Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Researchers unveiled the world’s first wooden satellite last month, billing it as clearing a path for more uses of wood in outer space. The material will be more sustainable and less polluting than the metals used in conventional satellites, they say.

Researchers at Kyoto University in Japan and the Tokyo-based logging company Sumitomo Forestry showed off the satellite, called LignoSat, in late May. The roughly 10-centimetre-long cube is made of magnolia-wood panels and has an aluminium frame, solar panels, circuit boards and sensors. The panels incorporate Japanese wood-joinery methods that do not rely on glue or metal fittings.

Wood might seem counterintuitive for use in space because it is combustible — but that feature can be desirable. To curb the growing problem of space junk threatening spacecraft and space stations, rocket stages and satellites are deliberately plunged into the Earth’s atmosphere to burn up. But during combustion, they release particles of aluminium and other metals. Many more spacecraft launches are planned, and scientists have warned that the environmental effects of this pollution are unknown.

When LignoSat plunges back to Earth, after six months to a year of service, the magnolia will incinerate completely and release only water vapour and carbon dioxide, says Takao Doi, an astronaut and engineer at Kyoto University, who is part of the research team. He points to other benefits of wood: it’s resilient in the harsh environment of space and does not block radio waves, making it suitable for enclosing an antenna.

And there is a precedent for spacecraft with wooden parts. Launched in 1962, NASA’s Ranger 3 lunar probe had a balsa-wood casing intended to protect its capsule as it landed on the lunar surface (the probe malfunctioned, missed the Moon and began orbiting the Sun).
Timber pioneers

LignoSat will cost about US$191,000 to design, manufacture, launch and operate. Sensors onboard will evaluate strain on the wood, temperature, geomagnetic forces and cosmic radiation, as well as receive and transmit radio signals. The satellite has been handed over to the Japan Aerospace Exploration Agency (JAXA) and will be transferred to the International Space Station in September, before being launched into orbit in November.

Growth has been slow for the project, which began in 2020 with speculation about the wider potential for wood in space for better sustainability.

“In our first conversations, Dr Doi proposed we build wooden housing on the Moon,” says team member Koji Murata at the biomaterials-design laboratory at Kyoto University’s Graduate School of Agriculture. “We have also discussed the possibility of building domes on Mars out of wood in order to grow timber forests.”

Martian and lunar colonists, like all pioneers, would have to make use of local materials — regolith (rocky material on the surface), silicon dioxide and other minerals, in the case of Mars. But wood could play a part in crafting temporary or permanent shelters. Murata points to plans by JAXA and industrial partners to develop shelters made partly of wood that could be used in Antarctica or on the Moon.

“The natural radiation-shielding properties of wood could be used effectively to design walls or outer shells of space habitats to provide protection,” says Nisa Salim, who specializes in engineered materials at Swinburne University of Technology in Melbourne, Australia, and is not part of the project. “Wood is an effective insulator, capable of regulating temperature and minimizing heat transfer to maintain a comfortable indoor environment. Wood is easy to work with, renewable and biodegradable, aligning with sustainability goals for space exploration.”

Salim noted that the structural integrity, safety and longevity of wood need to be confirmed in space.

Wood consists of cellulose held together by lignin, a kind of organic polymer. That makes it a naturally occurring member of the class of materials known as composites, says Scott J McCormack, a materials engineer at the University of California, Davis, who is not involved in the project. Composites are often used in the aerospace industry, so he does not find it surprising that their use in satellites might be explored.

“Composites are ideal for the aerospace industry — and also satellites — due to their high strength-to-weight ratio,” says McCormack. But he has doubts about how wood will fare as a structural material on the Moon or Mars. “The first concern that comes to mind is galactic cosmic radiation [GCR] and how it might degrade the mechanical properties of wood over time. GCR isn’t that big of problem for us here on Earth, thanks to our atmosphere.”

But Murata says that the team has studied measurements of GCR and solar energetic particles — high-energy particles that are released from the Sun — taken by NASA’s Curiosity rover on Mars, as well as the effects of gamma rays on wood on Earth. He thinks that wood on Mars could potentially last for thousands of years. “Radiation on Mars is a big problem for living organisms, including humans,” he says. “I don't think this is going to be much of an issue for wood.”

Source

Today, Microsoft offered more details on the new Automatic Super Resolution feature, which will be included in the recently announced Copilot+ PC notebooks that run on the Qualcomm Snapdragon X Plus and Elite processors.

In a developer blog post, Microsoft stated:

Auto SR stands out in two exciting ways: it’s applied automatically and enhances existing games. Like other SR technologies, Auto SR uses AI to improve frame rates and image quality. However, Auto SR focuses on bringing the benefits of super resolution to your existing game library with no manual configurations needed. This makes it ideal for players who prefer a straightforward experience. Simply start your game, and Auto SR instantly enhances it, allowing you to effortlessly enjoy visuals that surpass native 1080p quality with the fast frame rates typically seen at lower resolutions.

Because the AI model that is uses to run the Auto SR feature needs a Neural Processing Unit, that's why this feature is making its debut on the first Copilot+ PC laptops with the Snapdragon X chips, which have an NPU processor.

Microsoft said its approach for AutoSR is different than the AI game upscaling features that are available from Nvidia's DLSS Super Resolution, AMD's FidelityFX Super Resolution, and Intel's XeSS

These approaches require games to alter their rendering, for example, by using jitter and MIP bias to add finer details. In contrast, Auto SR tackles the task of enhancing games without the extra information to improve games as they exist today. It relies on larger models and the NPU to create stunning visuals.

The blog post offered up some screenshots of a test game to show the benefits of Auto SR. The screenshot above shows a game running normally at 1440p resolution and at 24 frames per second. The second is running at 720p resolution but with the Auto SR feature turned on and at 40 frames per second. The game looks nearly identical in both screenshots,

Microsoft does admit that using Auto SR on games could create "some softening of text and HUD elements" which is why you won't see this tech used on text-heavy games. Copilot+ PC owners can toggle Auto SR on or off, or adjust it for individual games.

Microsoft previously released the system requirements for Auto SR PCs. When those first Copilot+ PC notebooks launch on June 18, 11 games will be set up by default to run on Auto SR:

• BeamNG.drive
• Borderlands 3
• Control (dx11)
• Dark Souls III
• God of War
• Kingdom Come: Deliverance
• Resident Evil 2
• Resident Evil 3
• Sekiro Shadows Die Twice
• Sniper Ghost Warrior Contracts 2
• The Witcher 3

Microsoft stated that in the future, it will add improvements for running Auto SR games with additional resolutions, multi-monitor support, and HDR content.

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

It's a nightmare scenario for Microsoft. The headlining feature of its new Copilot+ PC initiative, which is supposed to drive millions of PC sales over the next couple of years, is under significant fire for being what many say is a major breach of privacy and security on Windows. That feature in question is Windows Recall, a new AI tool designed to remember everything you do on Windows.

On paper, it's a cool idea. As CEO Satya Nadella described it, Windows now has a photographic memory that uses AI to triage and index everything you've ever done on your computer, enabling you to semantically search for things you've seen using natural language. It's a new and improved way of finding things on Windows, and in our testing of the feature, it works really well.

However, for a tool like this to be feasible, trust between the user and the platform is required, a luxury Microsoft doesn't appear to have with its Windows user base right now. Recall operates by taking and storing captures of your screen every few seconds to build a database that the user can later search, with screenshots as visual aids. That database is stored locally on your device and never uploaded to the cloud.

In fact, Microsoft goes so far as to promise that it cannot see the data collected by Windows Recall, that it can't train any of its AI models on your data, and that it definitely can't sell that data to advertisers. All of this is true, but that doesn't mean people believe Microsoft when it says these things. In fact, many have jumped to the conclusion that even if it's true today, it won't be true in the future.

Microsoft eroded user trust on Windows with bad practices that are now biting them back

Microsoft employs some bad practices on Windows 11 to squeeze money out of its users.

(Image credit: Microsoft)

Microsoft is fully aware that the concept of Windows Recall sounds creepy. I know that the company spent a lot of time internally figuring out how to communicate this feature to the world, but it turns out there's no good way to communicate something like this when your users don't trust you.

Users are describing the feature as literal spyware or malware, and droves of people are proclaiming they will proudly switch to Linux or Mac in the wake of it. Microsoft simply doesn't enjoy the same benefit of the doubt that other tech giants like Apple may have.

People think Windows Recall is malware or spyware.

(Image credit: Windows Central)

Had Apple announced a feature like Recall, there would have been much less backlash, as Apple has done a great job building loyalty and trust with its users, prioritizing polished software experiences, and positioning privacy as a high-level concern for the company.

Don Estridge literally gifted the entire personal computing market to IBM. When we first saw him in this series, he was competing with Lore Harp to build the first personal computer. We’ve seen that story from her perspective—now it's time for Estridge's. You might think that the man who ushered IBM into the PC age would have been a central figure within the company, but in reality, he worked in the relative backwater of its Boca Raton, Florida office. This is the story of how Estridge successfully navigated complicated corporate politics to build the iconic IBM PC—and his shocking betrayal. It's the latest installment of The Crazy Ones, Gareth Edwards's monthly column about the forgotten men and women who built the future of technology. Subscribe to Every so that you don't miss out.—Kate Lee

In a burnished-oak corridor outside the committee room at IBM’s headquarters in August 1980, two engineers pace nervously. Eventually, a door opens. Their boss, Bill Lowe, emerges from the board room next door. Before they can say anything, he smiles and nods. They laugh. They can’t quite believe it. It’s official. IBM is going to try and build a home computer.

Bill Lowe kicked off this ambitious project, but he wouldn’t be the person who would finish it. That role would fall to his successor, a humble, cowboy boot-wearing mid-level executive, out of favor and kicking his heels in the IBM corporate backwater of Boca Raton, Florida. He would take Lowe’s project forward, one nobody else in the company wanted. Just 12 months later, on August 15, 1981, a computer would launch that would change the world: the IBM PC.

This is the story of Don Estridge, the man who brought the IBM PC to market and changed business and home computing forever. In just five years he created an IBM division that almost nobody else in the company wanted to exist. By 1983, it had seized 70 percent of the microcomputer market and was valued at over $4 billion ($12 billion today). Under Estridge, IBM’s PC division sold over 1 million machines a year, making it the third largest computer manufacturer in the world on its own. This story is based on contemporary accounts in publications such as InfoWorld, PC magazine, Time, and the New York Times, as well as books such as Blue Magic by James Chposky and Ted Leonsis; Big Blues by Paul Carroll; and Fire in the Valley by Michael Swaine and Paul Frieberger.

‘Where’s my Apple?’

In 1980, no one senior at IBM wanted to build a microcomputer, as home computers were often called, except its CEO Frank Cary. But one man’s will was far from enough to get things done at IBM, even if he was the CEO.

In the late seventies, IBM was vast. Known as “Big Blue” to its friends and enemies, the company had almost 350,000 employees, ran so many branches, and operated in so many markets that one commentator at the time described IBM as a country, not a company. It made mainframe and minicomputers that filled rooms—sometimes, whole buildings—and cost vast sums of money. It made business electronics, dominated the global market for typewriters, and made millions selling a wide variety of office supplies. A former employee would later describe it even more accurately: IBM didn’t stand for “International Business Machines.” It stood for “International Business Mafia.”

IBM’s huge divisions, spread across the U.S. and the rest of the world, operated like a series of global mafia families. They cooperated when under threat from the outside, and occasionally fought. The heads of those corporate families would regularly come together at the company headquarters in Armonk, New York to form the Management Committee. There, under the watchful eye of IBM’s CEO, they made decisions that would shape IBM’s business empire globally.

Cary started at IBM in 1948. He’d been its CEO since 1972. He had helped it dominate the world of mainframe computing, selling machines that could cost tens—or even hundreds—of thousands of dollars. But he also didn’t believe that the company should rest on its laurels. The Apple II personal computer had been selling in enormous numbers since 1977. At these meetings, he had one question for IBM’s divisional capos.

“Where’s my Apple?”

His divisional heads always had the same answer. Microcomputers—home computing—were a fad. They were low-cost and low-profit. Let others scrabble around in the metaphorical dirt of home computing. The real money was in the markets that IBM’s divisions already dominated—selling vast mainframes and minicomputer systems to large businesses. Cary was even told to buy Atari, which by then had established itself as America’s home video game system of choice. That’s all home computers were good for: gaming.

Frank Cary in 1972. Credit: IBM.

But Cary didn’t back down. In July 1980, he finally forced a concession out of John Rogers, the head of IBM’s General Products Division (GPD). Rogers agreed to set up a project to create a prototype IBM microcomputer—as long as it didn’t come out of his budget.

The answer lay in Boca Raton, Florida. There, lurking near-forgotten in his vast empire of typewriters, photocopiers and other low-level business electronics, was the Entry Level Systems (ELS) Facility, headed up by one of Rogers’s lieutenants, Bill Lowe. It was a leaky, rundown concrete facility that existed primarily to research and prototype basic electronics. Lowe, who was one of the few microcomputer supporters at IBM, was summoned before the Management Committee in July 1980. He was assigned to “Project Chess” and told to create a prototype of a microcomputer. Lowe was given one month to do so, after suggesting to the Committee that one could be assembled from off-the-shelf parts. And he would report directly to Cary. After all, Cary was footing the bill, not Rogers.

To pull off this minor miracle of engineering, Lowe assembled a core team known as the dirty dozen (there were 13 of them in total). They were a mix of engineers and planners taken from a number of different divisions. Mostly, they were individuals who struggled to fit in with the risk-averse, rules-heavy way of working within IBM. Bill Snydes, for example, was a talented engineer who had taken the initiative on a failing project and spun off parts of it into a success. His reward for doing so was a reprimand for going outside of IBM’s standard operating practices. Other misfits, like systems engineer Lew Eggebrecht and software engineer Jack Sam, joined for similar reasons—a chance to work on a project with few limitations.

The project seemed doomed to fail. Mainframe computers were very different from microcomputers, and nobody at IBM had any real experience with the latter. Even if suitable off-the-shelf parts could be found, there was no guarantee those parts would work well together. They’d also need to program their prototype to do something that proved it would work. The risk that problems would occur was high, and they’d been given little margin to fix those that did. But it would be a fun way to spend a month. And as most of them were out of favor within IBM, they had little to lose.

To their own surprise, after a frantic month of development, this team pulled it off. They created a machine that worked. The computer didn’t do much, but it was enough to show the Management Committee.

"The system would do two things. It would draw an absolutely beautiful picture of a nude lady, and it would show a picture of a rocket ship blasting off the screen,” Bill Snydes confessed later. “We decided to show the Management Committee the rocket ship."

This demo, held in August 1980, convinced Cary and the Committee that the work should continue—for now. They told ELS to turn its prototype into a product—and to do everything else needed to take it to market.

Lowe was under no illusions about the difficulty of this task. He was convinced that the project was valuable, but Cary still seemed to be the only senior figure within IBM who agreed. The various division heads wanted nothing to do with the project. But as long as the cost (and any eventual blame) sat squarely on the ELS facility in Boca Raton, they didn’t feel a need to kill it. What the Committee did, instead, was impose a deadline for launch: August 1981. It was another impossibly tight deadline, just one year away.

Lowe headed back from the headquarters in Armonk, New York to Boca Raton with Snydes and Eggebrecht. There, the dirty dozen began the necessary work to turn their prototype into a real machine. But Lowe, who was seen as a rising star within the company, was already being considered for a new, bigger role elsewhere in IBM. Soon his promotion to that position was confirmed.

Lowe believed in the project but didn’t believe he was the right person to do it. Taking it forward would require a unique set of leadership skills. It had to be led by someone who thrived on breaking the rules and who could use the ELS’s newfound position outside the regular financial and management structure to its maximum. Luckily, Lowe knew just the man within IBM for the job—Philip “Don” Estridge.

A company man

Don Estridge turned 43 in August 1980. He had been an IBM man his entire professional life. He loved being able to tell people he worked there. He would tell people that his blood ran big, and it ran blue.

Don Estridge in 1983. Source: IBM.

Yet in many ways, Estridge ran as counter to the grain at IBM as the rest of the dirty dozen did. Tall with a mop of well-kept hair, Estridge wore the crisp suit, shirt, and tie that was the standard uniform for an IBM executive. But he coupled that look with a pair of expensive cowboy boots that often drew frowns from more senior executives. He was a humble man who had no time for flattery and a strong distaste for corporate politics. In IBM, this made him unusual. His fellow senior executives would frequently complain that he failed to return their phone calls, and rarely agreed to join golf sessions or long lunches. And his junior staff noted that he made a deliberate effort never to sit at the head of a table. Estridge believed in collegiality, not seniority.

Estridge was hugely charismatic, although his personal magnetism was drawn from a different source than that of Steve Jobs at Apple or Adam Osborne at Osborne Computing. They were men who sold their own visions. They could create “a reality distortion field” around themselves (a term coined by Apple’s Bud Tribble about Jobs). They could convince you of the value of their ideas and inspire you to give everything in their service.

Estridge was the polar opposite. He listened and supported. His role was to set objectives and provide people with the resources or political cover they needed. This approach inspired a different kind of loyalty from those who worked for him, but it was just as fierce. And it delivered results.

Lowe had seen how effective Estridge had been on the IBM Series/1 minicomputer project. The Series/1 was intended to be flexible enough to serve a variety of big business computing needs. It would be a workhorse that businesses would buy in large numbers, making IBM healthy profits on its five-figure price tag. But its development was plagued with problems and project overruns, and the end result was criticized for being underpowered. As a result, the Series/1 project was regarded as a failure within IBM.

However, one small part of it was successful. IBM’s sales department had secured a large contract to convert all of State Farm Insurance’s operations to run on Series/1 machines. That part of the project had been led by Don Estridge, who had managed the production of these specific machines, complete with required software. They had been delivered to State Farm on budget and on time.

When the Series/1 project ended, Estridge found himself tied to its wider failure—and out of favor. But IBM didn’t fire people who might still be useful. It simply sidelined them. They would be sent somewhere quiet and remote in case the company decided it needed them again. So Estridge was thanked for his work and sent into the Florida wilderness, to kick his heels in the corporate backwater of Boca Raton.

Estridge was a self-confessed computer nut. In his early IBM years, he had worked on radar systems for the military and software for NASA’s Apollo programme. Estridge was amazed by the Altair 8080, the world’s first home computer. He never believed that he might one day have a computer at home. And when Apple released the Apple II in 1977, Estridge bought one for himself. During the day, he would help IBM build mainframes and minicomputers. In the evenings, he would tinker with a computer of his own.

When Estridge heard about Project Chess, he wanted in. Lowe was happy to recommend him to the Management Committee as his replacement. In the fall of 1980, Estridge took over. He brought his Apple II with him, setting it up in his office at Boca Raton. If he felt that visitors weren’t seeing the value of microcomputers, they would be subject to an extended personal demonstration of what the Apple II could do until they conceded his point.

Going it alone

Nobody at the ELS would later remember how the machine they were building acquired the nickname “Acorn,” just that it was a sideways nod to “Apple.” Whatever the name, no one outside of the department seemed to care.

Lowe had warned Estridge to expect this. In early September, ELS approached the management of IBM’s new assembly plant in Boulder, Colorado. The new plant was significantly under capacity, and Lowe had offered them the chance to build the Acorn. He had received a polite but firm refusal. Boulder’s senior management would rather let their plant sit idle than associate themselves with Project Chess.

However, although IBM senior management clearly didn’t want anything to do with the Acorn, plenty of junior staff did. This would become a trend—senior management would first tell Lowe, then Estridge, to go away. Junior staff would be more interested. Those with useful skills would be lured down south to join Estridge’s ever-growing group of misfits in Entry Level Systems.

“We got some excellent people out of Boulder,” Bill Snydes confessed later.

Estridge and his senior team quickly realized that the only way they could get their home computer to market was if they did everything themselves. Whenever Estridge was told that yet another IBM department or division refused to cooperate, he would fly from Florida to IBM’s corporate headquarters in Armonk, New York. There, he’d petition Frank Cary for more money or people before returning home once again. It was difficult to secure additional resources at IBM—even if upper management believed in what you were doing. Yet, to the astonishment of his senior staff, Don Estridge was almost always successful. His charisma and quiet persuasiveness got him the concessions he needed. At their height, these trips were happening two or three times a week.

Finding a market

Estridge was lucky enough to inherit two things from Lowe: a viable design for a mass-market microcomputer and a core team of competent and enthusiastic engineers. It was Estridge, however, who realized that they should enter the small-to-medium business market, not home computing.

The design that Lowe’s team had developed was perfect for this. Nothing about their design was pioneering, but it provided enough power, at a low enough price, to run basic small business software, such as word processors and spreadsheets. Historically, IBM had ignored this market because these companies didn’t have the five- or six-figure budgets needed for mainframes or minicomputers.

However, they did have enough money to buy a single machine for $5,000 dollars or less. These same business users associated IBM with good things. As the saying went, no one ever got fired for buying an IBM.

However, to turn a profit, they would need to sell in volume, which meant making a lot of machines. Estridge’s options for manufacturing support within IBM were limited, so he took a trip to California to meet Lore Harp, CEO of Vector Graphic. Under Harp’s leadership, Vector had carved out a niche for itself making computers for the small business market. The company also had a reputation for quality that matched IBM’s. Estridge toyed with a partnership. He asked Harp to sign an NDA, and then revealed that IBM was considering a small move into this market itself. The discussion was friendly, but Harp was wary. Estridge left with five Vector microcomputers and a vague agreement that they should talk again.

While the offer to partner had been genuine, it would never take place. Instead, Harp would find herself rushing to save her business as Estridge filled out his senior team with individuals interested in going up to the challenge of the ELS alone.

A well-led team

Word spread quickly that Cary had given Estridge and the ELS freedom to operate outside of IBM’s byzantine structures and processes. As operations expanded rapidly in Boca Raton, it became clear that Estridge wasn’t afraid to use this authority.

Dan Wilkie, who came on board to lead production, experienced first hand what a difference this freedom made. With Estridge helping to cut through red tape around budgets, staffing, and processes, Wilkie created an assembly facility from scratch in about four months. The previous unofficial record for doing so at IBM stood at three years. He became one of Estridge’s most loyal and effective lieutenants.

“With the PC project, I saw the whole pie for the first time,” Wilkie said later. “We saw the costs, we solved the problems ourselves. We lived with the good and the bad. It’s no exaggeration to say that I made more decisions in the first 30 days with that group than I made during my first 14 years with IBM.”

Also critical was the recruitment of H.L. Sparks, known to his friends and colleagues as “Sparky.” A born salesman and a rising star within IBM, Sparks first encountered Estridge after the Series/1 project ended. Sparks could see that Estridge was a talented manager. When the Series/1 project had shown signs of failure, others had shifted the consequences elsewhere. Sparks was impressed that Estridge had done the opposite, shielding his team from blame, even though it meant accepting corporate exile to Boca Raton for himself. Then, in December 1980, Sparks received a call from Estridge, who offered him a job.

“A job doing what?” Sparks replied.

“Don’t ask. It’s a really great deal. Trust me and just say yes.”

Sparks didn’t hesitate, abandoning 18 years of climbing the ladder the IBM way to join the project. Estridge put him in charge of marketing and distribution. There was no point in having a machine if users didn’t know about it or have a way to buy it.

Sparks contacted the Office Products Division (OPD) within IBM, which was responsible for selling office items—such as typewriters and photocopiers—to businesses. He’d previously worked with an advertising executive in the division named Jim D’Arezzo and had been impressed with his work. Sparks lured D’Arezzo down to Boca Raton, where he was subjected to one of Estridge’s personal Apple II demonstrations. Sparks and Estridge ran D’Arezzo through the market potential they felt the Acorn could tap into.

“I saw right away that this thing was fantastic, and I knew that it was a barn-burner as one of the biggest items of all time in the IBM low-end product category,” D’Arezzo said later.

He returned to the OPD offices in New York where he met with Richard Young, the division’s president.

“Dick, we need this product,” D’Arezzo insisted.

“No, we don’t,” Young replied. Young didn’t like new things. And, as the capo of this particular IBM family, his word was final. Conversation over.

D’Arezzo was shocked. Sparks and Estridge weren’t. While Estridge provided senior management cover, Sparks did something heretical instead. He ignored the OPD and worked with mass-market stores directly, negotiating retail partnerships for the Acorn with Sears Roebuck and ComputerLand. Both jumped at the opportunity to sell a product from IBM.

The Tramp

Sparks suggested that D’Arezzo join them in Boca Raton. Angry and astonished that his division had rejected the Acorn, D’Arezzo agreed. In January 1981, Estridge welcomed D’Arezzo to Boca Raton as the advertising lead for the Acorn. He told D’Arezzo to take a few weeks and come back with a full marketing plan and advertising campaign, which he would bring to the Management Committee for sign-off in July.

“I was sitting there taking notes and so I asked what the product was going to be called when it was introduced. He said he didn’t know. I asked how the product was going to be packaged and he said, ‘Packaged?! We don’t even know what its size will be!’

“About this time,” D’Arezzo continued, “I started to get a bit apprehensive and I wondered if I should go home and pack my bags and head back north.”

D’Arezzo suggested they stop trying to persuade another division to handle their marketing, and work directly with an external advertising agency instead. This wasn’t how things were meant to be done within IBM, but Estridge again smoothed the way. Freed from IBM’s normally restrictive creative shackles, D’Arezzo and the agency came up with a campaign that featured Charlie Chaplin’s depiction of the Tramp in the 1915 film of the same name. In it, the Tramp is confronted by overwhelming giant machines. This was the IBM of old.

Afterward, the Tramp would be shown happily using the Acorn. This was the new IBM.

Filling the software gap

As Project Chess’s misfit reputation grew, other divisions started sending their more vocal problem children Estridge’s way. One such arrival was Joyce Wren.

Wren worked at one of IBM’s smaller offices in Silicon Valley. She was outspoken on the subjects of sourcing external development, and IBM’s slow pace of software development and management.

At one point, needing a program written in a hurry, Wren contracted a couple of programmers directly. She paid for them to stay in a hotel for eight weeks while they wrote it. Her reward was a slap on the wrist from the IBM head office in Armonk. The division that should have gotten the work had complained.

But Estridge liked Wren’s approach. He formed a new software commissioning arm within the ELS for Wren to lead. Software was critical to the Acorn project. The biggest requirements were a spreadsheet program and a word processor. These two tools, which were popular requests when buying a computer, delivered the most value for small businesses. The makers of VisiCalc agreed to port their software to IBM’s machine, solving the first problem. A distinctly un-IBM solution was required to deal with the second. They paid the creator of EasyWriter to port his word processor from the Apple II to the new IBM machine.

EasyWriter’s creator was John Draper, better known to the world as the infamous “phone phreaker” (and proto-hacker) Captain Crunch. Phone phreaking involved using tricks and tools to make free phone calls illegally. Draper achieved notoriety for being one of the best phreakers in the business. He had only just been released from prison for phreaking when he received the request to convert EasyWriter for Project Chess. He was amused to find himself contracted now to IBM, one of the most respectable (and uptight) companies in America. In Boca Raton, it was decided that it was probably best if Armonk never discovered who the writer of their flagship word processor was.

Making a deal with Microsoft

The final piece of the puzzle for Estridge’s machine was the operating system—something IBM didn’t have time to develop itself.

As with many key moments in computing history, the process that led to the adoption of MS-DOS as IBM’s main operating system is veiled in myth. The most popular telling pitches it as a straightforward battle between Bill Gates at Microsoft and Gary Kildall at software company Digital Research for the soul of the new machine—a battle Kildall lost by choosing to take a pleasure flight in his personal plane rather than meet with the men from IBM to discuss a deal.

The originator of that story is an unreliable narrator—Bill Gates himself. The reality is far more complex.

As early as September 1980, Gates was approached by Jack Sams, one of Lowe’s “dirty dozen” engineers, to provide a version of BASIC for IBM’s new machine. Sams and Gates were old school friends. Sams asked Gates for his opinion on what operating system they should use. Gates recommended that they talk to Digital Research, run by Gary Kildall and Dorothy McEwen Kildall.

Gates and the Kildalls had an informal agreement not to compete. Microsoft focused on programming languages like BASIC and other software, while Digital Research focused on operating systems, CP/M. Although Gary wasn’t present at the first meeting, there was no need for him to be there—Dorothy handled the business side of Digital Research.

CP/M was the leading operating system for business-focused microcomputers at the time. It was regarded as the market leader in this sector, and a lot of software had been written for it already. As a result, Digital Research was in a strong negotiating position. When IBM offered $250,000 for a fixed-cost universal license to use CP/M, the Kildalls refused. Adam Osborne, the charismatic founder of Osborne Computers, had persuaded Gary to give him a license like that only a few months prior for his new microcomputer, the Osborne 1. As sales exploded, the Kildalls were filled with regret. They watched as Osborne reaped the profits due to the fixed cost they’d agreed to up front and decided they would never make the same mistake again.

However, back in Boca Raton, the team was growing frustrated with the pace of negotiations with Digital Research. The Kildalls didn’t fully appreciate how tight deadlines were for the Acorn team. When Gates became aware of this disconnect, he sensed an opportunity. After discussing it with his leadership team at Microsoft—Steve Ballmer, Paul Allen and Kazuhiko Nishi,—the men decided it was time to unilaterally end their truce with Digital Research.

Gates revealed to Sams that he was aware of another operating system that could be ported to the IBM machine. Sams confirmed that IBM would be interested—but that the company had no desire to get involved in a series of backroom deals or betrayals. If Microsoft did offer them a viable alternative to CP/M, however, they would consider the offer carefully. Emboldened, Gates negotiated the purchase of the QDOS operating system from fellow Seattle software developers at SCP and renamed it MS-DOS. It was promptly offered to IBM as an alternative option to CP/M.

There would be more twists and turns—and a number of legal cases—in the MS-DOS story, but this sequence of events would cement Microsoft as Estridge’s preferred partner during the Acorn’s development and lead to MS-DOS becoming one of the dominant operating systems of the era. Unlike the Kildalls, Gates was prepared to turn over Microsoft’s entire operation in service of the Acorn ahead of launch. And that was exactly what Estridge needed.

When the new machine was finally announced to the public it came with a choice of three operating systems: MS-DOS, CP/M, or p-System from the University of California. The choice was the result of a settlement between IBM and Digital Research, which was threatening legal action over the events that had led to the creation of MS-DOS and Microsoft’s relationship with IBM. But with MS-DOS priced at $40 in comparison to CP/M’s $240, it was clear to buyers which version of the machine Big Blue felt they should choose.

Family matters

By March 1981, the hardware for the Acorn was ready to go. By June, the software and operating system were finalized. In July, Estridge received permission to take the product to market from the Management Committee. They would officially launch it in August, right on time.

Estridge’s management style had delivered. He pushed his staff hard but was careful to walk the fine line between hard work and burnout.

Those who worked in Boca Raton during this period remember that Estridge frowned upon late nights in the office unless they were absolutely necessary. He was a devoted family man and had been married to his college girlfriend, Mary Ann. They had four daughters, one of whom they had adopted after her parents died in a car crash while she was young. He believed that work should not come at the expense of family time.

To Mary Ann’s amusement, he would sometimes drive back to the office late at night to check that everyone had gone home. If he found people staying too late, he would demand to know why. If they needed more resources, he would try and find a way to provide them. New transfers to Boca Raton discovered that leisure time was for everyone, not just executives.

However, as pressure to meet their launch date increased, Estridge did ask more of his senior team. Wilkie and Sparks recall informal Saturday morning conclaves in the office. It had become impossible to find uninterrupted moments to debate and address problems during the week. These unofficial get-togethers were Estridge’s answer. Attendance was not compulsory for anyone—music recitals and Little League games had priority. Nor did they have set durations. What they offered was a chance for the senior team to get together, take a breath, and work through crises.

It was perhaps at one of those sessions that they finally decided on a name. Acorn had always been a placeholder. Internally, its official designation was the IBM 5150. At some point people just started calling it the Personal Computer, “PC” for short. It was a name that seemed like a natural fit.

On August 12, 1981, IBM revealed the IBM 5150 to the press and the world. At a grand event in a New York City hotel, journalists were given the opportunity to test it out while Estridge and others stood by, available for questions. Estridge confirmed it would go on sale at Sears and ComputerLand stores from August 15 onward.

Source: Byte magazine.

The coverage the next day was overwhelmingly positive. The machine was quick, reliable, and well built. The software was good and the price point didn’t break the bank. The press had already fallen in love with the PC, a viable contender to the Apple II for small businesses.

“It appears that IBM has a better understanding of why the Apple II is successful than Apple,” the New York Times wrote on the day after its announcement.

Changing the world

The launch of the IBM PC changed computing. Prices started at about $1,500 for a basic model running MS-DOS (about $5,000 today). Most people opted for extras, such as a monitor or additional disk drives, that brought the price up to around $3,000. Still, the price was easily in the range of a similarly fitted-out Apple II and cheaper than a Vector. It was well within the price range of most small businesses. And IBM stood for quality and reliability. It was less risky than buying an Apple.

“They didn’t know what to buy until they heard about the IBM PC,” Martin Alpert, the founder of computer hardware manufacturer Tecmar, said, “And then they figured they couldn’t go wrong because it was an IBM.”

In July 1981, when the PC was announced, Alpert sensed which way the market was likely to go. By the time it had launched, he had already started the process of pivoting Tecmar to focus entirely on the PC. It was a smart move, one made possible by a design choice led by Estridge. Estridge realized that whoever set the new computing standard would be able to dominate the market for some time to come. The easiest way to set that standard wasn’t just to sell machines; it was to let other companies sell parts, software, and even whole computers that would be compatible with your machine. Unlike all of its major rivals—including the Apple II—the IBM PC was built with an open architecture.

The PC swept all before it. Within months, demand exceeded production capacity at Boca Raton by roughly 800 percent.

“We can only handle so many factors of two,” Estridge told Byte magazine shortly after launch, when asked how work was going on increasing supply. They’d had a gut feeling that their machine would sell well, but not this well.

By the end of 1981, they were shipping 13,000 machines a month out of Boca Raton. It still wasn’t enough to meet demand. By the end of 1983, they had sold 750,000 machines and controlled an estimated 70 percent of the entire home and business computing markets. If IBM had spun the ELS off into a separate business at that time, they would have been the third-largest computer manufacturer on the planet.

“I think we’re in an era where the public has adopted computing the same way it adopted the automobile.” Estridge told Byte in 1983.

That same year, Time didn’t declare a “Man of the Year.” The magazine declared it a “Machine of the Year.” The PC had conquered the world.

The original Personal Computer. Source: IBM.

The Woodstock of computing

There are few things in business that can change minds faster than profit. So it was within IBM. In less than two months, the ELS went from the ignored and unwanted child of IBM’s computing arm to its favored son.

It was a vindication for Estridge and his team. While Estridge’s management had limited the impact of burnout, getting the PC to market had taken long hours and weekend work. Many—Estridge included—had believed that they’d be able to mentally pause post-launch.

Instead, they found themselves under even more pressure than before—to increase production as fast as possible and provide direct consumer support in a way that nobody at IBM had ever needed to do before.

Securing the resources to achieve these things took even more trips to Armonk. These were just as successful as before, not least because Estridge was no longer dealing with a Management Committee that viewed the IBM PC with suspicion. But more resources also meant more paperwork and more management responsibility, both for Estridge and his senior staff and engineers. People like Snydes or Wilkie now found themselves having to rely on others to solve technical problems as their teams—and responsibilities—grew. They could no longer dive straight in and fix things themselves.

It also wasn’t long before Boca Raton was regarded as the place you needed to work if you were an ambitious IBM executive. Suddenly, everyone was a PC evangelist, including many who the original PC team knew had been actively hostile to their efforts before. One member of the original “dirty dozen” later described Boca Raton as IBM’s Woodstock: Everyone would claim to have been there when it started, and the vast majority were lying.

The betrayal

As operations at Boca Raton grew exponentially, it became harder and harder for the original ELS staff to maintain their sense of identity. For many, the joy of working there had come from their status as ignored and unloved misfits. Those days seemed to be over. Boca Raton was changing, and not for the better. They felt like they were being reabsorbed and forgotten. The pressure to reintegrate with IBM’s traditional way of operating began to build. It didn’t help that IBM’s “salarymen” culture didn’t enable financial rewards for exceptional achievements.

The ELS had invented an entire industry—the PC business, one that IBM and others were now making a lot of money from. Very little of that money filtered back to the men and women of ELS. Not because IBM didn’t recognize what they’d achieved, but because the company had no way of providing major financial bonuses, at least not ones that weren’t tied to promotion into management.

Wilkie, for example, couldn’t help but comment on the difference on a visit to Tandon, one of the companies that supplied IBM with disk drives. “When I passed the executive parking lot, all I saw were Rolls-Royces, Mercedes, and Porsches,” he said. “I did consider that, after all, I was in Southern California, but I also remember thinking, ‘I am definitely in the wrong end of this business.’”

Estridge struggled to find ways to financially reward the core members of the PC team without promoting them into management positions—positions many of them didn’t want, because they took them further away from the hands-on engineering they loved. While nobody blamed Estridge for the lack of an alternative reward scheme, its absence bred resentment.

And then, in September 1982, the resentment imploded.

A phone call arrived, from Martin Alpert, the Tecmar CEO. A group of IBM staff members had contacted Tecmar, offering to sell the company commercially sensitive designs and specifications for the PC. Alpert was an honorable, intelligent man. As a major peripheral manufacturer, his relationship with IBM was the cornerstone of his business. He informed IBM of their potential security breach and then went a step further—he offered to assist them in a sting operation.

On September 4, 1982, Alpert met with the rogue IBM employees in a hotel at Cleveland Airport. He listened to their pitch. He told them he was interested and they agreed to talk. He was also wearing a wire.

A week later, IBM fired William Erdman and Peter Stearns, two newer members of the IBM PC operation for attempting to sell corporate secrets—along with Lew Eggebrecht, one of the original dirty dozen members and forefathers of the PC.

The discovery of Eggebrecht’s betrayal sent shock waves through the original members of the PC project. He had been there from the beginning. He was one of them. They believed that his actions were naive rather than malicious. But it brought all the feelings that something had been lost in Boca Raton back to the fore.

It wasn’t even about money. They just wanted someone—something—to recognize that what they had achieved as a group had been special.

One day shortly after the incident, Wilkie turned up to a meeting with Estridge and noticed that he was wearing a red rosette on his jacket lapel.

“What’s that for?” Wilkie asked.

Wordlessly, Estridge opened his briefcase and pulled out another red rosette. He handed it to Wilkie.

“What’s this mean?” Wilkie asked again, confused.

“It means you’re a member of the finest, most professional and most loyal team that’s ever been assembled in the history of IBM,” Estridge told him. “Put the rosette in your lapel and wear it everywhere you go, so people will know who you are and what you belong to. Besides, we all need to stop and smell the roses from time to time.”

Wilkie, a man not prone to displays of emotion, began to well up. It was a small, silly thing, but Estridge had somehow managed to distill all of their frustrations into something tangible.

“Now state your business, then get out of here,” Estridge told him. “I’ve got more of these things to hand out.”

From that point onward, you could always spot a member of the original PC team at Boca Raton. They were the ones proudly wearing one of Estridge’s red rosettes.

The final act

The more successful the PC became, the more IBM’s other divisions—and its more ambitious executives —wanted in. Having finally realized the potential profits to be made in the microcomputer industry, IBM’s most senior management were pushing the ELS for more. Estridge pulled together his weary senior team. They agreed that one of three projects looked promising—a cheaper, cut-back version of the PC (the PC Junior); a high-end version of the existing PC, complete with a hard disk (the PC/XT); or an improved version using newer Intel chips (the PC AT).

With everyone at Boca Raton already stretched thin, Dan Wilkie asked the question nobody else wanted to ask.

“My god, what happens if the Management Committee approves all three programs?”

There was a long, drawn-out silence. Eventually, Estridge spoke. “That’s something we’ll worry about if and when we get that order.”

As Wilkie had feared, the order was given.

The decision to push all three post-PC projects proved to be the final nail in the coffin for what Estridge had built at Boca Raton. They had become too successful.

"You couldn't have a business that might account for 25 percent of your revenue not to be tied into the formal system," a member of the Management Committee from this time later explained.

Estridge had fought to maintain the independence of the ELS and keep the rest of IBM away. Now, even his supporters appeared to be working against him. The more resources they gave him, the more he was pressured to take on more standard IBM ways of operation to deal with the size of the operation.  

Every promotion inserted additional levels of management and bureaucracy between him and the factory floor. His swift decision-making, approachable nature, and ability to see the full picture had always been his strengths. These became significantly harder to wield once layers of secretaries and junior managers asserted control over his schedule.  

All three of the PC successor projects needed Estridge’s managerial hands-on touch. But there was only one Estridge.

On August 1, 1983, IBM tacitly confirmed to the company that the misfits of Boca Raton were misfits no more. They were so important and successful that the Management Committee had upgraded Entry Level Systems at Boca Raton. The media and press were informed that ELS had been replaced by the Entry Systems Division in Boca Raton. Its president was Don Estridge.

They were a fully-fledged IBM family. That brought with it an implicit requirement to behave the IBM way. Like sharks circling, other divisions began to take back responsibility for activities that they had been happy to let the ELS do before. Distribution was returned to the IBM National Distribution Division. Marketing went to the IBM National Marketing Division. Each change was meant to bring stability; instead, it created complexity and established a more conservative and risk-averse way of operating. That brought delays, which would be blamed on the ESD and Estridge, miscast as problems related to any freedom the division retained.

In the end, the failure of the PC Junior in the home market triggered Estridge’s final fall. The project, led by Bill Sydnes, suffered a catalog of delays and failures—in part, because Estridge, still used to managing large projects directly rather than stepping back as a senior executive, could never quite relinquish control. Sydnes, for example, recognized that competing for the home computer market meant going up against Atari and Commodore, and facing off against sharp operators like Commodore’s ruthless founder, Jack Tramiel. In order to beat the master of bargain-basement computing, IBM would have to sink down to Tramiel’s level and sell machines in the kind of discount electronics chains that IBM had always shunned. Estridge, too distant from the minutiae of the project to understand this, vetoed Sydnes’s plans.

In its wake, Estridge was finally forced to agree to something he had been trying to avoid for almost a year: accept one final promotion. In early 1985, IBM announced that Don Estridge was now the Director of Worldwide Manufacturing, responsible for 116,000 people at 41 IBM plants in 15 countries worldwide.

When Estridge announced the news in the giant corporate auditorium at Boca Raton, the assembled members of the ESD gasped. Wilkie, like many of the old guard present, instinctively reached for the red rosette on his lapel. Some of them began to cry. So did Estridge. On paper, it was a promotion. However, it was a role with no real power. They all knew what it meant. Their leader had been quietly benched, just as he had been in 1979 on the Series/1.

The rise of the clones

What happened next wasn’t hard to predict. IBM failed to remember one of the key things that made Estridge so successful—his humility. The company assumed that once it controlled 70 percent of the PC market, nobody would be able to compete again. Employees labored away at projects like the PC Junior using the same generation of PC technology, rather than focusing on what was happening in the market around them—the rise of the clones.

Estridge’s decision to base the PC on an open architecture was a smart one, but it had been built on a bad assumption—that IBM would continue to move fast and innovate. The clones would always follow in its wake. But IBM took its eye off the ball. The development of a true PC successor, the PS/2, which used Intel’s next-generation processor—the more powerful 80286—proceeded at a snail’s pace.

Finally, in April 1987, IBM announced the release of the IBM PS/2, built around the “286” (as the 80286 had come to be known). When it launched, it was already a generation of processors behind. Six months earlier, Compaq, an aggressive PC clone that had lured many of Estridge’s former team away, released the Deskpro. It was the first mass-market machine to use Intel’s new 80386 processor—a “386.”

By the end of 1987 market share fell from 70 percent down to 20 percent. IBM finally realized its mistake. It was time to bring Don Estridge off the bench and push IBM to the forefront of the PC business once again.

Perhaps, in another timeline, that’s what happened. In this one, Don Estridge was no longer there to save the day.

The flight

At 7:30 p.m. on Friday, August 2, 1985, Dan Wilkie was enjoying a quiet evening at home when his phone rang. He was the site manager for Boca Raton and the designated contact in case of an emergency. He listened in horror to the voice at the other end of the phone. An hour earlier, Delta Flight 191 had crashed as it approached the runway in Dallas, Texas, and 137 passengers and crew were now dead.

After demand for the PC had soared, the ESD opened an additional manufacturing facility in Texas. The Dallas evening flight had become popular with IBM staff heading between the two sites. For the rest of the night, Wilkie stayed by the phone, coordinating with Delta to identify who from Boca Raton had been onboard. He spoke to worried relatives and promised to use the corporate weight of IBM to support them however they needed, logistically or financially.

At 11:30 p.m., he received another call. It was from the manager of an IBM sales office in Fort Lauderdale. Wilkie had been fielding calls from various IBM managers all evening, checking on missing staff. He asked the manager who he wanted to check on.

“No, It’s Patty Estridge, Don’s daughter,” the manager said, sounding distraught. “She just called me and said her folks were on the plane.”

Wilkie recoiled, as if he’d received a punch to the stomach. For a brief moment, he refused to believe it. Estridge had already left Entry Level Systems. He was preparing to move to New York. There was no reason for him to be on a flight between Texas and Boca Raton. But something deep in his soul told him it was true. He hung up the phone and placed his head in his hands.

Four hours later, Delta was finally able to confirm it. Don Estridge was dead.

The last goodbye

Don Estridge had always been a family man. After news of his benching by IBM got out, in 1985, Steve Jobs made him an unbelievable offer: $1 million to take over from Jobs and run Apple. Estridge refused. IBM was his family. So were his employees. Even when he’d been made the president of an entire IBM division, he’d continued to drive around the buildings at Boca Raton at night to check on his staff.

Most of all, he loved his real family—his wife Mary Ann and his four daughters. They were his life. So, before leaving Florida forever for his new role at IBM headquarters in Armonk he decided to take Mary Ann on a long-overdue camping holiday. They decided to travel via Dallas to meet up with their daughter who lived there.

"I think he went to his grave with the love affair still going on with his wife," Ed Faber, the CEO of ComputerLand, said. "I know that sounds corny, but you just had to see the two of them swooping around the dance floor to understand how in love they were."

The funeral service for Don and Mary Ann Estridge was held on August 5, 1985, attended by their friends and family. They were not alone. Over 1,000 people crowded the church and graveyard that day. They had come, en masse, from Boca Raton.

The crowd was a sea of black and gray, flecked with the occasional small flash of red from a lapel. Every one of the pall-bearers for Don’s casket also wore the red rosette.

Even after the service ended, the crowd seemed reluctant to leave. Leaving meant admitting that Don Estridge was gone. The man who had always stood behind them, championed them within the conservative culture of IBM, and helped change the world forever. He had given so much of himself to them. On some level, they knew they needed to give some part of themselves back to him. They didn’t want him facing the next world alone.

Eventually, it was Wilkie who made the first move. Overwhelmed with emotion, his eyes red and swollen with grief, he stepped forward and detached the red rosette from the lapel of his suit jacket. It was the same one Don had given him years before. Leaning down, he gently placed the rosette on the casket.

From behind him, he heard someone else moving through the crowd of mourners. Wilkie didn’t turn around. He knew what was coming. The mourner reached past Wilkie and another red rosette was placed gently, next to his own, on Don’s casket.

More movement. Another rosette. Then another.

Then, after a while, it was done.

Wilkie stepped back, not daring to make eye contact with the friends and colleagues who stood alongside him. He turned and slowly walked away. His colleagues did the same. The spell broken, the other mourners started to disperse, too.

Within minutes, all that remained was the silence of the graveyard. On the casket of Don Estridge, a collection of red rosettes cast pink shadows in the pale morning light.

Gareth Edwards is a digital strategist, writer, and historian. He has worked for startups and corporations in both the UK and U.S. He is an avid collector of old computers, rare books and interviews, and abandoned cats. Follow him on X, Mastodon, and BlueSky.

To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.

Source

Generative artificial intelligence (AI) has shown a remarkable ability to answer questions on structured tests, including achieving well above a passing score on the United States Medical Licensing Examination.

But in an unstructured setting, when the AI models are fed a stream of novel questions crafted by humans, the results can be terrible, the models often returning several inaccurate or outright false assertions, in the phenomenon known as 'hallucinations'.

Researchers at Heidelberg University Hospital in Heidelberg, Germany, reported in the prestigious New England Journal of Medicine (NEJM) this week that hooking generative AI models up to a database of relevant information vastly improved the model's ability to answer unstructured queries in the domain of oncology, the treatment of cancer.

The approach of retrieval-augmented generation (RAG), letting the large language models tap into external sources of information, dramatically improved the spontaneous question-answering, according to authors Dyke Ferber and the team at Heidelberg in a study they describe this week in NEJM, "GPT-4 for Information Retrieval and Comparison of Medical Oncology Guidelines." (A subscription to NEJM is required to read the full report.)

The study was prompted by the fact that medicine faces a unique information overload -- there are more recommendations for best practices being generated all the time by medicine's professional organizations. Staying current on those suggestions burdens physicians trying to handle a population that is living longer and expanding the demand for care.

Groups such as the American Society of Clinical Oncology (ASCO), Ferber and team related, "are releasing updated guidelines at an increasing rate," which requires physicians to "compare multiple documents to find the optimal treatments for their patients, an effort in clinical practice that is set to become more demanding and prevalent, especially with the anticipated global shortage of oncologists."

Ferber and team hypothesized that an AI assistant could help clinicians sort through that expanding literature.

Indeed, they found that GPT-4 can reach levels of accuracy with RAG sufficient to serve at least as a kind of first pass at summarizing relevant recommendations, thus lightening the administrative burden on doctors.

The authors tested OpenAI's GPT-4 by having human oncology experts submit 30 "clinically relevant questions" on pancreatic cancer, metastatic colorectal cancer, and hepatocellular carcinoma, and having the model produce a report in response with statements about recommended approaches for care.

The results were disastrous for GPT-4 on its own. When asked in the prompt to "provide detailed and truthful information" in response to the 30 questions, the model was in error 47% of the time, with 29 out of 163 statements being inaccurate, as reviewed by two trained clinicians with years of experience, and 41 statements being wrong.

"These results were markedly improved when document retrieval with RAG was applied," the authors reported. GPT-4 using RAG reached 84% accuracy in its statements, with 60 of 71, 62 of 75, and 62 of 72 correct responses to the three areas of cancer covered in the 30 questions.

"We showed that enhancing GPT-4 with RAG considerably improved the ability of GPT-4 to provide correct responses to queries in the medical context," wrote Ferber and team, "surpassing a standard approach when using GPT-4 without retrieval augmentation."

To compare native GPT-4 to GPT-4 with RAG, they used two prompting strategies. In its native, non-RAG form, GPT-4 was asked, "Based on what you have learned from medical oncology guidelines, provide detailed and truthful information in response to inquiries from a medical doctor," and then one of the questions about how to treat a particular instance of cancer.

GPT-4 in this native prompting was used both with what's called 'zero-shot' question answering, where only the prompt question is offered, then with few-shot prompting, where a document is inserted into the prompt, and the model is shown how the document can answer a similar question.

A RAG approach allows GPT-4 to tap into a database of clinical knowledge.
Heidelberg University Hospital

In the RAG approach, the prompt directs GPT-4 to retrieve "chunks" of relevant medical documents provided by ASCO and the European Society for Medical Oncology (ESMO) from a database. Then, the model must reply to a statement such as, "What do the documents say about first-line treatment in metastatic MSI tumors?"

The two human clinicians at Heidelberg University Hospital scored the responses for accuracy by manually comparing GPT-4's replies to the supplied documents.

"They systematically deconstructed each response into discrete statements based on the bullet points provided by GPT-4," wrote Ferber and team.

"Each statement was carefully evaluated according to its alignment with the respective information from the ASCO and ESMO documents," and, "for each question, the clinicians performed a detailed manual review of the guidelines corresponding to each query to define our ground truth."

That manual evaluation shows an important aspect of the RAG approach, Ferber and team noted: it can be checked. "By providing access to the retrieved guideline documents, the RAG mechanism facilitated accuracy verification, as clinicians could quickly look up the information in the document chunk," they wrote.

The conclusion is promising: "Our model can already serve as a prescreening tool for users such as oncologists with domain expertise," write Ferber and team.

There are limitations, however, to RAG. When GPT-4 used RAG to retrieve relevant passages that provided conflicting advice about treatment, the model sometimes replied with inaccurate suggestions.

"In cases in which GPT-4 must process information from conflicting statements (clinical trials, expert views, and committee recommendations), our current model was not sufficient to reliably produce accurate answers," write Ferber and team.

It turns out you still have to do some prompt engineering. Ferber and team were able to mitigate inaccuracies by asking GPT-4 to identify the conflicting opinions in the literature, and then provide a revised response, which turned out to be correct.

Source

What you need to know

• NVIDIA CEO Jensen Huang says robots will be the next wave of AI.
• Huang indicated that self-driving cars and humanoid robots will dominate this category.
• This happens as NVIDIA recently ranked as the third most valuable company in the world. 
• The CEO says humanoid robots are the easiest to adapt to the world "because we built the world for us."

NVIDIA CEO Jensen Huang has had some pretty interesting insights regarding the progression of AI, including the imminent death of coding as a career option for the next generation. 

While presenting at the Taiwan developer conference, Huang indicated that we might be on the verge of hitting the next wave with AI. The CEO's sentiments indicate we might be transitioning from AI-powered chatbots like Copilot and ChatGPT. Huang had humanoid robots on stage while making his presentation and stated physical AI is the next phase (via Business Insider )

"The next wave of AI is physical AI. AI that understands the laws of physics. AI that can work among us. Everything is going to be robotic. All of the factories will be robotic. The factories will orchestrate robots and those robots will be building products that are robotic."

As you might already know, huge tech corporations hopping onto the AI bandwagon heavily depend on NVIDIA's GPUs for AI advances. Microsoft and OpenAI have invested $100 billion in a new project dubbed Stargate to support their resource-hungry AI advances while simultaneously meeting their high demand for AI chips. 

NVIDIA's software and hardware are broadly used for sophisticated AI advances like the production of humanoid robots. As such, the chipmaker stands to benefit if physical AI and humanoid robots are the next phase for AI in development. 

Huang says self-driving cars and humanoid robots will be among the first and widely adopted robots as we venture into the physical AI phase. Self-driving cars aren't entirely a new phenomenon, especially with the rapid adoption of Tesla. Based on this premise, Huang's predictions could be partly true at the very least.

Huang's also backed up his humanoid robots reference by indicating:

Nvidia's market cap currently stands at $3.019 trillion, after its stock price went up by $5.16 percent to settle at $1,224.40 a share. Apple's market cap ended today at $2.99 trillion. Nvidia is still behind the biggest US publically traded company, Microsoft, which ended trading today at $3.15 trillion.

Nvidia's stock price has gone up by a whopping 24 percent since it revealed its latest quarterly financial numbers just a few weeks ago on May 22, when it revealed that it had brought in revenues of $26 billion, well ahead of the $7.19 billion in revenues it generated from the same quarter in 2023. The company also announced it predicts to bring in $28 billion in revenues in its current quarter which ends in late July.

Nvidia's massive stock market price rise is, of course, due to the demand of its AI GPUs which are powering the AI data centers for Microsoft's Copilot, OpenAI's ChatGPT, and others. At Computex 2024 this week, Nvidia revealed a new AI GPU architecture, which has the code name Rubin. That chip is expected to launch sometime in 2025.

Nvidia is still working on its next AI GPU chip, code named Blackwell. That chip will begin to ship to customers like Microsoft sometime later in 2024. Nvidia stated this week that it now expects to release new AI chips every year, instead of every couple of years.

While Nvidia is the king of AI chips, competitors like Intel and AMD both announced new AI-themed processors during Computex this week and Microsoft is reportedly developing its own in-house AI chips as well.

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Micron's GDDR7 memory chips use 1-beta DRAM technology to deliver up to 32Gbps speeds with over 1.5TB per second of system bandwidth in a 384-bit bus configuration. The company claims that is about 60% more bandwidth than the existing GDDR6 memory solutions. Those improvements should result in faster response times, "smoother gameplay," and quicker processing of high amounts of data.

Energy efficiency is another aspect where GDDR7 outperforms previous-gen memory. Technological advancements have allowed for the improvement of energy efficiency by 50% over GDDR6 while reducing thermals for better battery life in mobile devices. These chips also feature a new sleep mode, which consumes 70% less energy.

In the official announcement post, Micron said it expects GDDR7 memory chips to provide up to 30% improvements in FPS for ray tracking and rasterization over the current GDDR6 and GDDR6X-based graphics cards. As for AI, which is now everywhere, better-performing GDDR7 memory will increase throughput by up to 33% and lower response times by up to 20%.

Micron is once again at the forefront of memory innovation, developing the highest bandwidth solutions available, built with advanced process and interface technology to enable continued graphics performance leadership. The best-in-class capabilities of Micron GDDR7 memory help bring new levels of realism and performance to the most demanding applications.

Micron's GDDR7 memory chips will be available to partners in the second half of 2024. AMD will utilize the next-generation VRAM from Micron for its products, and other partners, such as PNY, are already using GDDR7 samples to test and validate their upcoming graphics cards and other solutions.

You can check out our entire Computex 2024 coverage here.

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

What you need to know

• GOG, CD Projekt's DRM-free PC gaming distribution service, is implementing a new cloud storage policy after August 31, 2024.
• The new policy will add a hard 200 MB per game cloud save storage limit, with CD Projekt planning to delete cloud saves (in order from oldest to newest) that exceed this limit until the remaining files fit within it.
• Players can manage cloud saves on the GOG website after logging in; additionally, within the GOG Galaxy software, users can also back up their cloud saves and save them locally on their PC.
• By backing their cloud saves up to local storage, players can keep over 200 MB of saves after the new policy goes into effect.

CD Projekt's digital game distribution platform GOG is one of the best PC gaming services available, and is generally considered to be an excellent DRM-free alternative to other popular platforms like Steam and the Epic Games Store. If you use it, though, you should be aware of a policy change regarding its online cloud saves — these give you a fallback if local saves get corrupted or you're planning to play on a PC other than your main rig — that's slated to go into effect later this year.

While GOG currently allows you to keep over 200 MB of cloud saves per game stored on its service, that will no longer be the case after August 31, 2024. In a recent blog post, CD Projekt announced its intent to automatically delete any cloud saves that exceed GOG's default 200MB allocation limit after this date, with the company explaining that this hard limit is being implemented to reduce storage costs while still giving players plenty of save storage per game.

"As the size and number of games increase, so does the demand for Cloud Storage. These limits ensure that all players have access to sufficient and manageable space for their game progress, and that we keep the associated costs under control," reads the publisher's post. "By optimizing our storage allocation, we aim to continue providing a reliable and user-friendly platform for everyone."

My Fallout 3 cloud saves currently exceed 200 MB, so I guess it's time to get rid of some old ones.

(Image credit: Bethesda Softworks LLC)

Notably, CD Projekt says that the first files it deletes from game save storage that exceeds 200 MB will be "unnecessary files," or files that managed to get saved in these folders despite not being related to your actual game saves. If a game's cloud storage is still over 200 MB after this, game saves will then be deleted in order from oldest to newest until the remaining files fit within the storage limit.

You can review and manage your cloud saves on the GOG website here if you'd like to get the save storage for each of the games you play under 200 MB before August 31. If you want to back up your cloud saves locally to your PC, you can do so within the GOG Galaxy launcher software by selecting a game, selecting "Extras," and then clicking the download arrow in the "Cloud Saves Backup" section (GOG has an official guide for this). Note that backed up cloud saves saved locally won't be affected by GOG's auto-deletion, so if you have over 200 MB of saves for a game you want to keep, make sure you do this.

In order to ensure that GOG users aren't caught by surprise when the new policy goes into effect after August 31, CD Projekt says it will regularly notify any players with game save folders over 200 MB about the hard cap "until all your Cloud Save files are within the allocated limits." Keep an eye out for these notifications, as even if you don't have any cloud save folders that exceed 200 MB now, you might push one beyond that between now and the end of August while playing the best PC games you're into right now.

The Samsung 990 Pro, which is considered to be one of the best SSDs for gaming in 2024. Along with full

game installations, you could store save files on a drive like this. (Image credit: Chuong Nguyen)

While it's unfortunate that you won't be able to keep more than 200 MB of cloud saves per game on GOG's storage service after August, CD Projekt's position here is very reasonable, and it's nice that it's given players a lengthy three-month warning. That said, it's still bad news for PC gamers with tons of cloud saves, but limited amounts of local storage available.

In 2012, when Microsoft unveiled the original Surface and Surface Pro, the company had the simple task of managing firmware for just two devices. Each computer had its custom firmware tailored to its specific needs. However, Microsoft later expanded the lineup with more devices, form factors, and device categories, each with its features and quirks. Addressing issues and implementing new features became a tedious task.

The Surface team decided to tackle the challenge with a new approach of using a shared firmware architecture with a common core for all Surface devices and device-specific extensions. That allowed the creation, for example, of a single fix that would apply to multiple models across the lineup. Quick and efficient, this method standardized firmware updates when it was implemented nine years ago.

However, new problems appeared as Microsoft was expanding its Surface portfolio. Hardware scalability and better flexibility, plus the need for a more efficient way to deliver reliable updates, required Microsoft to come up with a new approach.

As a result, developers implemented "a robust firmware architecture," which is now in use in almost every Surface device. It features improved interoperability across platforms and chip makers, robust automation, code reusability, and a consistent user experience across different form factors, such as tablets, laptops, docks, etc.

Despite the improvements, Microsoft says its Surface firmware journey is far from over, and the company is constantly looking for new ways to improve its device ecosystem and firmware platform. Those include things like better sensor integration, RUST-based security measures, and various convenient features.

Surface firmware updates we often cover here at Neowin may not look as exciting as Windows feature updates. Still, behind those updates, a team of engineers is working tirelessly to deliver a great user experience.

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

A tool that functions like a Google for DNA has demonstrated its promise for making all of the world’s biological sequence data cheaply and easily searchable, according to the Swiss team that developed it. In a proof of principle study, the researchers say they successfully indexed 10% of the world’s known DNA, RNA, and protein sequences—and the same method could be used to do the rest.

The advance, posted last month on bioRxiv, used a computational tool the group recently developed called MetaGraph to organize and compress publicly available sequence data into a searchable format—much as internet search engines do for web pages and their content. The resulting indexes, available for download and via a web portal, allow users to scan sequences comprising trillions of base pairs and billions of amino acids.

The research “represents a massive achievement and a landmark in our ongoing pursuit of the grand challenge of indexing all publicly available sequencing data,” says Rob Patro, a computational biologist at the University of Maryland who wasn’t involved in the pilot effort.

Such a resource could aid myriad areas of research, from identifying novel viruses to revealing disease-associated RNA sequences. Although MetaGraph isn’t the only project aiming for this goal, the team has created some of the largest indexes so far and calculates that its tool will be relatively inexpensive to use.

The need is pressing, Patro and others note. Repositories storing DNA, RNA, and protein sequence data are expanding exponentially. The Sequence Read Archive (SRA), a genetic database run by the National Institutes of Health’s National Center for Biotechnology Information (NCBI) and collaborators, already contains more than 50 thousand trillion base pairs (50 petabases) from organisms including humans and other animals, plants, and bacteria.

Current bioinformatics tools can’t scan this much data all at once, especially for sequences that haven’t yet been assembled into genomes. Researchers have to narrow down the sequence collections before they can search them. Several groups hope to solve this problem by compressing sequences from larger databases into a more organized data structure, or index, designed for easy searching in downloadable files or online portals.

In 2020, bioinformatician André Kahles, computer scientist Gunnar Rätsch, both at ETH Zürich, and their colleagues presented an early version of MetaGraph. The team used its tool, in which mathematical structures known as de Bruijn graphs represent overlaps between sequences, to index more than 1 million records from the SRA, totaling about 3 petabases. They have already employed MetaGraph in projects including identifying the microbial makeup of different cities.

Now, the team has an improved version of MetaGraph, and has harnessed it to index 5 petabases from the SRA and other databases, comprising sequences from microbes, fungi, plants, humans, and the human gut microbiome. Some indexes in the new paper reduce tens of terabases of data into about 10 gigabytes—small enough to work with on a personal computer. Although building the initial indexes is
expensive—hundreds of thousands of dollars for all the SRA, the researchers say—users can query the data sets much more cheaply than with existing techniques.

The work is “hugely exciting,” says Lesley Hoyles, a bioinformatician and microbiologist at Nottingham Trent University. With data repositories ballooning in size, “anything that can reduce the compute storage and energy costs … is a massive plus for researchers worldwide.” Such approaches could lessen barriers to genomic research for scientists in low- and middle-income countries, she adds. “Work could easily be done on cheap laptops.”

Other groups are also making progress. Last year, the Pasteur Institute won €2 million from the European Research Council to launch its IndexThePlanet project to catalog all data in the SRA. And researchers at NCBI are working on their own indexing tool, called Pebblescout. “It’s a very, very active field at the moment,” says Zamin Iqbal, a computational biologist at the University of Bath who worked on AllTheBacteria, a project that assembled bacterial sequence data to make them more easily searchable.

Patro suggests that because of MetaGraph’s index sizes, it could be slower than other tools on some particularly large tasks, such as looking up millions of sequences from a sample simultaneously. It’s also not yet clear how best to update the indexes with new sequence data, he adds. There’s also the challenge of funding the project, as well as all the computational costs that accompany it. Indeed, whether the tool ends up being widely adopted will partly depend on “addressing the social and administrative questions of how such a substantial resource should be hosted, updated, and maintained,” Patro says, adding that it seems “infeasible (and unfair) to expect an individual research group” to take on this enormous task.

Kahles and Rätsch agree, saying they hope the work will inspire other groups, and larger organizations such as NCBI or the SRA, to pick up the project and help index the remaining 90% of sequence data for use by researchers. “We show them here: ‘It’s possible—please do it,’” Rätsch says.

Source

An artificial intelligence (AI) model developed by Microsoft can accurately forecast weather and air pollution for the whole world — and it does it in less than a minute.

The model, called Aurora, is one of a slew of AI weather-forecasting tools being developed by tech giants, including GraphCast from Google DeepMind in London and FourCastNet from Nvidia, based in Santa Clara, California. But Aurora’s ability to quickly predict air pollution globally is pioneering, say researchers.

“This, for me, is the first big step in a journey of atmospheric chemistry and machine learning,” says machine-learning researcher Matthew Chantry at the European Centre for Medium-Range Weather Forecasts (ECMWF) in Reading, UK.

Conventional weather forecasting uses mathematical models of physical processes in the atmosphere, land and sea. To predict air-pollution levels, researchers have previously used machine learning along with conventional mathematical models, says Chantry. Aurora seems to be the first entirely AI model to generate a global pollution forecast — which is a much more complex task than weather forecasting, says Chantry.

“That was the thing where I went: wow, that’s a really cool result,” he says. The benefit of AI models is that they often require less computational power to make predictions than do conventional models, says Chantry.

AI researcher Paris Perdikaris at Microsoft Research AI for Science in Amsterdam and his colleagues found that Aurora could in less than a minute predict the levels of six major air pollutants worldwide: carbon monoxide, nitrogen oxide, nitrogen dioxide, sulfur dioxide, ozone and particulate matter. Its predictions span five days. It can do it “at orders of magnitude smaller computational cost” than a conventional model used by the Copernicus Atmosphere Monitoring Service at the ECMWF, which predicts global air-pollution levels, the team wrote in a preprint1 published on arXiv on 20 May.

Aurora’s predictions were of a similar quality to those of the conventional model. Policymakers use such predictions to track air pollution and protect against the related health harms. Air pollution has been linked to an increased risk of asthma, heart disease and dementia.

The researchers trained Aurora on more than a million hours of data from six weather and climate models. After training the model, the team tweaked it to predict pollution and weather globally. The model generates a ten-day global weather forecast alongside the air-pollution prediction.

The team says that, on some tasks, Aurora could outperform other AI weather-forecasting models, such as GraphCast — which can outperform conventional models and make global weather predictions in minutes. But it is too early to make a definitive comparison, says Chantry. “You’d have to spend a lot of time, and probably have access to the models themselves, to be able to really go into detail and say with some certainty that model A is better than model B,” he says.

Further research will reveal whether ‘foundational’ AI models trained on diverse data sets, such as Aurora, perform better than those trained on a single data set, such as GraphCast. “There’s lots of cool science to be done,” he says.

Source

Tokyo-based semiconductor startup Premo (a Latin word meaning "to be near") has unveiled what it claims is the world's first CPU prototype with wireless inter-chip connections. 

This new chip uses Premo's proprietary technology, Dualibus, which was developed in collaboration with the Irie and Kadomoto Laboratory at the Graduate School of Information Science and Technology, University of Tokyo.

Conventional semiconductor chips require physical boards and wiring to relay signals between chips, but Premo's Dualibus technology harnesses the principles of magnetic field coupling to allow the chips to communicate wirelessly.

Physical connection for power

Premo says its chip integrates a CPU, sensor, power supply, and communication module and uses chip-to-chip wireless connection technology and the company’s own CPU design to deliver a miniaturized device that reduces the need for printed circuit boards and wiring. 

The company, which was established in February 2020, envisions the chip's application in a number of industries, such as infrastructure, vehicles, consumer goods, animal husbandry, and IoT. It could even be installed in locations which aren't suitable for traditional, more bulky processors.

Dualibus-equipped chips could reduce pads on silicon wafers, making more efficient use of the semiconductor area, and resulting in smaller, flexible device shapes. Premo suggests, “By taking advantage of the close proximity and wireless connection between the transmitting coil in the thumb and the receiving coil in the knuckle, it can be used as a new user interface in AR/VR environments.”

Premo's breakthrough could potentially pave the way for more cost-effective chips that use fewer raw materials. The company is currently looking into creating post-packaged chiplets by arranging millimeter-sized chips.

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Note: this firmware update is not available for the regular consumer version of the Surface Pro 7, which has different hardware and firmware.

Here is the official changelog:

The following update is available for Surface Pro 7 Plus devices running Windows 10 October 2022 Update, Version 22H2, or greater.

 

Improvements and fixes:

 

• Addresses camera's performance issues caused by inadequate memory or storage capacity.

New drivers in this update include the following:

Here is extra information about the release:

Surface firmware updates are available from Windows Update and in the form of cumulative packages from the official Surface support website (does not apply to ARM devices). They are non-uninstallable, so back up important data before installing new firmware.

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Paul's Hardware (1.48M subscribers)

June 5, 2024

Video length: 36m 08s

0:00 Welcome to ASUS @ Computex 2024

0:53 ROG Thor III Power Supplies / Mjolnir 1200W UPS

6:08 Concept BTF Mini ITX Chassis Designs / RTX 4080 Super GPUs

8:45 AIO w/ Flexible (motorized) OLED Display, “Minimal” Tubing

10:35 Asetek/Fabric8 3D-printed Coldplates for next-gen ASUS Water Blocks

13:38 ROG Azoth Extreme Keyboard

18:14 ROG Harpe Ace Extreme Carbon Fiber Mouse

22:05 ROG Delta II Headset

25:09 ROG Rapture GT-BE19000 Wifi 7 Gaming Router

28:28 ASUS Prime AIO CPU cooler series

29:17 ProArt Monitors - 5K and 8K Panels

32:00 ProArt BTF Concept Chassis is SO CUTE

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Generative AI tools such as OpenAI’s ChatGPT and Microsoft’s Copilot are rapidly evolving, fueling concerns that the technology could open the door to multiple privacy and security issues, particularly in the workplace.

In May, privacy campaigners dubbed Microsoft’s new Recall tool a potential “privacy nightmare” due to its ability to take screenshots of your laptop every few seconds. The feature has caught the attention of UK regulator the Information Commissioner’s Office, which is asking Microsoft to reveal more about the safety of the product launching soon in its Copilot+ PCs.

Concerns are also mounting over OpenAI’s ChatGPT, which has demonstrated screenshotting abilities in its soon-to-launch macOS app that privacy experts say could result in the capture of sensitive data.

The US House of Representatives has banned the use of Microsoft’s Copilot among staff members after it was deemed by the Office of Cybersecurity to be a risk to users due to “the threat of leaking House data to non-House approved cloud services.”

Meanwhile, market analyst Gartner has cautioned that “using Copilot for Microsoft 365 exposes the risks of sensitive data and content exposure internally and externally.” And last month, Google was forced to make adjustments to its new search feature, AI Overviews, after screenshots of bizarre and misleading answers to queries went viral.

Overexposed

For those using generative AI at work, one of the biggest challenges is the risk of inadvertently exposing sensitive data. Most generative AI systems are “essentially big sponges,” says Camden Woollven, group head of AI at risk management firm GRC International Group. “They soak up huge amounts of information from the internet to train their language models.”

AI companies are “hungry for data to train their models,” and are “seemingly making it behaviorally attractive” to do so, says Steve Elcock, CEO and founder at software firm Elementsuite. This vast amount of data collection means there’s the potential for sensitive information to be put “into somebody else’s ecosystem,” says Jeff Watkins, chief product and technology officer at digital consultancy xDesign. “It could also later be extracted through clever prompting.”

At the same time, there’s the threat of AI systems themselves being targeted by hackers. “Theoretically, if an attacker managed to gain access to the large language model (LLM) that powers a company's AI tools, they could siphon off sensitive data, plant false or misleading outputs, or use the AI to spread malware,” says Woollven.

Consumer-grade AI tools can create obvious risks. However, an increasing number of potential issues are arising with “proprietary” AI offerings broadly deemed safe for work such as Microsoft Copilot, says Phil Robinson, principal consultant at security consultancy Prism Infosec.

“This could theoretically be used to look at sensitive data if access privileges have not been locked down. We could see employees asking to see pay scales, M&A activity, or documents containing credentials, which could then be leaked or sold.”

Another concern centers around AI tools that could be used to monitor staff, potentially infringing their privacy. Microsoft’s Recall feature states that “your snapshots are yours; they stay locally on your PC” and “you are always in control with privacy you can trust.”

Yet “it doesn’t seem very long before this technology could be used for monitoring employees,” says Elcock.

Self-Censorship

Generative AI does pose several potential risks, but there are steps businesses and individual employees can take to improve privacy and security. First, do not put confidential information into a prompt for a publicly available tool such as ChatGPT or Google’s Gemini, says Lisa Avvocato, vice president of marketing and community at data firm Sama.

When crafting a prompt, be generic to avoid sharing too much. “Ask, ‘Write a proposal template for budget expenditure,’ not ‘Here is my budget, write a proposal for expenditure on a sensitive project,’” she says. “Use AI as your first draft, then layer in the sensitive information you need to include.”

If you use it for research, avoid issues such as those seen with Google’s AI Overviews by validating what it provides, says Avvocato. “Ask it to provide references and links to its sources. If you ask AI to write code, you still need to review it, rather than assuming it’s good to go.”

Microsoft has itself stated that Copilot needs to be configured correctly and the “least privilege”—the concept that users should only have access to the information they need—should be applied. This is “a crucial point,” says Prism Infosec’s Robinson. “Organizations must lay the groundwork for these systems and not just trust the technology and assume everything will be OK.”

It’s also worth noting that ChatGPT uses the data you share to train its models, unless you turn it off in the settings or use the enterprise version.

List of Assurances

The firms integrating generative AI into their products say they’re doing everything they can to protect security and privacy. Microsoft is keen to outline security and privacy considerations in its Recall product and the ability to control the feature in Settings > Privacy & security > Recall & snapshots.

Google says generative AI in Workspace “does not change our foundational privacy protections for giving users choice and control over their data,” and stipulates that information is not used for advertising.

OpenAI reiterates how it maintains security and privacy in its products, while enterprise versions are available with extra controls. “We want our AI models to learn about the world, not private individuals—and we take steps to protect people’s data and privacy,” an OpenAI spokesperson tells WIRED.

OpenAI says it offers ways to control how data is used, including self-service tools to access, export, and delete personal information, as well as the ability to opt out of use of content to improve its models. ChatGPT Team, ChatGPT Enterprise, and its API are not trained on data or conversations, and its models don’t learn from usage by default, according to the company.

Either way, it looks like your AI coworker is here to stay. As these systems become more sophisticated and omnipresent in the workplace, the risks are only going to intensify, says Woollven. “We're already seeing the emergence of multimodal AI such as GPT-4o that can analyze and generate images, audio, and video. So now it's not just text-based data that companies need to worry about safeguarding.”

With this in mind, people—and businesses—need to get in the mindset of treating AI like any other third-party service, says Woollven. “Don't share anything you wouldn't want publicly broadcasted.”

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

AI has become a hot topic again these last few years since the launch of ChatGPT and rival services. Not looking to get left out of this boom, the Raspberry Pi foundation has launched the Raspberry Pi AI Kit in collaboration with Hailo for just $70 at Approved Resellers. The package includes a M.2 HAT+ pre-assembled with a Hailo-8L AI accelerator module atop a Raspberry Pi 5.

If you do decide to buy this piece of kit, you’ll be able to build complex AI vision applications capable of running in real-time with low latency and low power requirements. The Hailo-8L co-processor is able to run neural networks for object detection, semantic and instance segmentation, pose estimation, facial landmarking, and more while leaving the main CPU to process other tasks.

Commenting on the partnership, Hailo CEO and Co-Founder Orr Danon, said:

“We are thrilled to support Raspberry Pi and empower its vibrant community of professional engineers and creative makers with cutting-edge AI capabilities. Our partnership with the world’s leading single-board computer provider will inspire a new era of computing, enhanced by our high-performance AI processing capacity.”

To help users get started, Hailo has created a “model zoo” on GitHub where you can find a whole bunch of pre-trained neural network models ready to deploy, they’ve also been optimised to run on the AI Kit.

Other key features of the Raspberry Pi AI Kit include:

• 13 tera-operations per second (TOPS) of inferencing performance
• Single-lane PCIe 3.0 connection running at 8Gbps
• Full integration with the Raspberry Pi image software subsystem
• Compatibility with first-party or third-party cameras
• Efficient scheduling of the accelerator hardware: run multiple neural networks on a single camera, or single/multiple neural networks with two cameras concurrently

For those that do decide to pick up the Raspberry Pi AI Kit, a getting started guide has been created to help you set up and begin using AI software on your device quickly so you can move onto building your own projects.

Source: Raspberry Pi and Hailo

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

The Recall feature as it currently exists in Windows 11 24H2 preview builds.

Andrew Cunningham

 

Microsoft’s Windows 11 Copilot+ PCs come with quite a few new AI and machine learning-driven features, but the tentpole is Recall. Described by Microsoft as a comprehensive record of everything you do on your PC, the feature is pitched as a way to help users remember where they’ve been and to provide Windows extra contextual information that can help it better understand requests from and meet the needs of individual users.

 

This, as many users in infosec communities on social media immediately pointed out, sounds like a potential security nightmare. That’s doubly true because Microsoft says that by default, Recall’s screenshots take no pains to redact sensitive information, from usernames and passwords to health care information to NSFW site visits. By default, on a PC with 256GB of storage, Recall can store a couple dozen gigabytes of data across three months of PC usage, a huge amount of personal data.

 

The line between “potential security nightmare” and “actual security nightmare” is at least partly about the implementation, and Microsoft has been saying things that are at least superficially reassuring. Copilot+ PCs are required to have a fast neural processing unit (NPU) so that processing can be performed locally rather than sending data to the cloud; local snapshots are protected at rest by Windows’ disk encryption technologies, which are generally on by default if you’ve signed into a Microsoft account; neither Microsoft nor other users on the PC are supposed to be able to access any particular user’s Recall snapshots; and users can choose to exclude apps or (in most browsers) individual websites to exclude from Recall’s snapshots.

 

This all sounds good in theory, but some users are beginning to use Recall now that the Windows 11 24H2 update is available in preview form, and the actual implementation has serious problems.

“Fundamentally breaks the promise of security in Windows”

Security researcher Kevin Beaumont, first in a thread on Mastodon and later in a more detailed blog post, has written about some of the potential implementation issues after enabling Recall on an unsupported system (which is currently the only way to try Recall since Copilot+ PCs that officially support the feature won’t ship until later this month). We've also given this early version of Recall a try on a Windows Dev Kit 2023, which we've used for all our recent Windows-on-Arm testing, and we've independently verified Beaumont's claims about how easy it is to find and view raw Recall data once you have access to a user's PC.

 

To test Recall yourself, developer and Windows enthusiast Albacore has published a tool called AmperageKit that will enable it on Arm-based Windows PCs running Windows 11 24H2 build 26100.712 (the build currently available in the Windows Insider Release Preview channel). Other Windows 11 24H2 versions are missing the underlying code necessary to enable Recall.

 

• 

• 

The short version is this: In its current form, Recall takes screenshots and uses OCR to grab the information on your screen; it then writes the contents of windows plus records of different user interactions in a locally stored SQLite database to track your activity. Data is stored on a per-app basis, presumably to make it easier for Microsoft’s app-exclusion feature to work. Beaumont says “several days” of data amounted to a database around 90KB in size. In our usage, screenshots taken by Recall on a PC with a 2560×1440 screen come in at 500KB or 600KB apiece (Recall saves screenshots at your PC's native resolution, minus the taskbar area).

 

Recall works locally thanks to Azure AI code that runs on your device, and it works without Internet connectivity and without a Microsoft account. Data is encrypted at rest, sort of, at least insofar as your entire drive is generally encrypted when your PC is either signed into a Microsoft account or has Bitlocker turned on. But in its current form, Beaumont says Recall has “gaps you can drive a plane through” that make it trivially easy to grab and scan through a user’s Recall database if you either (1) have local access to the machine and can log into any account (not just the account of the user whose database you’re trying to see), or (2) are using a PC infected with some kind of info-stealer virus that can quickly transfer the SQLite database to another system.

 

Beaumont says admin access to the system isn’t required to read another user’s Recall database. Another user with an admin account can easily grab any other user’s Recall database and all the Recall screenshots by clicking through a simple UAC prompt. The SQLite database is stored in plain text, and data in transit isn’t encrypted, either, making it trivially easy to access both the stored database of past activity and to monitor new entries as Recall makes them. Screenshots are stored without a file extension, but they're regular old image files that can easily be opened and viewed in any web browser or image editor.

 

The other big problem is that because Recall is on by default and you have to manually exclude specific apps or websites from being scraped by it, the SQLite database will keep records of activities that are explicitly meant to be hidden or temporary. That includes viewing pages in Incognito mode in some browsers, emails or messages that you delete from your device, and files that you edit or delete.

 

Beaumont says he is holding off on publishing some details to “give [Microsoft] time to do something” about the feature as it is currently implemented. But he has pointed to efforts like this “TotalRecall” script as an example of how quickly and easily Recall data can be stolen and searched.

 

• 

• 

• 

There are mitigating factors here. Recall will begin by shipping on just a handful of new Windows 11 systems. It can be turned off entirely if you don’t want to use it, and the controls to disable Recall snapshots for certain apps or sites theoretically give users enough control that they can use Recall as intended without storing overly sensitive information in the database.

 

But given the sheer amount of data that Recall scrapes, the minimal safeguards Microsoft has put in place to protect that database once a malicious user has access to your PC, and the fact that many PC users never touch the default settings, the risks to user data seem far higher than the potential benefits of this feature.

 

Microsoft has struggled with security and privacy in its products. Not even a month ago, CEO Satya Nadella pledged to make security the most important thing at the company, following multiple high-profile data breaches and poorly handled information disclosures. Executive pay is being tied in part to security; rank and file employees are being told to “do security,” even when “faced with the tradeoff between security and another priority,” Nadella said. To launch Recall with such obviously exploitable security holes flies in the face of that directive.

Microsoft is hard to trust right now

Even if Recall were locked down better, another problem is that Windows 11 has eroded its users' trust and patience over time by endlessly pushing Microsoft's other products and services and refusing to respect user choices once they've been made. The company frequently finds new places to put ads; a "clean install" of the operating system comes with unasked-for third-party apps and ongoing notifications about other Microsoft services; the Bing Chat feature and then Copilot were rolled out quickly to the entire user base despite being "preview" products prone to problems.

 

None of this is directly related to Recall, but it demonstrates Microsoft's willingness to put revenue-squeezing ahead of the user experience, and that makes me inherently skeptical of Windows' AI features in general and Recall in particular. A huge searchable database of PC activity would be a holy grail for advertisers, and given how willing Microsoft has been to muck up Windows 11 in the last two and a half years, it's hard for me to trust that the company will stay committed to keeping the data collected by Recall completely private.

 

Recall isn’t finalized. It won’t be available on the vast majority of Windows PCs. Even once Copilot+ PCs from Intel and AMD hit the market, it will take at least a year or two for compatible NPUs to show up in midrange and low-end PCs. Those who truly hate it will be able to turn it off, though the list of “at least you can uninstall it/turn it off!” things in Windows 11 is getting frustratingly long.

 

We’ve also contacted Microsoft to ask about these concerns and whether the version of Recall that you can test on an Arm PC right now is the same one that end users will get on Copilot+ PCs; as of this writing, the company hasn’t responded.

 

But let's put it this way: Microsoft is building a feature into Windows that is monitoring and logging a ton of data about you and the way you use your PC. Traditionally, we’d call this “spyware.” The difference is that Microsoft is giving this particular data collection feature its blessing and advertising it as a banner feature of its upcoming wave of Copilot+ PCs.

 

The fact that the data is processed locally rather than in the cloud is a good first step, but it's also the bare minimum. Based on both the permissive default settings and the ease with which this data can be accessed, Recall’s security safeguards as they currently exist just aren't good enough.

 

If Microsoft really does intend for everyone at the company to “do security,” it needs to put these concerns ahead of its apparently all-consuming drive to insert generative AI features into every single one of its products. Improving Recall before it becomes generally available needs to take priority, even if it delays the launch.

 

Listing image by Jason Redmond/AFP via Getty Images

 

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts

Source

Hope you enjoyed this news post.

Thank you for appreciating my time and effort posting news every single day for many years.

2023: Over 5,800 news posts | 2024 (till end of May): Nearly 2,400 news posts