healthcarereimagined

Fuzziness may rule the quantum realm, but it can be manipulated to our advantage.

By Sophia Chen

February 29, 2024

When two black holes spiral inward and collide, they shake the very fabric of space, producing ripples in space-time that can travel for hundreds of millions of light-years. Since 2015, scientists have been observing these so-called gravitational waves to help them study fundamental questions about the cosmos, including the origin of heavy elements such as gold and the rate at which the universe is expanding. 

But detecting gravitational waves isn’t easy. By the time they reach Earth and the twin detectors of the Laser Interferometer Gravitational-Wave Observatory (LIGO), in Louisiana and Washington state, the ripples have dissipated into near silence. LIGO’s detectors must sense motions on the scale of one ten-thousandth the width of a proton to stand a chance. 

LIGO has confirmed 90 gravitational wave detections so far, but physicists want to detect more, which will require making the experiment even more sensitive. And that is a challenge. 

“The struggle of these detectors is that every time you try to improve them, you actually can make things worse, because they are so sensitive,” says Lisa Barsotti, a physicist at the Massachusetts Institute of Technology.

Nevertheless, Barsotti and her colleagues recently pushed past this challenge, creating a device that will allow LIGO’s detectors to detect far more black hole mergers and neutron star collisions. The device belongs to a growing class of instruments that use quantum squeezing—a practical way for researchers dealing with systems that operate by the fuzzy rules of quantum mechanics to manipulate those phenomena to their advantage. 

Physicists describe objects in the quantum realm in terms of probabilities—for example, an electron is not located here or there but has some likelihood of being in each place, locking into one only when its properties are measured. Quantum squeezing can manipulate the probabilities, and researchers are increasingly using it to exert more control over the act of measurement, dramatically improving the precision of quantum sensors like the LIGO experiment.  

“In precision sensing applications where you want to detect super-small signals, quantum squeezing can be a pretty big win,” says Mark Kasevich, a physicist at Stanford University who applies quantum squeezing to make more precise magnetometers, gyroscopes, and clocks with potential applications for navigation. Creators of commercial and military technology have begun dabbling in the technique as well: the Canadian startup Xanadu uses it in its quantum computers, and last fall, DARPA announced Inspired, a program for developing quantum squeezing technology on a chip. Let’s take a look at two applications where quantum squeezing is already being used to push the limits of quantum systems.

Taking control of uncertainty

The key concept behind quantum squeezing is the phenomenon known as Heisenberg’s uncertainty principle. In a quantum-mechanical system, this principle puts a fundamental limit on how precisely you can measure an object’s properties. No matter how good your measurement devices are, they will suffer a fundamental level of imprecision that is part of nature itself. In practice, that means there’s a trade-off. If you want to track a particle’s speed precisely, for example, then you must sacrifice precision in knowing its location, and vice versa. “Physics imposes limits on experiments, and especially on precision measurement,” says John Robinson, a physicist at the quantum computing startup QuEra. 

By “squeezing” uncertainty into properties they aren’t measuring, however, physicists can gain precision in the property they want to measure. Theorists proposed using squeezing in measurement as early as the 1980s. Since then, experimental physicists have been developing the ideas; over the last decade and a half, the results have matured from sprawling tabletop prototypes to practical devices. Now the big question is what applications will benefit. “We’re just understanding what the technology might be,” says Kasevich. “Then hopefully our imagination will grow to help us find what it’s really going to be good for.” 

LIGO is blazing a trail to answer that question, by enhancing the detectors’ ability to measure extremely tiny distances. The observatory registers gravitational waves with L-shaped machines capable of sensing tiny motions along their four-kilometer-long arms. At each machine, researchers split a laser beam in two, sending a beam down each arm to reflect off a set of mirrors. In the absence of a gravitational wave, the crests and troughs of the constituent light waves should completely cancel each other out when the beams are recombined. But when a gravitational wave passes through, it will alternately stretch and compress the arms so that the split light waves are slightly out of phase.

The resulting signals are subtle, though—so subtle that they risk being drowned out by the quantum vacuum, the irremovable background noise of the universe, caused by particles flitting in and out of existence. The quantum vacuum introduces a background flicker of light that enters LIGO’s arms, and this light pushes the mirrors, shifting them on the same scale as the gravitational waves LIGO aims to detect.

Barsotti’s team can’t get rid of this background flicker, but quantum squeezing allows them to exert limited control over it. To do so, the team installed a 300-meter-long cavity in each of LIGO’s two L-shaped detectors. Using lasers, they can create an engineered quantum vacuum, in which they can manipulate conditions to increase their level of control over either how bright the flicker can be or how randomly it occurs in time. Detecting higher-frequency gravitational waves is harder when the rhythm of the flickering is more random, while lower-frequency gravitational waves get drowned out when the background light is brighter. In their engineered vacuum, noisy particles still show up in their measurements, but in ways that don’t do as much to disturb the detection of gravitational waves.“ You can [modify] the vacuum by manipulating it in a way that is useful to you,” she explains. 

The innovation was decades in the making: through the 2010s, LIGO incorporated incrementally more sophisticated forms of quantum squeezing based on theoretical ideas developed in the 1980s. With these latest squeezing innovations, installed last year, the collaboration expects to detect gravitational waves up to 65% more frequently than before.

Quantum squeezing has also improved precision in timekeeping. Working at the University of Colorado Boulder with physicist Jun Ye, a pioneer in atomic clock technology, Robinson and his team made a clock that will lose or gain at most a second in 14 billion years. These super-precise clocks tick slightly differently in different gravitational fields, which could make them useful for sensing how Earth’s mass redistributes itself as a result of seismic or volcanic activity. They could also potentially be used to detect certain proposed forms of dark matter, the hypothesized substance that physicists think permeates the universe, pulling on objects with its gravity. 

The clock Robinson’s team developed, a type called an optical atomic clock, uses 10,000 strontium atoms. Like all atoms, strontium emits light at specific signature frequencies as electrons around the atom’s nucleus jump between different energy levels. A fixed number of crests and troughs in one of these light waves corresponds to a second in their clock. “You’re saying the atom is perfect,” says Robinson. “The atom is my reference.” The “ticking” of this light is far steadier than the vibrating quartz crystal in a wristwatch, for example, which expands and contracts at different temperatures to tick at different rates.

In practice, the tick in the Robinson team’s clock comes not from the light the electrons emit but from how the whole system evolves over time. The researchers first put each strontium atom in a “superposition” of  two states: one in which the atom’s electrons are all at their lowest energy levels and another in which one of the electrons is in an excited state. This means each atom has some probability of being in either state but is not definitively in either one—similar to how a coin flipping in the air has some probability of being either heads or tails, but is neither.

Then they measure how many atoms are in each state. The act of measurement puts the atoms definitively in one state or the other, equivalent to letting the flipping coin land on a surface. Before they measure the atoms, even if they intend to wind up with a 50-50 mixture, they cannot precisely dictate how many atoms will end up in each state. That’s because in addition to the system’s change over time, there is also inherent uncertainty in the state of the individual atoms. Robinson’s team uses quantum squeezing to more reliably determine their final states by reducing these intrinsic fluctuations. Specifically, they manipulate the uncertainties in the direction of each atom’s spin, a property of many quantum particles that has no classical counterpart. Squeezing improved the clock’s precision by a factor of 1.5.

To be sure, gravitational waves and ultra-precise clocks are niche academic applications. But there is interest in adapting the approach to other, potentially more mainstream uses, including quantum computers, navigation, and microscopy.

The increased use of quantum squeezing is part of a wider technological trend toward higher precision—one that encompasses cramming more transistors on chips, studying the universe’s most elusive particles, and clocking the fleeting time it takes for an electron to leave a molecule. Squeezing benefits only measurements so subtle that the randomness of quantum mechanics contributes significant noise. But it turns out that physicists have more control than they think. They may not be able to remove the randomness, but they can engineer where it shows up.

Article link: https://www-technologyreview-com.cdn.ampproject.org/c/s/www.technologyreview.com/2024/02/29/1089092/how-scientists-are-using-quantum-squeezing-to-push-the-limits-of-their-sensors/amp/

Only a year ago, Pat Gelsinger intended for Intel to compete with Nvidia, but the new CEO, Lip-Bu Tan, is burying that ambition.
He does not aim at the Client Computing Group when discussing funding new engineering ideas and innovations.
The Day 1 principle is a start-up philosophy; the new AI strategy after Gaudi has been virtually written off.

And it makes sense. While Intel was playing CEO roulette, Nvidia increased its R&D budget dramatically, and the AI giant did not need to fund other hobby projects like Intel. To catch Nvidia, Intel needed a big bet, which Gaudi never was, and now it is too late.

In 2024, Intel was spending 45% of its operating budget on Intel Foundry, and what was left of the R&D budget was prioritised for the Client Computing Group.

This is insufficient to compete with Nvidia, so the Intel Periscope has turned towards the two smaller vessels involved in XPU designs.

If the New Intel and the 2025 Vision are important to you, read more here: https://lnkd.in/drXUPZXJ

Article link; https://www.linkedin.com/posts/clausaasholm_only-a-year-ago-pat-gelsinger-intended-for-activity-7314629995451064321-RGiv?

The Lip-Bu Tan 2025 Vision of The New Intel

CLAUS AASHOLM

APR 05, 2025

Intel’s new CEO, Lip-Bu Tan, brought many new words that can give us insight into Intel’s strategy. While words don’t easily move supertankers like Intel, words are important, and all strategy and transformation begin with words. 

Under Pat Gelsinger, there were several strategic pivots. While he was the master of creative external CapEx deals to fund Intel Foundry, the strategy was retrenched to a focus on the x86 franchise and the Client Computing group under Michelle Johnston Holthaus as she protected her CCG tribe.

The appointment of Lip-Bu Tan does not continue the current Intel Strategy, and it is vital to understand what the new CEO brings to the party. You can think what you want about Intel, but it still powers most of the world’s computing even though it is struggling to do so profitably. Intel is still a key factor in the development of the semiconductor industry.

In his first keynote, Lip-Bu Tan directly discussed not being happy with where Intel is. While this is not the same as signalling a strategic change, his statements about ruthlessly focusing on the truth are. This is an acceptance of the Intel culture of “glossing over” rather than facing the brutal truths. 

The most radical idea that Lip-Bu Tan introduced was that Intel needs to adopt a day 1 mentality. While it sounds like more words, the implications of a “Day 1 mentality” can mean dramatic change for the Intel organisation and culture, as Intel is currently the poster child of a “Day 2” company.

Amazon’s “Day One”: A Philosophy of Perpetual Innovation

At the heart of Amazon’s relentless drive lies its “Day One” philosophy, which emphasises maintaining a startup-like mentality, even within a massive corporation. Introduced by Jeff Bezos in his 1997 shareholder letter, “Day One” is about constant curiosity, nimbleness, and a willingness to experiment. It’s the antithesis of “Day Two,” where complacency sets in, and companies become slow, bureaucratic, and less customer-focused.

Bezos understood that customer obsession is the key to maintaining “Day One” vitality. As customers always seek better, faster, and more innovative solutions, companies must continuously adapt and evolve to meet their needs. As tech reporter Alex Kantrowitz explains in his book “Always Day One,” this philosophy encourages building for the future, not resting on past achievements. It means treating even a sprawling enterprise like a small, agile startup, fostering a culture of innovation and a relentless focus on the customer. In essence, “Day One” is about embracing change, challenging the status quo, and always acting with the urgency and passion of a company just starting.

Lip-Bu keeps flirting with “Day 1” as he calls for customer focus.

“My number one priority was spending time with customers, and they gave me honest feedback. We have a lot of work ahead of us.” 

While talking to customers and getting feedback is essential, it is not the same as becoming customer-centric. A company like Apple is customer-centric and has delivered the most successful consumer product ever, even though nobody ever asked for an iPhone.

Talking to customers isn’t enough; you’ll need to understand their future situation and latent needs. This is not Intel’s traditional starting point, as the company has its cultural roots deep in products and technology, and Lip-Bu Tan wants to challenge that.

When discussing his background, Lip-Bu Tan highlights times when he successfully changed the culture of the companies he was involved with. This signals that cultural change is high on the agenda, and he believes he is the right person to drive it.

This also reveals that Lip-Bu Tan believes that corporate culture can be designed within a reasonable timeframe, which sounds more straightforward than it is. 

Anybody who has worked in a company with several acquisitions knows that the acquired culture does not die. It lives on in the acquired people and influences the existing culture. 

One of the key features of a large corporation is what I call management-dampening. All top management decisions must pass through the hierarchical layers in a sense-making process. The simple CEO decisions have complex impacts on the organisation, and each level needs to adapt to them in a way that makes sense for that layer and their internal and external customers. 

The company’s suspension system ensures that radical changes don’t break the system and prevent change from happening. 

I have seen several examples of how the management-dampening system has saved (prevented) senior management’s panic-flip-flopping decisions. Unfortunately, it also dampens necessary changes and complicates cultural change. 

An overview of the “Day One” philosophy can be seen below, along with the key message.

As Lip-Bu Tan talks about the truth, it is necessary to highlight Intel’s current approach to these principles. While you might disagree with my assessment, it is hard to characterise Intel as anywhere near a “Day One” company.

Lip-Bu Tan recognises this as he states that he is unhappy with things and wants to correct past mistakes. He wants to create the “New Intel”, an engineering-first company, suggesting that Intel is currently a “something else-first” company. 

The truth

The foundation of everything I do starts with a data-driven assessment of the current situation. While I know it is easier to sell happy data, it is a principle of mine to stay clear of liking and disliking data. 

There can be many truths, and data does respond to torture. Still, it is always in the company’s best interest to begin the strategic formulation based on an honest assessment of the situation.

Happy data rarely creates happy endings.

Intel is undoubtedly in a difficult spot, and Lip-Bu Tan’s ambitions are at odds with reality. Scarcity creates tension between different objectives but is innovation’s greatest friend. 

As I pointed out in The Corporate Sandwich Model, Intel externally has a problem with the truth. It is all rah-rah and up and to the right. This might be different internally, but I doubt it.

It’s time to explore the tension between reality and the Intel 2025 Vision. 

The x86 Franchise

The most apparent change in strategy from Gelsinger to Holthaus was the introduction of the “x86 franchise”. That the Queen of the Client Computing Group praised the holiest of relics in CCG was unsurprising and looked like the usual internal divisional power game. Lip-Bu Tan adds colour to the x86 concept and positions it more as an asset in the Foundry strategy as part of the differentiation strategy: IP and localisation. The x86 is also positioned in the future AI strategy built on the smoulders of Gaudi that Holthaus effectively killed by calling it “difficult” to use

The Innovation perspective.

Intel is trying to reduce operating costs to achieve positive cash flow. However, it is becoming evident that Intel’s current AI data centre product strategy is non-existent. 

In the 2024 Intel Vision keynote, Pat Gelsigner called Gaudi the only alternative to Nvidia H100 in data centre learning applications. As Michelle Johnston Holthaus took the co-CEO chair, she demonstrated the rift between the client group and the other divisions and called Gaudi challenging. This was a virtual death sentence, and Lip-Bu Tan is not reviving the architecture. 

He clearly stated that Intel is unhappy with its current position and needs to develop a new approach. Intel needs higher performance, lower power and lower cost. In one year, The focus shifted from competing with Nvidia to competing with Broadcom and Marvell in XPU accelerators in inference rather than Nvidia in Learning and inference. 

This also correlates with his statements about becoming an outside-in company that starts with the customer’s problem rather than the hardware in the old inside-out approach. 

Nvidia’s AI strategy builds on a well-executed product strategy, while Broadcom and Marvell are working on customer problems. Customers are telling Li-Bu Tan that they would love an Intel AI Platform (cheaper, faster, lower power—surprise!), and he has started deep-dive dives into reimagining the products. 

This is the key area to win as AI accelerates tech adoption and is the only driver behind the upturn. Other focus areas, such as photonics, quantum, and robotics, are on the agenda, but the data centre AI needs to be revived.

The competitive situation in AI is hopeless for Intel.

Excluding Nvidia revenue makes it even more embarrassing for Intel. For now, it is evident that Intel is not aiming at Nvidia and learning but at the competitors in sight that are focusing on the accelerator designs.

Article link; https://clausaasholm.substack.com/p/from-a-something-else-first-to-an?

What the firm found challenges some basic assumptions about how this technology really works.

By Will Douglas Heaven

March 27, 2025

The AI firm Anthropic has developed a way to peer inside a large language model and watch what it does as it comes up with a response, revealing key new insights into how the technology works. The takeaway: LLMs are even stranger than we thought.

The Anthropic team was surprised by some of the counterintuitive workarounds that large language models appear to use to complete sentences, solve simple math problems, suppress hallucinations, and more, says Joshua Batson, a research scientist at the company.

It’s no secret that large language models work in mysterious ways. Few—if any—mass-market technologies have ever been so little understood. That makes figuring out what makes them tick one of the biggest open challenges in science.

But it’s not just about curiosity. Shedding some light on how these models work exposes their weaknesses, revealing why they make stuff up and why they can be tricked into going off the rails. It helps resolve deep disputes about exactly what these models can and can’t do. And it shows how trustworthy (or not) they really are.

Batson and his colleagues describe their new work in two reports published today. The first presents Anthropic’s use of a technique called circuit tracing, which lets researchers track the decision-making processes inside a large language model step by step. Anthropic used circuit tracing to watch its LLM Claude 3.5 Haiku carry out various tasks. The second (titled “On the Biology of a Large Language Model”) details what the team discovered when it looked at 10 tasks in particular.

“I think this is really cool work,” says Jack Merullo, who studies large language models at Brown University in Providence, Rhode Island, and was not involved in the research. “It’s a really nice step forward in terms of methods.”

Circuit tracing is not itself new. Last year Merullo and his colleagues analyzed a specific circuit in a version of OpenAI’s GPT-2, an older large language model that OpenAI released in 2019. But Anthropic has now analyzed a number of different circuits inside a far larger and far more complex model as it carries out multiple tasks. “Anthropic is very capable at applying scale to a problem,” says Merullo.

Eden Biran, who studies large language models at Tel Aviv University, agrees. “Finding circuits in a large state-of-the-art model such as Claude is a nontrivial engineering feat,” he says. “And it shows that circuits scale up and might be a good way forward for interpreting language models.”

Circuits chain together different parts—or components—of a model. Last year, Anthropic identified certain components inside Claudethat correspond to real-world concepts. Some were specific, such as “Michael Jordan” or “greenness”; others were more vague, such as “conflict between individuals.” One component appeared to represent the Golden Gate Bridge. Anthropic researchers found that if they turned up the dial on this component, Claude could be made to self-identify not as a large language model but as the physical bridge itself.

The latest work builds on that research and the work of others, including Google DeepMind, to reveal some of the connections between individual components. Chains of components are the pathways between the words put into Claude and the words that come out.  

“It’s tip-of-the-iceberg stuff. Maybe we’re looking at a few percent of what’s going on,” says Batson. “But that’s already enough to see incredible structure.”

Growing LLMs

Researchers at Anthropic and elsewhere are studying large language models as if they were natural phenomena rather than human-built software. That’s because the models are trained, not programmed.

“They almost grow organically,” says Batson. “They start out totally random. Then you train them on all this data and they go from producing gibberish to being able to speak different languages and write software and fold proteins. There are insane things that these models learn to do, but we don’t know how that happened because we didn’t go in there and set the knobs.”

Sure, it’s all math. But it’s not math that we can follow. “Open up a large language model and all you will see is billions of numbers—the parameters,” says Batson. “It’s not illuminating.”

Anthropic says it was inspired by brain-scan techniques used in neuroscience to build what the firm describes as a kind of microscope that can be pointed at different parts of a model while it runs. The technique highlights components that are active at different times. Researchers can then zoom in on different components and record when they are and are not active.

Take the component that corresponds to the Golden Gate Bridge. It turns on when Claude is shown text that names or describes the bridge or even text related to the bridge, such as “San Francisco” or “Alcatraz.” It’s off otherwise.

Yet another component might correspond to the idea of “smallness”: “We look through tens of millions of texts and see it’s on for the word ‘small,’ it’s on for the word ‘tiny,’ it’s on for the French word ‘petit,’ it’s on for words related to smallness, things that are itty-bitty, like thimbles—you know, just small stuff,” says Batson.

Having identified individual components, Anthropic then follows the trail inside the model as different components get chained together. The researchers start at the end, with the component or components that led to the final response Claude gives to a query. Batson and his team then trace that chain backwards.

Odd behavior

So: What did they find? Anthropic looked at 10 different behaviors in Claude. One involved the use of different languages. Does Claude have a part that speaks French and another part that speaks Chinese, and so on?

The team found that Claude used components independent of any language to answer a question or solve a problem and then picked a specific language when it replied. Ask it “What is the opposite of small?” in English, French, and Chinese and Claude will first use the language-neutral components related to “smallness” and “opposites” to come up with an answer. Only then will it pick a specific language in which to reply. This suggests that large language models can learn things in one language and apply them in other languages.

Anthropic also looked at how Claude solved simple math problems. The team found that the model seems to have developed its own internal strategies that are unlike those it will have seen in its training data. Ask Claude to add 36 and 59 and the model will go through a series of odd steps, including first adding a selection of approximate values (add 40ish and 60ish, add 57ish and 36ish). Towards the end of its process, it comes up with the value 92ish. Meanwhile, another sequence of steps focuses on the last digits, 6 and 9, and determines that the answer must end in a 5. Putting that together with 92ish gives the correct answer of 95.

And yet if you then ask Claude how it worked that out, it will say something like: “I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95.” In other words, it gives you a common approach found everywhere online rather than what it actually did. Yep! LLMs are weird. (And not to be trusted.)

This is clear evidence that large language models will give reasons for what they do that do not necessarily reflect what they actually did. But this is true for people too, says Batson: “You ask somebody, ‘Why did you do that?’ And they’re like, ‘Um, I guess it’s because I was— .’ You know, maybe not. Maybe they were just hungry and that’s why they did it.”

Biran thinks this finding is especially interesting. Many researchers study the behavior of large language models by asking them to explain their actions. But that might be a risky approach, he says: “As models continue getting stronger, they must be equipped with better guardrails. I believe—and this work also shows—that relying only on model outputs is not enough.”

A third task that Anthropic studied was writing poems. The researchers wanted to know if the model really did just wing it, predicting one word at a time. Instead they found that Claude somehow looked ahead, picking the word at the end of the next line several words in advance.  

For example, when Claude was given the prompt “A rhyming couplet: He saw a carrot and had to grab it,” the model responded, “His hunger was like a starving rabbit.” But using their microscope, they saw that Claude had already hit upon the word “rabbit” when it was processing “grab it.” It then seemed to write the next line with that ending already in place.

This might sound like a tiny detail. But it goes against the common assumption that large language models always work by picking one word at a time in sequence. “The planning thing in poems blew me away,” says Batson. “Instead of at the very last minute trying to make the rhyme make sense, it knows where it’s going.”

“I thought that was cool,” says Merullo. “One of the joys of working in the field is moments like that. There’s been maybe small bits of evidence pointing toward the ability of models to plan ahead, but it’s been a big open question to what extent they do.”

Anthropic then confirmed its observation by turning off the placeholder component for “rabbitness.” Claude responded with “His hunger was a powerful habit.” And when the team replaced “rabbitness” with “greenness,”Claude responded with “freeing it from the garden’s green.”

Anthropic also explored why Claude sometimes made stuff up, a phenomenon known as hallucination. “Hallucination is the most natural thing in the world for these models, given how they’re just trained to give possible completions,” says Batson. “The real question is, ‘How in God’s name could you ever make it not do that?’”

The latest generation of large language models, like Claude 3.5 and Gemini and GPT-4o, hallucinate far less than previous versions, thanks to extensive post-training (the steps that take an LLM trained on text scraped from most of the internet and turn it into a usable chatbot). But Batson’s team was surprised to find that this post-training seems to have made Claude refuse to speculate as a default behavior. When it did respond with false information, it was because some other component had overridden the “don’t speculate” component.

This seemed to happen most often when the speculation involved a celebrity or other well-known entity. It’s as if the amount of information available on a subject pushed the speculation through, despite the default setting. When Anthropic overrode the “don’t speculate” component to test this, Claude produced lots of false statements about individuals, including claiming that Batson was famous for inventing the Batson principle (he isn’t).

Still unclear

Because we know so little about large language models, any new insight is a big step forward. “A deep understanding of how these models work under the hood would allow us to design and train models that are much better and stronger,” says Biran.

But Batson notes there are still serious limitations. “It’s a misconception that we’ve found all the components of the model or, like, a God’s-eye view,” he says. “Some things are in focus, but other things are still unclear—a distortion of the microscope.”

And it takes several hours for a human researcher to trace the responses to even very short prompts. What’s more, these models can do a remarkable number of different things, and Anthropic has so far looked at only 10 of them.

Batson also says there are big questions that this approach won’t answer. Circuit tracing can be used to peer at the structures inside a large language model, but it won’t tell you how or why those structures formed during training. “That’s a profound question that we don’t address at all in this work,” he says.

But Batson does see this as the start of a new era in which it is possible, at last, to find real evidence for how these models work: “We don’t have to be, like: ‘Are they thinking? Are they reasoning? Are they dreaming? Are they memorizing?’ Those are all analogies. But if we can literally see step by step what a model is doing, maybe now we don’t need analogies.”

Article link: https://www-technologyreview-com.cdn.ampproject.org/c/s/www.technologyreview.com/2025/03/27/1113916/anthropic-can-now-track-the-bizarre-inner-workings-of-a-large-language-model/amp/

BY JULIA MANCHESTER – 03/26/25 6:42 PM ET

Businessman Mark Cuban is making it his “mission” to shake up U.S. health care. 

Cuban is known for owning the Dallas Mavericks and for his role as an angel investor on the ABC reality TV show “Shark Tank.” 

He hasn’t shied away from the political arena as a prominent critic of President Trump. But Cuban says he has no interest in running for higher office, opting instead to disrupt health care. 

“If I could change health care in this country, that would be amazing,” Cuban said in an interview for The Hill’s “Health Next Summit.”

For Cuban, that starts with his company Cost Plus Drugs, which launched in 2022 with the goal of pharmacies leapfrogging middleman wholesalers. 

“I’m not saying I’m going to be able to pull it off, but I know we’ve had a significant impact with CostPlusDrugs.com, and I think getting there and focusing on that and just changing people’s lives for the better on the health care side, that’s my mission,” he said. 

Cuban said the company carries more than 2,500 medications and that when customers type in the name of a medication, they immediately see its actual cost followed by the markup, which is always 15 percent. The businessman said the company’s transparency is what sets it apart from its competitors, which include Amazon Prime, Costco and GoodRX.

“As crazy as it sounds, we are the only company that publishes their entire price list,” he said.

However, Cuban said Cost Plus Drugs has run into the roadblock of pharmacy benefit managers (PBMs). Those are the intermediaries that manage drug coverage for businesses across the country. PBMs and brand drug manufacturers negotiate discounts in the form of rebates, and the PBM then passes most of the rebates on to employers.

Critics argue PBMs do not always find the best deals and are forcing independent drug stores out of business by not paying them enough to cover their costs. The PBM industry, says it saves employers and patients billions on drug costs. 

“When we talk to manufacturers and we say, ‘Look, why can’t we get in the same rebate programs as these big PBMs?’ They tell us without telling us that they don’t want to lose their position on the formulary, which is hundreds of millions or billions of dollars in sales,” Cuban said. 

“The big PBMs just won’t let them deal with us. That is our problem. Period. End of story,” he added. 

But Cuban said he and his team are slowly making progress through going to drug manufacturer CEOs and working with pass-through PBMs, which eliminate discounts, rebates and fees directly to the payer. 

“All of the rebates are passed through so you actually get significant savings, and slowly but surely we’re having inroads,” he said. 

Cuban is also touting the company’s partnership with independent pharmacies through paying them a $12 fill fee per prescription, fully reimbursing the pharmacy and allowing them to make a profit. 

“I want independent pharmacies to stay in business,” Cuban said. “I think as a country, we don’t want to see pharmacy deserts. We need them to stay in business. All those people, senior citizens that have been going to the same pharmacist for decades and that pharmacist knows them, they know their family, that’s important. That saves lives.”

“The problem is that the biggest PBMs won’t fully reimburse,” he said. 

Cuban argued that Cost Plus Drugs would be able to dramatically cut the cost of brand medication if the pharmaceutical industry allows the company to participate in rebate programs. 

“The amount of money that we can save taxpayers and patients will be, to paraphrase, like you’ve never seen before. Most in history ever,” he said, appearing to mimic Trump. 

Cuban said he is hopeful the Trump administration will work to lower prices through the Centers for Medicare and Medicaid Services (CMS) and the Department of Health and Human Services (HHS). However, he added that he believes Republicans have been afraid to work with him, particularly those from the Department of Government Efficiency (DOGE)

Cuban described the approach of the Trump administration and DOGE head Elon Musk as “ready, fire, aim,” which, he added, “is no way to govern.” 

“Particularly when the you-know-what, you know, rolls downhill onto the small-to-medium-sized communities and cities where all of a sudden, who knows how many people are losing jobs, who knows how many companies have to close because their grants have been cut, and who knows the impact on that community in terms of services they’re going to be able to offer, raising taxes,” he said. 

Cuban is one of the most outspoken critics of Trump from the business community, and he opted to campaign with then-Vice President Kamala Harris last year.

Cuban said he does not particularly care about the future of the Democratic Party because of his status as an Independent. However, he encouraged Democrats to go into communities to get a sense of how Trump’s tariffs and DOGE cuts are playing out locally. 

“You’re not going to get people to all of the sudden turn on Donald Trump, but what you can do, as we’ve seen with some of the town halls, is get people to turn on some of the Republican House members who are going to have to make a really tough choice,” he said. 

“If their communities are being negatively impacted by the all-at-once cuts of DOGE and the impact of the tariffs, well, put the pressure on those House Republicans to make a choice. Either you support Donald Trump, or you go against Donald Trump and say these tariffs and these cuts are awful for my town, or you lose your job,” he said. 

Article link: https://thehill.com/policy/healthcare/5216267-mark-cuban-health-care/

What happens when quantum computers can finally crack encryption and break into the world’s best-kept secrets? It’s called Q-Day—the worst holiday maybe ever.

AMIT KATWALA

MAR 24, 2025 6:00 AM

ILLUSTRATION: NICHOLAS LAW

ONE DAY SOON, at a research lab near Santa Barbara or Seattle or a secret facility in the Chinese mountains, it will begin: the sudden unlocking of the world’s secrets. Your secrets.

Cybersecurity analysts call this Q-Day—the day someone builds a quantum computer that can crack the most widely used forms of encryption. These math problems have kept humanity’s intimate data safe for decades, but on Q-Day, everything could become vulnerable, for everyone: emails, text messages, anonymous posts, location histories, bitcoin wallets, police reports, hospital records, power stations, the entire global financial system.

“We’re kind of playing Russian roulette,” says Michele Mosca, who coauthored the most recent “Quantum Threat Timeline” report from the Global Risk Institute, which estimates how long we have left. “You’ll probably win if you only play once, but it’s not a good game to play.” When Mosca and his colleagues surveyed cybersecurity experts last year, the forecast was sobering: a one-in-three chance that Q-Day happens before 2035. And the chances it has already happened in secret? Some people I spoke to estimated 15 percent—about the same as you’d get from one spin of the revolver cylinder.

The corporate AI wars may have stolen headlines in recent years, but the quantum arms race has been heating up too. Where today’s AI pushes the limits of classical computing—the kind that runs on 0s and 1s—quantum technology represents an altogether different form of computing. By harnessing the spooky mechanics of the subatomic world, it can run on 0s, 1s, or anything in between. This makes quantum computers pretty terrible at, say, storing data but potentially very good at, say, finding the recipe for a futuristic new material (or your email password). The classical machine is doomed to a life of stepwise calculation: Try one set of ingredients, fail, scrap everything, try again. But quantum computers can explore many potential recipes simultaneously.

So, naturally, tech giants such as Google, Huawei, IBM, and Microsoft have been chasing quantum’s myriad positive applications—not only for materials science but also communications, drug development, and market analysis. China is plowing vast resources into state-backed efforts, and both the US and the European Union have pledged millions in funding to support homegrown quantum industries. Of course, whoever wins the race won’t just have the next great engine of world-saving innovation. They’ll also have the greatest code-breaking machine in history. So it’s normal to wonder: What kind of Q-Day will humanity get—and is there anything we can do to prepare?

If you had a universal picklock, you might tell everyone—or you might keep it hidden in your pocket for as long as you possibly could. From a typical person’s vantage point, maybe Q-Day wouldn’t be recognizable as Q-Day at all. Maybe it would look like a series of strange and apparently unconnected news stories spread out over months or years. London’s energy grid goes down on election day, plunging the city into darkness. A US submarine on a covert mission surfaces to find itself surrounded by enemy ships. Embarrassing material starts to show up online in greater and greater quantities: classified intelligence cables, presidential cover-ups, billionaires’ dick pics. In this scenario, it might be decades before we’re able to pin down exactly when Q-Day actually happened.

Then again, maybe the holder of the universal picklock prefers the disaster-movie outcome: everything, everywhere, all at once. Destroy the grid. Disable the missile silos. Take down the banking system. Open all the doors and let the secrets out.

ILLUSTRATION: NICHOLAS LAW

SUPPOSE YOU ASK a classical computer to solve a simple math problem: Break the number 15 into its smallest prime factors. The computer would try all the options one by one and give you a near-instantaneous answer: 3 and 5. If you then ask the computer to factor a number with 1,000 digits, it would tackle the problem in exactly the same way—but the calculation would take millennia. This is the key to a lot of modern cryptography.

Take RSA encryption, developed in the late 1970s and still usedfor securing email, websites, and much more. In RSA, you (or your encrypted messaging app of choice) create a private key, which consists of two or more large prime numbers. Those numbers, multiplied together, form part of your public key. When someone wants to send you a message, they use your public key to encrypt it. You’re the only person who knows the original prime numbers, so you’re the only person who can decrypt it. Until, that is, someone else builds a quantum computer that can use its spooky powers of parallel computation to derive the private key from the public one—not in millennia but in minutes. Then the whole system collapses.

The algorithm to do this already exists. In 1994, decades before anyone had built a real quantum computer, an AT&T Bell Labs researcher named Peter Shor designed the killer Q-Day app. Shor’s algorithm takes advantage of the fact that quantum computers run not on bits but on qubits. Rather than being locked in a state of 0 or 1, they can exist as both simultaneously—in superposition. When you run an operation on a handful of qubits in a given quantum state, you’re actually running that same operation on those same qubits in all their potential quantum states. With qubits, you’re not confined to trial and error. A quantum computer can explore all potential solutions simultaneously. You’re calculating probability distributions, waves of quantum feedback that pile onto each other and peak at the correct answer. With Shor’s algorithm, carefully designed to amplify certain mathematical patterns, that’s exactly what happens: Large numbers go in one end, factors come out the other.

In theory, at least. Qubits are incredibly difficult to build in real life, because the slightest environmental interference can nudge them out of the delicate state of superposition, where they balance like a spinning coin. But Shor’s algorithm ignited interest in the field, and by the 2010s, a number of projects were starting to make progress on building the first qubits. In 2016, perhaps sensing the nascent threat of Q-Day, the US National Institute for Standards and Technology (NIST) launched a competition to develop quantum-proof encryption algorithms. These largely work by presenting quantum computers with complex multidimensional mazes, called structured lattices, that even they can’t navigate without directions.

In 2019, Google’s quantum lab in Santa Barbara claimed that it had achieved “quantum supremacy.” Its 53-qubit chip could complete in just 200 seconds a task that would have taken 100,000 conventional computers about 10,000 years. Google’s latest quantum processor, Willow, has 105 qubits. But to break encryption with Shor’s algorithm, a quantum computer will need thousands or even millions.

There are now hundreds of companies trying to build quantum computers using wildly different methods, all geared toward keeping qubits isolated from the environment and under control: superconducting circuits, trapped ions, molecular magnets, carbon nanospheres. While progress on hardware inches forward, computer scientists are refining quantum algorithms, trying to reduce the number of qubits required to run them. Each step brings Q-Day closer.

That’s bad news not just for RSA but also for a dizzying array of other systems that will be vulnerable on Q-Day. Security consultant Roger A. Grimes lists some of them in his book Cryptography Apocalypse: the DSA encryption used by many US government agencies until recently, the elliptic-curve cryptography used to secure cryptocurrencies like Bitcoin and Ethereum, the VPNs that let political activists and porn aficionados browse the web in secrecy, the random number generators that power online casinos, the smartcards that let you tap through locked doors at work, the security on your home Wi-Fi network, the two-factor authentication you use to log in to your email account.

Experts from one national security agency told me they break the resulting threats down into two broad areas: confidentiality and authentication. In other words, keeping secrets and controlling access to critical systems. Chris Demchak, a former US Army officer who is a professor of cybersecurity at the US Naval War College and spoke with me in a personal capacity, says that a Q-Day computer could let an adversary eavesdrop on classified military data in real time. “It would be very bad if they knew exactly where all of our submarines were,” Demchak says. “It would be very bad if they knew exactly what our satellites are looking at. And it would be very bad if they knew exactly how many missiles we had and their range.” The balance of geopolitical power in, say, the Taiwan Strait could quickly tilt.

Beyond that real-time threat to confidentiality, there’s also the prospect of “harvest now, decrypt later” attacks. Hackers aligned with the Chinese state have reportedly been hoovering up encrypted data for years in hopes of one day having a quantum computer that can crack it. “They wolf up everything,” Demchak told me. (The US almost certainly does this too.) The question then becomes: How long will your sensitive data remain valuable? “There might be some needles in that haystack,” says Brian Mullins, the CEO of Mind Foundry, which helps companies implement quantum technology. Your current credit card details might be irrelevant in 10 years, but your fingerprint won’t be. A list of intelligence assets from the end of the Iraq War might seem useless until one of those assets becomes a prominent politician.

The threat to authentication may be even scarier. “Pretty much anything that says a person is who they say they are is underpinned by encryption,” says Deborah Frincke, a computer scientist and national security expert at Sandia National Laboratories. “Some of the most sensitive and valuable infrastructure that we have would be open to somebody coming in and pretending to be the rightful owner and issuing some kind of command: to shut down a network, to influence the energy grid, to create financial disruption by shutting down the stock market.”

ILLUSTRATION: NICHOLAS LAW

THE EXACT LEVEL of Q-Day chaos will depend on who has access to the first cryptographically relevant quantum computers. If it’s the United States, there will be a “fierce debate” at the highest levels of government, Demchak believes, over whether to release it for scientific purposes or keep it secret and use it for intelligence. “If a private company gets there first, the US will buy it and the Chinese will try to hack it,” she claims. If it’s one of the US tech companies, the government could put it under the strict export controls that now apply to AI chips.

Most nation-state attacks are on private companies—say, someone trying to break into a defense contractor like Lockheed Martin and steal plans for a next-generation fighter jet. But over time, as quantum computers become more widely available, the focus of the attacks could broaden. The likes of Microsoft and Amazon are already offering researchers access to their primitive quantum devices on the cloud—and big tech companies haven’t always been great at policing who uses their platforms. (The soldier who blew up a Cybertruck outside the Trump International Hotel in Las Vegas early this year queried ChatGPT to help plan the attack.) You could have a bizarre scenario where a cybercriminal uses Amazon’s cloud quantum computing platform to break into Amazon Web Services.

Cybercriminals with access to a quantum computer could use it to go after the same targets more effectively, or take bigger swings: hijacking the SWIFT international payments system to redirect money transfers, or conducting corporate espionage to collect kompromat. The earliest quantum computers probably won’t be able to run Shor’s algorithm that quickly—they might only get one or two keys a day. But combining a quantum computer with an artificial intelligence that can map out an organization’s weakness and highlight which keys to decrypt to cause the most damage could yield devastating results.

And then there’s Bitcoin. The cryptocurrency is exquisitely vulnerable to Q-Day. Because each block in the Bitcoin blockchain captures the data from the previous block, Bitcoin cannot be upgraded to post-quantum cryptography, according to Kapil Dhiman, CEO of Quranium, a post-quantum blockchain security company. “The only solution to that seems to be a hard fork—give birth to a new chain and the old chain dies.”

But that would require a massive organizational effort. First, 51 percent of Bitcoin node operators would have to agree. Then everyone who holds bitcoin would have to manually move their funds from the old chain to the new one (including the elusive Satoshi Nakamoto, the Bitcoin developer who controls wallets containing around $100 billion of the cryptocurrency). If Q-Day happens before the hard fork, there’s nothing to stop bitcoin going to zero. “It’s like a time bomb,” says Dhiman.

THAT BOMB GOING off will only be the beginning. When Q-Day becomes public knowledge, either via grim governmental address or cheery big-tech press release, the world will enter the post-quantum age. It will be an era defined by mistrust and panic—the end of digital security as we know it. “And then the scramble begins,” says Demchak.

All confidence in the confidentiality of our communications will collapse. Of course, it’s unlikely that everyone’s messages will actually be targeted, but the perception that you could be spied on at any time will change the way we live. And if NIST’s quantum-proof algorithms haven’t rolled out to your devices by that point, you face a real problem—because any attempts to install updates over the cloud will also be suspect. What if that download from Apple isn’t actually from Apple? Can you trust the instructions telling you to transfer your crypto to a new quantum-secure wallet?

Grimes, the author of Cryptography Apocalypse, predicts enormous disruptions. We might have to revert to Cold War methods of transmitting sensitive data. (It’s rumored that after a major hack in 2011, one contractor purportedly asked its staff to stop using email for six weeks.) Fill a hard drive, lock it in a briefcase, put someone you trust on a plane with the payload handcuffed to their wrist. Or use one-time pads—books of pre-agreed codes to encrypt and decrypt messages. Quantum-secure, but not very scalable. Expect major industries—energy, finance, health care, manufacturing, transportation—to slow to a crawl as companies with sensitive data switch to paper-based methods of doing business and scramble to hire expensive cryptography consultants. There will be a spike in inflation. Most people might just accept the inevitable: a post-privacy society in which any expectation of secrecy evaporates unless you’re talking to someone in person in a secluded area with your phones switched off. Big Quantum is Watching You.

The best-case scenario looks something like Y2K, where we have a collective panic, make the necessary upgrades to encryption, and by the time Q-Day rolls around it’s such an anticlimax that it becomes a joke. That outcome may still be possible. Last summer, NIST released its first set of post-quantum encryption standards. One of Joe Biden’s last acts as president was to sign an executive order changing the deadline for government agencies to implement NIST’s algorithms from 2035 to “as soon as practicable.”

Already, NIST’s post-quantum cryptography has been rolled out on messaging platforms such as Signal and iMessage. Sources told me that sensitive national security data is probably being locked up in ways that are quantum-secure. But while your email account can easily be Q-proofed over the internet (assuming the update doesn’t come from a quantum imposter!), other things can’t. Public bodies like the UK’s National Health Service are still using hardware and software from the 1990s. “Microsoft is not going to upgrade some of its oldest operating systems to be post-quantum secure,” says Ali El Kaafarani, the CEO of PQShield, a company that makes quantum-resistant hardware. Updates to physical infrastructure can take decades, and some of that infrastructure has vulnerable cryptography in places it can’t be changed: The energy grid, military hardware, and satellites could all be at risk.

And there’s a balance to be struck. Rushing the transition risks introducing vulnerabilities that weren’t there before. “How do you make transitions slow enough that you can be confident and fast enough that you don’t dawdle?” asks Chris Ballance, CEO of Oxford Ionics, a quantum computing company. Some of those vulnerabilities might even be there by design: Memos leaked by Edward Snowden indicate that the NSA may have inserted a backdoor into a pseudorandom number generator that was adopted by NIST in 2006. “Anytime anybody says you should use this particular algorithm and there’s a nation-state behind it, you’ve got to wonder whether there’s a vested interest,” says Rob Young, director of Lancaster University’s Quantum Technology Centre.

Then again, several people I spoke to pointed out that any nation-state with the financial muscle and technical knowledge to build a quantum device that can run Shor’s algorithm could just as easily compromise the financial system, the energy grid, or an enemy’s security apparatus through conventional methods. Why invent a new computing paradigm when you can just bribe a janitor?

Long before quantum technology is good enough to break encryption, it will be commercially and scientifically useful enough to tilt the global balance. As researchers solve the engineering challenge of isolating qubits from the environment, they’ll develop exquisitely sensitive quantum sensors that will be able to unmask stealth ships and map hidden bunkers, or give us new insight into the human body. Similarly, pharma companies of the future could use quantum to steal a rival’s inventions—or use it to dream up even better ones. So ultimately the best way to stave off Q-Day may be to share those benefits around: Take the better batteries, the miracle drugs, the far-sighted climate forecasting, and use them to build a quantum utopia of new materials and better lives for everyone. Or—let the scramble begin.

---

Let us know what you think about this article. Submit a letter to the editor at mail@wired.com.

Article link: https://www.wired.com/story/q-day-apocalypse-quantum-computers-encryption/?

From 2023, Lesley Stahl’s report on AI, chatbots and a world of unknowns. From 2024, Stahl’s story on Kenyan workers training AI who say they’re overworked, underpaid and exploited by big American tech companies. Also from 2024, Anderson Cooper’s report on “nudify” sites that use AI to create realistic, revealing images of actual people. And from 2021, Bill Whitaker’s look at the use of artificial intelligence to create deepfakes.

Article link: https://youtu.be/LgUjLcxJxVg?si=dfIsAj3SyW4Jh1FY

MIT researchers developed a photon-shuttling “interconnect” that can facilitate remote entanglement, a key step toward a practical quantum computer.

Adam Zewe | MIT News

Publication Date: March 21, 2025

Quantum computers have the potential to solve complex problems that would be impossible for the most powerful classical supercomputer to crack.

Just like a classical computer has separate, yet interconnected, components that must work together, such as a memory chip and a CPU on a motherboard, a quantum computer will need to communicate quantum information between multiple processors.

Current architectures used to interconnect superconducting quantum processors are “point-to-point” in connectivity, meaning they require a series of transfers between network nodes, with compounding error rates.

On the way to overcoming these challenges, MIT researchers developed a new interconnect device that can support scalable, “all-to-all” communication, such that all superconducting quantum processors in a network can communication directly with each other.

They created a network of two quantum processors and used their interconnect to send microwave photons back and forth on demand in a user-defined direction. Photons are particles of light that can carry quantum information.

The device includes a superconducting wire, or waveguide, that shuttles photons between processors and can be routed as far as needed. The researchers can couple any number of modules to it, efficiently transmitting information between a scalable network of processors.

They used this interconnect to demonstrate remote entanglement, a type of correlation between quantum processors that are not physically connected. Remote entanglement is a key step toward developing a powerful, distributed network of many quantum processors.

“In the future, a quantum computer will probably need both local and nonlocal interconnects. Local interconnects are natural in arrays of superconducting qubits. Ours allows for more nonlocal connections. We can send photons at different frequencies, times, and in two propagation directions, which gives our network more flexibility and throughput,” says Aziza Almanakly, an electrical engineering and computer science graduate student in the Engineering Quantum Systems group of the Research Laboratory of Electronics (RLE) and lead author of a paper on the interconnect.

Her co-authors include Beatriz Yankelevich, a graduate student in the EQuS Group; senior author William D. Oliver, the Henry Ellis Warren (1894) Professor of Electrical Engineering and Computer Science (EECS) and professor of Physics, director of the Center for Quantum Engineering, and associate director of RLE; and others at MIT and Lincoln Laboratory. The research appears today in Nature Physics.

A scalable architecture

The researchers previously developed a quantum computing module, which enabled them to send information-carrying microwave photons in either direction along a waveguide.

In the new work, they took that architecture a step further by connecting two modules to a waveguide in order to emit photons in a desired direction and then absorb them at the other end.

Each module is composed of four qubits, which serve as an interface between the waveguide carrying the photons and the larger quantum processors.

The qubits coupled to the waveguide emit and absorb photons, which are then transferred to nearby data qubits.

The researchers use a series of microwave pulses to add energy to a qubit, which then emits a photon. Carefully controlling the phase of those pulses enables a quantum interference effect that allows them to emit the photon in either direction along the waveguide. Reversing the pulses in time enables a qubit in another module any arbitrary distance away to absorb the photon.

“Pitching and catching photons enables us to create a ‘quantum interconnect’ between nonlocal quantum processors, and with quantum interconnects comes remote entanglement,” explains Oliver.

“Generating remote entanglement is a crucial step toward building a large-scale quantum processor from smaller-scale modules. Even after that photon is gone, we have a correlation between two distant, or ‘nonlocal,’ qubits. Remote entanglement allows us to take advantage of these correlations and perform parallel operations between two qubits, even though they are no longer connected and may be far apart,” Yankelevich explains.

However, transferring a photon between two modules is not enough to generate remote entanglement. The researchers need to prepare the qubits and the photon so the modules “share” the photon at the end of the protocol.

Generating entanglement

The team did this by halting the photon emission pulses halfway through their duration. In quantum mechanical terms, the photon is both retained and emitted. Classically, one can think that half-a-photon is retained and half is emitted.

Once the receiver module absorbs that “half-photon,” the two modules become entangled.

But as the photon travels, joints, wire bonds, and connections in the waveguide distort the photon and limit the absorption efficiency of the receiving module.

To generate remote entanglement with high enough fidelity, or accuracy, the researchers needed to maximize how often the photon is absorbed at the other end.

“The challenge in this work was shaping the photon appropriately so we could maximize the absorption efficiency,” Almanakly says.

They used a reinforcement learning algorithm to “predistort” the photon. The algorithm optimized the protocol pulses in order to shape the photon for maximal absorption efficiency.

When they implemented this optimized absorption protocol, they were able to show photon absorption efficiency greater than 60 percent.

This absorption efficiency is high enough to prove that the resulting state at the end of the protocol is entangled, a major milestone in this demonstration.

“We can use this architecture to create a network with all-to-all connectivity. This means we can have multiple modules, all along the same bus, and we can create remote entanglement among any pair of our choosing,” Yankelevich says.

In the future, they could improve the absorption efficiency by optimizing the path over which the photons propagate, perhaps by integrating modules in 3D instead of having a superconducting wire connecting separate microwave packages. They could also make the protocol faster so there are fewer chances for errors to accumulate.

“In principle, our remote entanglement generation protocol can also be expanded to other kinds of quantum computers and bigger quantum internet systems,” Almanakly says.

This work was funded, in part, by the U.S. Army Research Office, the AWS Center for Quantum Computing, and the U.S. Air Force Office of Scientific Research. 

Article link: https://news.mit.edu/2025/device-enables-direct-communication-among-multiple-quantum-processors-0321

BYSHARON GOLDMAN

March 19, 2025 at 5:30 AM EDT

Last year, Nvidia’s annual GTC conference—hailed as the “Woodstock of AI”—drew a crowd of 18,000 to a packed arena befitting rock legends like the Rolling Stones. On stage, CEO Jensen Huang, clad in a shiny black leather jacket, delivered his keynote for the AI chip behemoth’s annual developer’s conference with the flair of a headlining act.

Today, a year later, Huang was onstage once again, shooting off a series of T-shirt cannons and clad this time in an edgy motorcycle black leather jacket worthy of a halftime show. This time, Nvidia-watchers tossed around the metaphor of the “Super Bowl of AI” like a football. Nvidia did not shy away from the pigskin comparison, offering a keynote “pre-game” event and a live broadcast that had guest commentators like Dell CEO Michael Dell calling plays on how Nvidia would continue to rule the AI world. 

As Huang took the stage in front of a stadium-sized image of the Nvidia headquarters—making sure to highlight the “gaussian splatting” 3D rendering tech behind it to his high-tech audience—his message was clear, even if unspoken: Nvidia’s best defense is a strong offense. With recent reasoning models from Chinese startup DeepSeek shaking up AI, followed by others from companies including OpenAI, Baidu and Google, Nvidia wants its business customers to know they need its GPUs and software more than ever. 

That’s because DeepSeek’s R1 model, which debuted in January, created some doubts about Nvidia’s momentum. The new model, its maker claimed, had been trained for a fraction of the cost and computing power of U.S. models. As a result, Nvidia’s stock took a beating from investors worried that companies would no longer need to buy as many of Nvidia’s chips. 

Reasoning models require more computing power

But Huang thinks those selling off made a big mistake. Reasoning models, he said, require morecomputing power, not less. A lot more, in fact, thanks to their more detailed answers, or in the parlance of AI folks, “inference.” The ChatGPT revolution was about a chatbot spitting out answers to queries—but today’s models must “think” harder, which requires more “tokens,” or the fundamental units text models use—whether it’s a word in a phrase or just part of a word. 

The more tokens used, the more efficiency customers demand, and the more computing power AI reasoning models will require. So making sure Nvidia customers can process more tokens, faster, is the not-so-secret Nvidia play—and Huang did not need to mention DeepSeek until one hour into the keynote to get that point across. 

All of the Nvidia GTC announcements that followed were positioned with that in mind. Stock-watchers might well have wanted to see an accelerated timeline for Nvidia’s new AI chip, the Vera Rubin, to be released at the end of 2026, or more details about the company’s short-term roadmap. But Huang focused on the fact that while AI pundits had insisted over the past year that the pace of AI once rapid improvements were slowing down, Nvidia believes getting AI improvements to “scale” is increasing faster than ever. Of course, that would be to Nvidia’s benefit in terms of revenue. “The amount of computation we need as a result of agentic AI, as a result of reasoning, is easily 100 times more than we thought we needed this time last year,” Huang said.  

Will Nvidia’s efforts to drive growth be enough to win? 

Nvidia’s announcements that followed were all about making sure customers understand they will have everything they need to keep up in a world where extreme speed at providing detailed answers and better reasoning will be the difference between a company’s AI success and failure. Blackwell GPUs, Nvidia’s latest, top of the line AI chips, are in full production—with 3.6 million of them already used. An upgraded version, the Blackwell Ultra, boasts 3x performance. The new Vera Rubin chip and infrastructure is on the way. Nvidia’s “world’s smallest AI supercomputer” is at the ready. Software for AI agents is quickly being used in the physical world, including self-driving cars, robotics, and manufacturing. 

But will Nvidia’s efforts to drive growth be enough to keep enterprise companies investing in Nvidia products? Will buying Nvidia’s costly AI chips—which can cost between $30,000 to $40,000 each, prove too expensive, given the still-unclear-ROI of AI investments? Ultimately, Nvidia’s premium picks and shovels require enough customers willing to keep digging. 

Huang is confident that there are enough—and that Nvidia’s Super Bowl win is not just a victory for the 31-year-old company. “Everyone wins,” he insisted.

Perhaps, but there is no doubt that as Nvidia seeks to establish a dynasty in the AI era, expectations remain higher than ever. Huang, for his part, appears undaunted even as the AI continues to evolve at high speed. He’s always reaching for the brass ring, it seems—or in this case, the Super Bowl ring.

Article link: https://fortune.com/2025/03/19/nvidia-jensen-gtc-super-bowl-ai-deepseek/