from the barrel cortex when scientists tickled the mice’s whiskers. It detected activity from single neurons and neural network activity associated with behavior.

MOTE seemed mostly harmless. None of the mice experienced seizures or other neurological issues sometimes seen in larger implants. They skittered around and chowed down on food as usual. There was also very little scarring around the implant, even after a year.

The devices aren’t just for decoding mouse brains. They could one day pick up electrical signals from organoids—so-called mini-brains. Organoids loosely mimic the early stages of brain development. Although tiny, they’re densely packed with multiple types of brain cells and connections, making it difficult for bulkier implants to record signals without damage.

Upgraded with better detection and light-emission hardware, MOTEs could theoretically work up to six millimeters deep in the brain, enough to record from the entire mouse brain and in organoids, wrote the team.

They’re still far from clinical use, but making implants wireless means they’re more compatible with brain imaging technologies, such as fMRI (functional magnetic resonance imaging), which could paint a wider picture of brain activity during tasks. Outside the brain, MOTE could tap into the spinal cord, heart, or other tissues and record dynamic movies of their health.

The post This Wireless Brain Implant Is Smaller Than a Grain of Salt appeared first on SingularityHub.

2025-11-15 17:39:58

The post This Week’s Awesome Tech Stories From Around the Web (Through November 15) appeared first on SingularityHub.