The vast majority of data used to train LLMs is scraped from the public internet. While this helps them to build knowledge and generate natural responses, it also puts them at risk from data poisoning attacks. It had been thought that as models grew, the risk was minimized because the percentage of poisoned data had to remain the same. In other words, it would need massive amounts of data to corrupt the largest models. But in this study, which is published on the arXiv preprint server, researchers showed that an attacker only needs a small number of poisoned documents to potentially wreak havoc.

To assess the ease of compromising large AI models, the researchers built several LLMs from scratch, ranging from small systems (600 million parameters) to very large (13 billion parameters). Each model was trained on vast amounts of clean public data, but the team inserted a fixed number of malicious files (100 to 500) into each one.

Next, the team tried to foil these attacks by changing how the bad files were organized or when they were introduced in the training. Then they repeated the attacks during each model's last training step, the fine-tuning phase.

What they found was that for an attack to be successful, size doesn't matter at all. As few as 250 malicious documents were enough to install a secret backdoor (a hidden trigger that makes the AI perform a harmful action) in every single model tested. This was even true on the largest models that had been trained on 20 times more clean data than the smallest ones. Adding huge amounts of clean data did not dilute the malware or stop an attack.

Build stronger defenses

Given that it doesn't take much for an attacker to compromise a model, the study authors are calling on the AI community and developers to take action sooner rather than later. They stress that the priorities should be making models safer, not just building them bigger.

"Our results suggest that injecting backdoors through data poisoning may be easier for large models than previously believed, as the number of poisons required does not scale up with model size—highlighting the need for more research on defenses to mitigate this risk in future models," commented the researchers in their paper.

Source

Just over two months since Electronic Arts (EA) revealed the game in late July and then blew everyone's socks off with a colossal Open Beta that broke a Call of Duty record in August, Battlefield 6 — the latest installment of the beloved, long-running multiplayer FPS — has finally come out on Xbox Series X|S, Windows PC, and PS5.

Between the return of the classic Battlefield class system, tons of new class-specific gadgets that expand the ways you can make an impact throughout your matches, the most advanced dynamic map destruction the series has ever had, and more, it's poised to rise as the dream game fans of the franchise have been yearning for.

But like many new AAA titles these days — seemingly nearly all of them, unfortunately — Battlefield 6's release has been mired by frustrating bugs and launch issues. Some of these are rather minor thankfully, but others are more serious, as they're impeding fans' ability to run and play the game.

EA and developer DICE are no doubt hard at work trying to resolve these as soon as possible, but in the meantime, you should familiarize yourself with the FPS' current issues and all known fixes and workarounds so you know what to look for and can deal with problems that arise. You'll find a full list of all of this below.

Battlefield 6 — Known bugs, launch issues, and fixes

Battlefield 6 "Game Not Released" error

Issue: Steam users have reported that they've been unable to download or update Battlefield 6 due to an erroneous "Game Not Released" error in their Steam installation queue.

Fix: Try fully exiting Steam (in the client, select Steam > Exit at the top left) and relaunching it to see if this fixes the problem. If not, try restarting your internet router, or restarting your PC. If the problem persists, you'll have to wait until a fix comes (or the bug potentially resolves itself).

Battlefield 6 won't launch

Issue: Some players on Windows PC have reported that they've been unable to launch the Battlefield 6 client, with it crashing to desktop shortly after being run initially.

It's been four years since we've had a big Battlefield release on our hands (and the last one didn't exactly pan out very well, to say the least), but that's changed today with the explosive launch of Battlefield 6 across Xbox, PC, and PS5. Going in, Electronic Arts' (EA) new FPS was expected to skyrocket to the top of player charts and become one of 2025's biggest games — and so far, it's living up to that expectation and then some.

Just hours after the game's arrival on Friday morning, Battlefield 6 soared to a peak concurrent player count of 747,440 on Steam alone, according to SteamDB data; that's just shy of a whopping 750K milestone, and it makes the shooter the 14th most-played game of all time on Steam, just behind Baldur's Gate 3 and ahead of Goose Goose Duck.

At the time of writing, it's slipped down to around ~600K concurrent players, but that number is expected to climb higher as the weekend starts and more fans can deploy to the battlefield once they're off work.

In fact, many are anticipating Battlefield 6 will shoot past the 1 million mark — a feat that would put it above FromSoftware's GOTY-winning 2022 ARPG Elden Ring and place it in the top 10 most-played Steam games list. That would be an incredible achievement, and quite symbolic of Battlefield's comeback as a series.

In response to the FPS' launch day success, Vince Zampella — head of Respawn Entertainment, Ripple Effect Studios, and the Battlefield franchise itself — took to social media to respond, thanking Battlefield 6's developers for their hard work as well as the Battlefield community for its continued and honest feedback.

"I need to take a moment to say congrats the incredible teams that brought BF6 to life. What an incredible journey we’ve been on the last few years," he wrote. "They poured themselves into this game. So amazing to show all their hard work. And they are just getting started, more to come."

"I also need to thank the Battlefield community. The feedback has been instrumental in helping us shape something we are all proud to show! Keep it coming, our journey together is just getting started," Zampella added, alluding to future content releases coming in Battlefield 6's post-launch seasons (Season 1 begins on October 28).

What’s happened? Google and Mercedes-Benz just gave us a first glimpse at what happens when Gemini leaves your phone and enters the car. Google and Mercedes took a CLA for a spin on Google’s campus, and interacted with Gemini through the car’s built-in systems, showing off seamless conversational commands and context awareness.

• The Gemini-powered MBUX assistant handled follow-ups naturally. When asked to find nearby cafés and whether they serve pastries, it responded in context.
• The system pulled data from Google Maps and business listings to answer nuanced queries (e.g., “Does this place have a wine list?”) and even offered to place a call.
• Mercedes says the CLA will be the first production car to ship with this Gemini tech, launching in the U.S. before the end of 2025.
• Though Gemini for cars has been discussed before, this is one of the clearest live demos showing how it differs from the existing Google Assistant in auto contexts.

Why this is important: This marks a shift from voice assistants to conversational AI agents in cars, which could make all the difference in redefining how we talk to vehicles. Gemini’s stronger context handling could make the car’s virtual assistant a more helpful, less frustrating part of the drive.

•  It moves us closer to cars that understand follow-ups and multi-step requests, rather than forcing one-off commands.
•  Supports Google’s broader strategy of phasing out Assistant in favor of Gemini across devices.
•  Gives Mercedes (and eventually other car makers) a path to making their in-car assistants smarter, without relying purely on command lists.

Why should I care? Well, for drivers, this could mean asking for things more naturally. One could say, “Find a coffee shop along my route that’s open now with desserts”, and they’d receive useful, context-aware answers. Gemini in-car could reduce friction in tasks like rerouting, exploring POIs, or managing calls, all hands-free.

Google / YouTube

•     Fewer “I’m sorry, I didn’t get that” responses, since the AI is built to keep context.
•     The car assistant may shift from a tool to a conversation partner, picking up on nuance.
•     Because it integrates with Google Maps and business data, it will likely give richer answers and suggestions.

Okay, so what’s next? We expect Mercedes to reveal official specs, deployment plans, and regional rollouts soon. Google’s blog says Gemini in cars will come via both Android Auto and Google-built cars later this year, opening the door for broader adoption across brands. If the end-user experience is stable, other OEMs could also adopt the same tech, so we’ll keep an eye on partnerships and performance tests.

Source

This discovery, published in Nature Materials, marks a breathtaking step toward the age of optoelectronics — where light, not electricity, drives the technology that powers our world. It’s a glimpse into a future where computers are faster, greener, and far more efficient than anything we’ve imagined before.

The Spark of Discovery

At the heart of this breakthrough is something both invisible and extraordinary: a two-dimensional electron gas. This gas forms naturally at the interface between certain materials, where electrons can move freely, almost like liquid. Traditionally, scientists could only create or manipulate such electron gases using electrical signals. But now, researchers have discovered a way to make this happen using nothing more than light.

By illuminating a carefully engineered material made of oxide layers, the team observed the spontaneous formation of this electron gas — and just as remarkably, it vanished when the light was turned off. It was like watching electricity bloom and fade at the flick of a switch, powered only by illumination.

This is more than a fascinating trick of physics. It’s a powerful demonstration that light can directly control electronic behavior in solid materials — a phenomenon sitting right at the intersection of optics and electronics.

When Light Replaces Electricity

What if your devices could operate with light as easily as they do with electricity? Light-controlled electronics could revolutionize the way we design and power digital systems. The implications are staggering.

Electronic components that respond to light — such as transistors — would operate much faster and more efficiently than today’s electrically driven ones. Imagine processors where light controls data flow instantaneously, without the resistance and heat losses of electrical current. Scientists estimate that light-based transistors could eliminate up to one-third of the electrical contacts on a computer chip — which translates to around a billion fewer electrical connections in a modern processor. The result? Cooler, faster, and far more energy-efficient computing.

In an era where every watt of energy counts, such efficiency could transform not just personal electronics but entire industries, from data centers to quantum devices.

The Science Behind the Glow

At first glance, this achievement may seem almost magical — light creating electricity-like behavior in solid materials. But behind the scenes lies meticulous scientific craftsmanship.

The researchers engineered materials with atomic-level precision. By layering oxides — materials that combine oxygen with metals — they created interfaces where electrons could behave in unique ways. These atomic boundaries were examined under high-resolution microscopes, allowing scientists to observe how atoms and electrons interacted under light.

Meanwhile, complex theoretical models were developed to describe how photons (particles of light) excite electrons in the material, setting them into motion to form the gas. This collaboration of experiment and theory revealed not just the “what,” but the “how” — an essential step toward harnessing the phenomenon for technology.

The discovery was a team effort, uniting experts from the Albert Fert Laboratory (CNRS/Thales), the Strasbourg Institute of Materials Physics and Chemistry (CNRS/Université de Strasbourg), and the Solid State Physics Laboratory (CNRS/Université Paris-Saclay). Their work represents one of the most exciting collaborations at the crossroads of photonics, electronics, and quantum materials.

The Promise of Light-Controlled Devices

The applications of this discovery stretch across the entire technological landscape. Imagine smartphones and computers that process information using light — devices that are smaller, faster, and vastly more energy-efficient than those we use today.

Beyond computing, this phenomenon could inspire ultra-sensitive optical detectors. In such devices, light acts as a booster: for the same electrical voltage, the electrical current becomes up to 100,000 times stronger than it would be in darkness. This opens the door to detectors capable of perceiving the faintest glimmers of light — tools that could revolutionize astronomy, medical imaging, and environmental sensing.

The implications extend into fields like spintronics, where the spin of electrons (a quantum property) is used to store and process data, and quantum computing, where light-controlled materials could make quantum bits more stable and efficient. The boundary between photonics and electronics is dissolving — and this discovery may well be the bridge that connects them completely.

The Beauty of Light and Matter

There’s something poetic about using light — the most ancient and fundamental form of energy — to shape the future of technology. Light has guided life on Earth for billions of years; now, it may soon guide the circuits that define human progress.

Every electronic device we use today relies on the flow of electrons driven by electrical signals. This discovery shows that light can achieve the same effect — faster, cleaner, and with less energy waste. It’s as if the invisible illumination of the universe is learning to speak the language of technology.

The light-induced electron gas behaves like a switch that turns on when photons arrive and vanishes when they leave. It’s elegant, responsive, and entirely controllable — a scientist’s dream come true.

A Step Toward Quantum Futures

Perhaps the most exciting aspect of this discovery lies in its potential for quantum technologies. Quantum computers, sensors, and communication systems depend on delicate control of particles at atomic scales. Materials that respond to light in controllable, reversible ways could become the foundation for next-generation quantum components.

In such systems, light could manipulate quantum states with incredible precision — making operations faster and more stable. The fusion of optics and quantum physics could usher in a new technological revolution, one that makes today’s digital world look primitive by comparison.

Energy Efficiency and Environmental Impact

The modern digital world consumes enormous amounts of energy. From global data centers to billions of mobile devices, our hunger for processing power has an environmental cost. Transitioning to light-driven electronics could help dramatically reduce this burden.

Because photons travel faster and lose less energy than electrons, light-based circuits would generate less heat and require fewer resources to operate. Removing billions of electrical contacts from processors could slash energy consumption while boosting speed. The result would be a sustainable technological ecosystem — one where innovation and environmental responsibility move hand in hand.

The Road Ahead

While this discovery is still at the experimental stage, its implications are immense. Before we can build fully light-powered computers or phones, scientists must refine the materials and understand the underlying physics even more deeply. They’ll need to integrate these oxide layers into real-world chips and ensure that the light-induced effects remain stable, scalable, and durable.

Yet the path is clear — and glowing. Each step forward brings us closer to an era where the boundaries between light and electronics disappear, giving birth to a new generation of devices that are faster, cooler, and infinitely more connected to the natural rhythm of light itself.

A Glimpse of Tomorrow

In the near future, your smartphone could process information using beams of light rather than electric current. Internet signals could travel faster and farther with less energy. Entire data centers could operate almost heat-free, powered by photonic circuits that dance with light.

It’s a vision that feels almost poetic — as if we are teaching our machines to think in brightness, to communicate in color, to compute in the same energy that fuels stars.

The CNRS team’s discovery is not just a milestone in materials science; it’s a turning point in our relationship with technology. It reminds us that the greatest breakthroughs often come from seeing the familiar — in this case, light — in a completely new way.

Conclusion: The Bright Future of Light

This discovery isn’t merely about improving technology; it’s about reimagining it. By using light to control electronic behavior in materials, scientists have opened a new frontier — one where the elegance of physics meets the promise of engineering.

From faster computing to energy-efficient systems and quantum technologies, the fusion of light and matter could illuminate a new chapter in human innovation.

For billions of years, light has been the driving force of life on Earth. Soon, it may also be the driving force of our digital world. The age of light-powered technology has begun — and its future shines brilliantly ahead.

Source

Amazon Web Services is launching a new artificial intelligence platform for everyday business users, moving further beyond core cloud infrastructure to compete more directly against Microsoft’s Copilot, Google’s Gemini, and AI startups.

The platform, Amazon Quick Suite, was announced Thursday morning. It’s a collection of agentic AI tools that can automate tasks by connecting with internal documents and databases, and third-party apps and services.

For example, Amazon says the tools can recap emails, send messages, update project tickets, automate content creation, and build customer strategies, using a company’s own data securely and privately. Quick Suite integrates with third-party services like Salesforce for customer data, Zendesk for support tickets, and Slack for team collaboration, according to the company.

One of Amazon’s advantages: its ability to test new enterprise tools with its own massive workforce before public release. 

The company said Quick Suite has already been deployed to tens of thousands of Amazon employees — reporting that the tool has reduced complex data analysis tasks from months to minutes, for example, based on internal use.

One of its challenges: Unlike its rivals, Amazon does not own a native productivity suite like Microsoft 365 or Google Workspace. 

This means it must convince users to adopt its platform as an overlay on top of the tools they already use. It’s a big hurdle in a market where its competitors can embed their AI assistants directly into their existing software.

Earlier on Thursday, Google announced a unified Gemini Enterprise AI agent subscription, consolidating its business-focused AI tools.

AWS marketing chief Julia White told Bloomberg that existing customers of the Amazon Q Business AI software, which launched 18 months ago, will be encouraged to migrate to the new platform. White was previously a longtime Microsoft executive. 

Amazon says Quick Suite will be available in two tiers: a Professional plan starting at $20 per user per month, and an Enterprise plan at $40 with advanced features.

Quick Suite includes a few individual tools: 

• Quick Sight, a business intelligence tool for data analysis and visualization that puts AWS in direct competition with platforms like Tableau and Microsoft Power BI. 
• Quick Research for generating cited reports from internal and external data.
•  Quick Flows for simple, repetitive tasks.
•  Quick Automate for handling complex, enterprise-wide workflows.

AWS says it has rolled out the platform to hundreds of corporate beta customers, citing examples of significant cost savings and efficiency improvements.

Source

They say there's no such thing as bad publicity, but the press that Intel has generated in the last year has certainly been testing the boundaries of the aphorism.

Is it good when your company posts an annual loss for the first time in almost 40 years? When you're doing multiple rounds of mass layoffs? When your board pushes your CEO into leaving, and starts arguing with the new CEO about major strategy decisions almost immediately? When you're either losing or failing to gain market share in areas critical to your bottom line? When you need to explain to the president why your CEO should keep his job, and then explain to investors the many possible downsides of a mercurial president deciding he wants a stake in your company?

I feel like the answer to these questions is "mostly no." Even Intel's recent investment from and partnership with Nvidia came with a tacit admission that Intel was mostly failing to make a dent in AI hardware and software and the gaming and workstation GPU markets. (Reports that Intel could start manufacturing chips for AMD would be good news, as bizarre as that arrangement would have been at any other point in the two companies' history with one another, but those talks could still fall apart.)

But before it was a company that primarily manufactured bad news, Intel was mainly known for making and selling computer processors. And Intel is hearkening back to that heritage today by taking the wraps off of a new series of chips, codenamed Panther Lake, that will form the basis of its next-generation Core Ultra laptop CPU lineup.

Intel says that systems with these chips in them should be shipping by the end of the year. In recent years, the company has launched a small handful of ultraportable-focused CPUs at the end of the year, and then followed that up with a more fully fleshed-out midrange and high-end lineup at CES in January—we'd expect Intel to stick to that basic approach here.

Panther Lake draws near

Panther Lake tries to combine different aspects of the last-generation Lunar Lake and Arrow Lake chips.

Intel

 

Panther Lake combines three functional chiplets using Intel's Foveros packaging technology.

Intel

 

The three Panther Lake dies use the same package design, making for easy interchangeability for PC manufacturers.

Intel

 

Intel's first Core Ultra chips, codenamed Meteor Lake, were introduced two years ago. There were three big changes that separated these from the 14th-generation Core CPUs and their predecessors: They were constructed of multiple silicon tiles, fused together into one with Intel's Foveros packaging technologies; some of those tiles were manufactured by TSMC rather than Intel; and they added a neural processing unit (NPU) that could be used for on-device machine learning and generative AI applications.

The second-generation Core Ultra chips continued to do all three of those things, but Intel pursued an odd bifurcated strategy that gave different Core Ultra 200-series processors significantly different capabilities.

The most interesting models, codenamed Lunar Lake (aka Core Ultra 200V), integrated the system RAM on the CPU package, which improved performance and power consumption while making them more expensive to buy and complicated to manufacture. These chips included Intel's most up-to-date Arc GPU architecture, codenamed Battlemage, plus an NPU that met the performance requirements for Microsoft's Copilot+ PC initiative.

But Core Ultra 200V chips were mostly used in high-end thin-and-light laptops. Lower-cost and higher-performance laptops got the other kind of Core Ultra 200 chip, codenamed Arrow Lake, which was a mishmash of old and new. The CPU cores used the same architecture as Lunar Lake, and there were usually more of them. But the GPU architecture was older and slower, and the NPU didn't meet the requirements for Copilot+. If Lunar Lake was all-new, Arrow Lake was mostly an updated CPU design fused to a tweaked version of the original Meteor Lake design (confused by all these lakes yet? Welcome to my world).

You'll still see multiple iterations of the new Panther Lake architecture, which will presumably come to market under the "Core Ultra 300" banner, but the chips will no longer use a hodgepodge of mixed-and-matched technologies. All Panther Lake chips are still assembled with Foveros; all Panther Lake chips get the same NPU capable of 50 trillion operations per second (TOPS), the same CPU and GPU architectures, the same media encoding and decoding capabilities, and external RAM (either soldered down or in SODIMM slots). The main difference is how many CPU and GPU cores you get, not what kind of cores you get.

Intel is introducing three distinct Panther Lake chips, all using the same package design.

Comparing the three different Panther Lake configurations.

Intel

 

The 8-core Panther Lake.

Intel

 

The 16-core Panther Lake.

Intel

 

A version of the 16-core chip with less I/O but a bigger GPU.

Intel

 

Each of these three chips is tailored toward a different kind of laptop. The 8-core version will clearly be the mainstream workhorse chip for most midrange Ultrabooks. The 16-core version's extra PCI Express lanes mean we'll probably see it the most frequently in bulkier workstations and gaming laptops with external GPUs. And the 16-core 12Xe version is aimed at high-end thin-and-lights without dedicated GPUs (it will also require soldered-down LPDDR5X, runs at faster speeds, and will maximize the larger integrated GPU's performance).

Intel is sticking to broad and high-level performance comparisons for its new "Cougar Cove" P-core and "Darkmont" E-core architectures and the new Xe3 GPU architecture—the company will probably get more specific when it's actually announcing the specific products, rather than the architecture. But Intel claims we can expect up to a 10 percent improvement in single-core CPU performance compared to Lunar Lake, and up to 50 percent better multi-core CPU performance compared to both Lunar Lake and Arrow Lake. The GPU is said to be roughly 50 percent faster. And Intel says the chip consumes 10 percent less power than Lunar Lake, and 40 percent less power than Arrow Lake.

New CPU architectures.

Intel

 

The media encoding engine makes some tweaks.

Intel

 

On the manufacturing side, Intel has returned to make many of the components of Panther Lake in-house, but as with Meteor/Lunar/Arrow Lake, TSMC is still handling some of the silicon tiles.

Each of these chips uses a total of three functional tiles, not counting the Foveros base tile that binds them together or the "filler tile" that makes them rectangular.

• The compute tile houses all the CPU cores, the NPU, and the media engine and is built on the new Intel 18A process.
• The platform controller tile that handles most I/O, including PCI Express and USB, is still built at TSMC.
• The four-core version of the graphics tile is made using the Intel 3 process (mostly unused in consumer chips, though Intel does use it for server processors). The 12-core version of the GPU tile is still being outsourced to TSMC.

Assuming Intel's performance figures all work out in the real world, early indicators suggest that Panther Lake should be something of a return to form. In older generations, Intel used consistent CPU and GPU architectures all up and down its desktop and laptop lineups, ensuring that chips had similar capabilities even though they targeted much different power and performance levels. Panther Lake returns to something approximating that level of simplicity, and as such requires a whole lot less explaining than the Lunar/Arrow Lake split did.

High-level performance and power comparisons for Panther Lake.

Intel

 

A collection of Panther Lake features.

Intel

 

One thing that remains to be seen is how Intel scales some version of this architecture up to the desktop, where the company has (in the words of CFO David Zinsner) "fumbled the football" in recent generations. That's not what Panther Lake is about, but hopefully we'll see these CPU, GPU, and NPU cores remixed into a compelling high-end desktop chip sooner rather than later.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Friday 10 October 2025 at 2:21 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

After four years of construction, Intel said on Thursday that its Fab 52 semiconductor plant in Chandler, Arizona is now turning out its first chips. The company also shared more details about the long-awaited CPUs that it will be producing in the facility using Intel’s brand new 18A process technology.

The announcement comes just six weeks after the Trump administration acquired a 9.9 percent stake in Intel in exchange for $8.9 billion in stock. The fab opening, while long in the works, is the first major opportunity for the struggling American chip maker to convince the broader tech industry that it can produce some of the world’s most advanced chips at scale—and that the White House’s investment might pay off.

Late last month, Intel invited dozens of analysts and business partners, along with a handful of journalists, to tour Fab 52. The tour offered an extremely rare glimpse into the world of modern chipmaking, where robots perform most tasks, lithography machines the size of school buses print microscopic patterns on silicon wafers, and workers shuffle around in anti-contamination “bunny suits,” booties, goggles, and gloves. (Guests are required to wear the suits, too.) Intel says that the air within the fab is recycled every six seconds.

All of this is to prevent contamination of the fragile silicon wafers that the entire computing industry runs on. If a single speck of anything lands on a wafer, it can be irreparably damaged.

Make or Break

Intel says that its Fab 52 has technically been operational since July, and the new generation of chips being made there, dubbed Panther Lake and Clearwater Forest, have been in the works for years at this point.

But Intel decided to show off its new fab at a critical moment for the company. The facility is designed to make chips using a new process, called 18A, that’s supposed to yield more powerful and efficient products. “Supposed to” is key: Intel’s near-term fate hangs on whether it can produce semiconductors that are impressive enough to not only serve its usual hardware and computer customers, but also attract AI companies with large sums of cash to spend on advanced chips and data centers.

During the tour, Intel executives emphasized that Fab 52 is the most advanced chip manufacturing plant in the world. That may technically be true—the company's fabs, or foundries, “have been long known and respected in the industry for making the next node possible,” says Austin Lyons, an analyst at Creative Strategies and founder of Chipstrat, a semiconductor publication. In the early 2010s, for example, Intel made another significant node, or process advancement, when it introduced 32-nanometer chip technology. (Its latest chips are 2-nanometer.)

“But while Intel is good at figuring out how the industry can build the next-gen technology in a way that can be manufactured, TSMC is good at that and then saying, ‘How can we do this at high volume, high yield?’” Lyons said, referring to Taiwan Semiconductor Manufacturing Company, widely considered the most advanced global chip firm.

Intel first pioneered the chips that revolutionized computing (and defined “Silicon Valley”) in the 1970s. The company started to falter in the mid-2000’s when it failed to capitalize on the mobile phone boom and lost ground to emerging competitors. By that point, the US semiconductor industry had already moved a significant amount of manufacturing overseas to places like Taiwan.

Over the past few years, the rise of generative AI has led to a spike in demand for high-performance computing chips from Intel competitors Nvidia and AMD, putting the company at an even bigger disadvantage. Since 2024, the chipmaker has laid off tens of thousands of employees. In July, Intel’s market capitalization was less than $100 billion dollars, while Nvidia’s was roughly $4 trillion.

On the heels of the Trump administration’s unorthodox bailout, Nvidia also made a $5 billion strategic investment in Intel. The not-so-subtle message: Intel must be saved, but it has proven it can’t save itself.

Inside the Fab

Fab 52 is located on a patch of flat desert land approximately 20 miles outside of Phoenix, Arizona. It’s part of a sprawling complex that also includes Fab 42, which began operating in 2020, and Fab 62, which is still under construction. The entire thing spans one million square feet.

The concrete and steel buildings stand in contrast to the green irrigated fields beside them, land that belongs to the Gila River Indian Community. Intel has assigned cutesy names to the roadways that wind around the facilities, like Data Drive, Transistor Terrace, and Silicon Street—but aside from that and the sheer size of the place, the exterior doesn't betray much of what goes on inside. At the time of the tour, a sign for “Construction Suicide Prevention Week” was affixed to a metal fence. A gigantic American flag hangs from an overpass at the entrance.

Everyone who enters the fabs has to wear a bunny suit, and get dressed—or be dressed—in a clean room. Makeup, hair products, perfumes, colognes, and any aerosol products are prohibited. Workers are separated by a metallurgical hierarchy: There are those who work with copper, and those who do not. The copper people wear orange suits, not white, and have to suit up and strip down in their own clean room.

The Intel fab worker who helped me suit up proudly told me that he has done the same for two US presidents: Obama, who visited Fab 42, and Biden, who visited Fab 52 while it was under construction. As of late September, Trump still hadn’t visited, though Intel spokesperson Cory Pforzheimer said, “We’d eagerly welcome President Trump to see the most advanced R&D and leading-edge semiconductor manufacturing in the US.”

The workers shuffling around are not pulling levers and grinding away at the gears of manufacturing as much as quietly managing robots. They stand at (sterilized) computer stations while containers called front-opening unified pods, or FOUPs, whoosh by overhead through a labyrinth of robotic tracks. The rows of equipment appear endless. The floor below has been reinforced, then reinforced again, because the tiniest of shakes can ruin a whole batch of chips.

The lithography section of the facility is awash in a strange glow, which turned our white suits neon green and the copper-suited people pink. Intel demanded that the fab tourists not share the names of its suppliers, with the exception of one: ASML, the Dutch manufacturer of the world’s most cutting-edge lithography machines. WIRED witnessed two massive ASML Twinscan machines that appeared to be operational. The floor next to them was tape-marked with space for two more.

Intel has not yet publicly said how many semiconductors it expects to successfully yield, or manufacture, at Fab 52 annually. For now, the chips produced there will be used in consumer devices like laptops. But what Intel really needs is the same thing the entire industry is chasing: A hyperscaler customer, a giant data center deal, someone looking to spend billions to get an edge in AI. A whale.

Design Overhaul

Intel’s Panther Lake and Clearwater Forest chips will be made using a manufacturing process that tosses aside decades of proven design techniques in favor of two new technologies the company calls RibbonFET and PowerVia. RibbonFet is an architecture for transistors, stacking them in a way that allows for more density, while PowerVia moves the power interconnects from above the silicon stacks in the chip to below them.

Intel began working on the new design approach in 2021, and early tests have shown that RibbonFet and PowerVia led to performance gains. Reports suggest these new chips also use 30 percent less energy than the prior generation.

“Credit where credit is due, Intel is introducing both of these—RibbonFET and backside power—before TSMC, so they are the ones taking the risk here,” says Lyons. “I hope that bet pays off for them.” Competitors like Samsung and TSMC are now planning to introduce technologies similar to RibbonFet and PowerVia in their 2-nanometer chips.

But the overall performance and reliability of the new chips won’t be clear until they are shipped to customers at scale. Panther Lake, which is Intel’s third series of premium CPUs with dedicated AI processors, will start shipping this quarter. Clearwater Forest, which is designed for servers in data centers, arrives in the first half of 2026.

As Intel ramps up Fab 52 and its cutting-edge 18A process, the company will remain under intense public scrutiny. Customers are watching for any advantage they can use to get ahead in the AI arms race, hoping Intel’s new capacity will give them a performance edge. The Trump administration is eager to champion American manufacturing, but remains a vocal and unpredictable partner.

Wall Street is focused on whether Intel can leverage these advances to lift chip prices, expand margins, and regain momentum in an increasingly competitive market. Analysts and amateur chip nerds, meanwhile, are dissecting Intel’s supply, demand, and capacity to manage risk. And everyone is still questioning whether the chipmaker, once a crown jewel of the American tech sector, has enough momentum to make a real comeback.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Friday 10 October 2025 at 2:19 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

Intel maintains a sizable presence in the United States, especially in Oregon and Arizona. I travelled to the latter for Intel Tech Tour 2025, where Intel shared a bevy of announcements near its absolutely massive Ocotillo campus.

I was able to take a tour of this campus, and most notably Fab 52 — the latest (and largest) addition packed with the most cutting-edge semiconductor fabrication technology in the world. It's this foundry that will produce Intel's next-gen processors using the new 18A processing node.

2026 could be a big year for Intel if its ambitions with Fab 52 are realized. Here's what you should know about Intel 18A, and I'll also share my experience strolling through the impressive Fab 52.

The heart of Intel Panther Lake is 18A

Intel 18A is the first processing node to feature both RibbonFET and PowerVia, two revolutionary semiconductor advancements.

(Image credit: Intel)

 

RibbonFET changes the fundamental design of the transistors at the heart of every processor.

(Image credit: Intel)

 

PowerVia makes use of the backside of the die for the first time.

(Image credit: Intel)

 

Intel 18A, a 2nm process, boasts some impressive gains over Intel 3 (a 3nm process).

(Image credit: Intel)

 

Intel 18A was designed from the ground up with PowerVia and RibbonFET in mind, optimizing costs across the board.

(Image credit: Intel)

 

I've spent some time breaking down Intel's next-gen Panther Lake mobile chips, and explained what you need to know about the new Intel Arc graphics being paired with them. I skimmed over Intel 18A, though, and why exactly it's important.

 

If you weren't aware, one of the most important metrics for gauging the progression of semiconductor fabrication is the size of individual transistors, measured in nanometers. The smaller the size, the more densely you can pack transistors. The more densely you can pack transistors, the more transistors you can have. The more transistors you can have, the better your chip's performance and efficiency (largely speaking).

 

Intel 18A is a 2nm-class processing node, which is on the cutting edge of what modern semiconductor fabrication can achieve. That's important, but Intel 18A is special because it finally debuts two unique technologies in chips that are actually going to reach consumers — RibbonFET and PowerVia.

 

First, RibbonFET is set to replace the "FinFET" transistor design that has been the standard for over a decade. While the latter uses vertical "fin" transistors with the gate (which controls the flow of electricity through the transistors) set on top, RibbonFET uses a new, flexible and scalable "ribbon" design with a gate that wraps completely around the transistors. This design allows for more precise control, less power leakage, superior performance-per-watt, and helps support more densely packed transistors.

 

PowerVia is a similarly crucial advancement. Traditional chips weave together signal and power routing for transistors through the front of the chip die, but this can lead to congestion and performance degradation as transistors become more densely packed. With Intel 18A, chips can now route power to transistors through the backside of the die, leaving the front reserved just for signal input/output.

 

The result of these two technologies is significantly improved performance, efficiency, and consistency with intense, high-power workloads (like artificial intelligence). Intel 18A can also be paired with other Intel processing nodes thanks to Intel's more modular Foveros-3D advanced packaging, which can stack multiple, separate chiplets on the same die.

 

Intel Panther Lake is the first platform to be built using Intel 18A, and takes full advantage of these upgrades. As production scales up, too, we could see other partners turn to Intel for their semiconductor needs — and it's all happening inside Fab 52.

Fab 52 is alive, and I went aside to see what it's all about

Panther Lake is the first platform built on Intel 18A.

(Image credit: Intel)

 

This is one of the cutting-edge EUVs Intel is employing in Fab 52, which etches silicon wafers in a vacuum.

(Image credit: Intel)

 

Much of Intel's foundry is automated, but Intel still employs thousands of engineers, technicians, and other employees in Arizona.

(Image credit: Intel)

 

Miles and miles of tracks can automatically carry wafer between these tools.

(Image credit: Intel)

 

It was fascinating to get a peek behind the scenes, but there's so much that Intel doesn't show regular people.

(Image credit: Intel)

 

Intel's Ocotillo campus is massive. Constantly growing since it first opened in the 1990s, this campus is built on over a square mile of land with its own 12-acre water treatment and recycling plant, and Fab 52 — the latest addition — is large even by Ocotillo standards.

 

Four years of construction required thousands of workers and tens of billions of dollars of investment. Intel had to excavate the equivalent of over 400 Olympic pools filled with soil and rock, and poured over 600,00 cubic meters of concrete — so much concrete that Intel built its own mixing plant on-site — reinforced by over 75,000 tons of steel.

 

It wasn't just about erecting the largest building possible; Intel had to carefully plan how and where to reinforce the ground to support the impossibly heavy Extreme Ultraviolet Lithography (EUV) machines that are crucial for Intel 18A. These machines are so heavy it apparently takes three Boeing 747 cargo planes to transport just one to Arizona, and Fab 52 uses moveable bridge cranes built into its ceiling for construction and maintenance.

 

I got to see these EUVs in person, but I couldn't just stroll in. I had to leave my phone and all electronics behind (not just for Intel's security, but also because any wireless signal can interfere with the absurdly precise and delicate tools inside the foundry), and I had to don a hair net, gloves, and shoe covers before I was even allowed to enter the clean room where I was properly gowned in a full-body suit, hood, boots, protective goggles, and a second pair of gloves.

 

Enter the foundry, and you may note the aggressively "average" feeling of the air, as Intel carefully manages the temperature, humidity, and even air mixture. Dedicated air extraction towers cycle the air in the foundry six times a minute, apparently, and the air is thousands of times cleaner than even your average surgery operating room. You may also note the strange, yellow-green hue of the lighting, which is apparently an additional precaution to protect Intel's chips from the effects of some spectrums of light (similar to a dark room for developing older film).

 

I remember being mesmerized by the literal miles of track suspended from the ceiling, on which hundreds of automated robots carry silicon wafers and dies from tool to tool and building to building, the rows of extraordinarily advanced equipment, and the knowledge that the foundry is constantly under construction inside and out. Intel carefully considers which tools to replace or upgrade on a daily basis, ensuring that every foundry and building plays a role.

 

I won't even pretend to fully understand how exactly humanity managed to make rocks think, but I did learn a lot about semiconductor fabrication during my time with Intel in Arizona. I'm excited to see just how good Panther Lake ends up being, and it'll be especially interesting now that I've seen where Panther Lake chips are born.

 

Source

---

Hope you enjoyed this news post. Feedback welcome.

Posted Friday 10 October 2025 at 2:16 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

Google's flagship AI model, Gemini, is under fire after cybersecurity researchers discovered a serious flaw known as an "ASCII smuggling" exploit. What's worse, the tech giant has stated it won't fix it.

The vulnerability exposes a potential security risk that could allow attackers to manipulate Gemini's responses through invisible commands embedded in text.

Hidden Commands Inside Gemini's Text System

In an early report by Bleeping Computer, cybersecurity researcher Viktor Markopoulos of FireTail was the one who initially identified the vulnerability.

The exploit takes place by putting secret control characters or Unicode points in the text that Gemini recognizes as commands, despite the fact that the users can't see them. The invisible commands can quietly change the behavior of the AI, making it produce false or unexpected results.

Markopoulos showed how the vulnerability could be used with mundane text-based inputs like emails or calendar invitations. For instance, an email that may seem harmless in the eyes of the recipient can have invisible commands.

As Gemini processes or summarizes the email, the AI may inadvertently alter meeting information, skew data, or generate inaccurate summaries, all based on text that appears normal to humans.

Gemini Fails Where Other AIs Succeed

When the same ASCII smuggling attack was tested on other leading AI platforms, including OpenAI's ChatGPT and Microsoft's Copilot, they immediately figured out what was wrong with the hidden inputs.

However, Google's Gemini, along with Elon Musk's Grok and China's DeepSeek, failed to block the exploit, leaving them open to manipulation. This only shows that Gemini's text interpretation engine is more permissive than its competitors, potentially making it a prime target for AI-driven social engineering attacks.

Google Calls It 'Social Engineering', Not a Security Flaw

Despite the security risk, Google has declined to patch the flaw. In response to FireTail's report, the company labeled the ASCII smuggling exploit as a "social engineering" issue rather than a system vulnerability, according to Android Police. Google maintains that the problem lies in users being deceived, not in the AI's code or design.

But cybersecurity professionals contend this move could leave users vulnerable. Gemini interacts with Google's toolset of Gmail, Docs, and Calendar. In theory, hackers could use the bug to disseminate fake information or compromise confidential corporate information.

AI Security Fears

Google has already patched a number of Gemini-related vulnerabilities, such as problems in its logs, summaries, and browsing history, the so-called "Gemini Trifecta." But dismissing ASCII smuggling as not being an issue is a huge red flag that is hard to ignore.

Aside from vulnerabilities, Gemini AI also became a subject of controversy when it reportedly produced different responses mid-sentence. Some questioned Google about the ethics of using an AI tool at that time.

Source

In another loss for early smart home adopters, Logitech has announced that it will brick all Pop switches on October 15.

In August of 2016, Logitech launched Pop switches, which provide quick access to a range of smart home actions, including third-party gadgets. For example, people could set their Pop buttons to launch Philips Hue or Insteon lighting presets, play a playlist from their Sonos speaker, or control Lutron smart blinds. Each button could store three actions, worked by identifying smart home devices on a shared Wi-Fi network, and was controllable via a dedicated Android or iOS app. The Pop Home Switch Starter Pack launched at $100, and individual Pop Add-on Home Switches debuted at $40 each.

A company spokesperson told Ars Technica that Logitech informed customers on September 29 that their Pop switches would soon become e-waste. According to copies of the email shared via Reddit, Logitech’s notice said:

As of October 15, 2025, your POP button(s) and the connected hub will no longer be supported and will lose all functionality.

As an attempt at compensation, Logitech gave affected customers a coupon for 15 percent off some Logitech products, including its Ultimate Ears speakers. The coupon is only valid in the US until March 31, 2026, and doesn't apply to Logitech’s Pro or RS racing wheels for gaming, videoconferencing products, its Logitech for Business line, or "newly released products,” according to the email.

Logitech’s neglected smart home

Logitech’s spokesperson didn't respond to Ars' questions regarding e-waste, the short cancellation notice, or whether Pop button owners can continue using the devices locally after October 15.

“For close to a decade we have been maintaining the POP ecosystem, but as technology evolves, we have made the decision to end support for this device,” Logitech’s representative told Ars, repeating messaging from the email sent to customers.

However, Logitech’s maintenance of the Pop ecosystem has deteriorated lately, per user reports. Over the past two years, people have reported problems in getting the buttons to work consistently with their app, Logitech's Harmony Hub, Samsung SmartThings, and Apple HomeKit.

Logitech originally marketed the Pop switches as being compatible with its Harmony-branded universal remotes. Logitech stopped manufacturing Harmony remotes in 2021 and stopped supporting first-generation remotes in March.

Overall, Logitech appears less committed to smart homes than when it eagerly joined the smart home hype of the 2000s, including by acquiring the company originally behind Harmony remotes, Intrigue, for $29 million in 2004. Logitech CEO Hanneke Faber suggested that the company was moving away from its Circle-series of smart doorbells, telling The Verge last year that she was “not even sure those are still being sold." Logitech quickly backpedaled from Faber's comments, and a couple of Circle products are currently listed on Logitech's website (one is "temporarily" out of stock as of this writing). However, the executive's comments don't instill confidence in the longevity of Logitech's smart home lineup, which hasn't seen a product launch since 2020.

Smart home struggles

The slow decline and ultimate discontinuation of Logitech’s Pop buttons is indicative of the challenges that smart home enthusiasts face in finding quality products that not only work well, but will also be supported long-term. Smart home users have been let down by plenty of new companies that were unable to stay in business. However, even more established companies, like Logitech and Belkin, that eagerly jumped into smart homes years ago are falling short.

This e-waste situation is a reminder of the value of smart home products that allow local control, which ensures that your property isn’t bricked due to changing corporate interests or a failing business.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Thursday 9 October 2025 at 4:34 pm AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

The kiosks will initially be located inside offices for One Medical, Amazon’s primary healthcare company, in locations across the greater Los Angeles area starting in December. Expansion to additional One Medical offices and other locations is expected soon after.

The goal of the freestanding kiosks is to facilitate easier filling of prescriptions, eliminating the need for an extra trip or waiting in line at a conventional pharmacy.

According to Amazon, the kiosks will be stocked with a wide range of commonly prescribed medications, including antibiotics, inhalers, and blood pressure medications. Controlled substances and medications requiring refrigeration are not available. The inventory is tailored to the prescribing patterns of each office location.

After a provider writes a prescription, patients can choose to have it sent to Amazon Pharmacy for in-office kiosk pickup. In the Amazon app, patients get a QR code to scan at the kiosk to pick up their medication, which is delivered within minutes with a custom label printed on the spot.

A patient can still connect with a pharmacist via the Amazon Pharmacy kiosks. (Amazon Photo)

The initiative is not without human intervention — cameras inside the kiosks allow an Amazon pharmacist to get a live view and review medications before they are dispensed, and those pharmacists can also answer any patient questions via a video or phone consultation.

Giving customers a pickup point is a departure for Amazon Pharmacy, which works via home delivery of medications. Hannah McClellan, vice president of operations for Amazon Pharmacy, told GeekWire that Amazon is prepared to scale the kiosks beyond the L.A. rollout.

“We do have many more kiosks ready to go,” McClellan said. “I see One Medical as a launch pad for the kiosk, but I think they have runway far beyond One Medical, and frankly, far beyond primary care offices. There’s so many ways that they can drastically improve the pharmacy pickup experience today.”

The Seattle area has eight One Medical locations, and McClellan said Amazon’s hometown could be “high on the list” for where kiosks are placed.

Drugs are dispensed from Amazon Pharmacy kiosks in minutes, in bottles with custom-printed labels. (Amazon Photo)

The tamper-resistant kiosks weigh approximately 1,700 pounds and are secured to the floor, with additional security features including vibration sensors and surveillance cameras to discourage break-ins. Amazon says patient information is protected in accordance with HIPAA regulations and the entire system has undergone rigorous testing and regulatory review.

There is no additional fee to use a kiosk for Prime and non-Prime members. Patients pay only for their prescription medications, and insurance is accepted just as with traditional pharmacy options.

Amazon’s ultimate hope is that convenience cuts back on the nearly one-third of prescriptions in the U.S. that are never filled each year, leading to better health outcomes and billions in preventable healthcare costs.

“This is not a watered-down version of a pharmacy,” McClellan said. “We have cut zero corners. Every single thing that’s available to you at a retail pharmacy pickup is available to you here. We really have just brought the pharmacy experience to your provider office, and made it incredibly fast and transparent.”

Amazon launched the Amazon Pharmacy service in November 2020, following its $753 million acquisition of prescription-by-mail company PillPack in 2018. The company announced its $3.9 billion acquisition of One Medical in July 2022 and closed the deal in February 2023. In February 2024, Amazon cut hundreds of jobs from both health care units.

Source

Last week, Amazon and Google kick-started what could be the next chapter of the smart home. Their new voice assistants, Alexa Plus and Gemini for Home, have been rebuilt from the ground up on generative AI and large language models to be more conversational, understand context, and take actions. This marks the biggest shift in home control since the companies launched their original smart speakers over a decade ago.

In the years since, smart home adoption has stalled — because it’s complicated and confusing, and the value isn’t always clear. Google and Amazon are betting on this new wave of AI-powered intelligence to deliver a smarter, simpler, more capable smart home. After spending the last week speaking with folks at both companies and seeing their new hardware and software strategies, I’m hopeful. But I see at least three major hurdles: reliability, speed, and proving it’s worth paying for.

While you might think of generative AI as creating text and images, it can also be used to analyze data collected from a home to identify patterns and interpret context. This can provide the intelligence layer the smart home needs to move us from the command-and-control world we’ve been stuck in toward the promised land of ambient computing. With this upgrade, in theory, your home can respond and react proactively to situations, without you needing to spend hours fiddling with apps, setting up automations, or grappling with the exact phrasing required by a voice assistant.

“The biggest gap we’ve had in the last decade is that intelligence layer,” Google Home’s Anish Kattukaran tells me, adding that the command-centric nature of the current Google Assistant, Alexa, and Siri has been a major limitation for the smart home. “As an industry, we solved that with a lot of hard-coding, a lot of if-this, then-that statements.” But with generative AI and LLMs, the assistants can become that intelligence layer that we can more easily interact with, he says. “This is undoubtedly an inflection point.”

For Google, that shift starts this month, with the rollout of Gemini for Home. A “foundational intelligence” that is being added throughout Google’s smart home ecosystem, Gemini for Home is coming to all the company’s existing hardware and its new Google Home app. It will work best, however, on its newest cameras and doorbells, along with an upcoming smart speaker and possible new smart display.

Last week, Google announced a new smart speaker and launched new Nest cameras, which it says are optimized for Gemini.

Photo by Jennifer Pattison Tuohy / The Verge

 

The new AI-powered capabilities primarily focus on a smart home assistant that can understand and use natural language, as well as comprehend, generate, and summarize descriptions of events within and around your home. The features are being added slowly, as part of a new opt-in early access program in the US and several other countries.

Over at Amazon, its AI-upgraded Alexa Plus assistant has been in its own early access program since March. But just last week, Amazon announced that Alexa Plus will be ready to go “out of the box” on its newest hardware in the US.

Panos Panay, who leads Amazon’s devices and services division, tells me that Alexa Plus, combined with the new hardware, will create “magically connected experiences” that he believes will start to transform the smart home.

Amazon launched a new Echo Studio and Echo Dot Max last week, along with two smart displays and new 

cameras from Ring, all designed to optimize Alexa Plus.

Photo by Owen Grove / The Verge

 

These may sound like empty promises for smart home users who have been let down year after year by the smart home’s failure to move beyond single-purpose devices and into a collectively intelligent space. Yet, having used Alexa Plus for a few months and sampled some of the features Gemini is bringing to Google Home, I can see the potential. In the distance.

The most significant immediate change is that both voice assistants can now understand what you mean — not just what you say. For example, you’ll no longer need to remember the name of every light bulb you installed or what you named the back door lock. Instead, phrases like “I’m going to cook dinner, turn the lights on” should turn the kitchen lights on, no matter which room you’re in.

They should also be better at managing your connected gadgets, making the smart home less complicated for everyone in your home. My husband has never created an Alexa Routine in his life, but recently, he wanted the lights to turn off every night at 10PM. He asked me if I could set it up, and I told him to just tell Alexa what he wanted. He did, and it worked. He was impressed — and not a little surprised.

Google has a similar, if more limited, experience. An Ask Home chatbot in the new Google Home app can respond to voice and text inputs to create routines and automations. I saw a demo where it was prompted with the query, “I want to feel safer,” and it suggested routines that simulate presence in the home, such as turning lights on and off, setting up notifications for doors or windows opening, and locking all the doors and turning off the lights when you leave.

This is all encouraging in terms of improving the usability of the smart home, but the industry still needs to ensure that those commands work every time. As many Alexa Plus and some early Gemini users have noted, the new voice assistants don’t always work with our old smart homes. Everyone admits this is a challenge. “LLMs are great at being creative, but not so good at doing the same thing over and over again with the same predictable output,” Kattukaran says. “That’s what the last generation of machine learning was reasonable at. The short-term gap for us is bringing the two together.”

This is why Google is launching two versions of Gemini: Gemini for Home, a more structured assistant that, while capable of understanding natural language, is designed specifically for the home. Then Gemini Live is a more creative, free-flowing chatbot that doesn’t require you to repeat the wake word. It will be available on select smart speakers with a subscription.

Unlike Alexa Plus, Gemini Live has no control over your smart home and can’t take any actions at all. Kattukaran says they plan to merge the two eventually, with Live’s capabilities being what he views as the future for smart home control. However, he cautions that with both Gemini versions, “there will be learning curves; it won’t be perfect on day one.”

Amazon took a different approach, going all in on its LLM for the home. Its Alexa Plus can control your home and have those free-flowing conversations. The company developed a way to connect the new LLMs with the more structured API pathways of the traditional smart home. “What the LLM is doing is thinking, making its round trips, making sure it calls the right API,” Panay says. “I think it’s our secret sauce. Nobody else is at this level.”

In practice, Alexa Plus doesn’t always do what it is told. He says the team is still working on fixing that disconnect, and it’s one of the reasons Alexa Plus is still in early access.

Using visual language models, Amazon says Alexa Plus can interpret footage from Ring cameras and summarize 

events around the home. That includes, in this instance, what didn’t happen — no one fed the dog.

Image: Owen Grove / The Verge

 

Reliably replacing the old with the new is the first hurdle these companies must overcome. In my testing of Alexa Plus so far, I’ve had several instances where things haven’t worked the way they used to. I’ve had to redo old routines because Alexa Plus doesn’t recognize them, and rename more obscure devices because it can’t parse the product name from a natural language command. (I have a smart mosquito repellent system called Liv, and when I ask it to “turn on Liv,” Alexa tries to show me a live stream of my cameras.)

The second hurdle is a heavy reliance on the cloud. During my testing of Alexa Plus and the brief demos I saw of Gemini for Home, the response times were noticeably slower than those of the original assistants, often taking upwards of 10 seconds.

Not everything is that slow — controlling Matter devices in Alexa, which connect locally, is faster, and Google Home should be the same. And Amazon said its new Omnisense platform can process some data locally on its Echo smart speakers.

But Alexa Plus works largely in the cloud. “I’m not worried about latency or security [from the cloud],” Panay says. “The team understands how to deliver the balance to our customers.” He added that he believes the cloud is the right solution. I disagree; a smart home that is reliant on an always-up internet connection is a hobbled smart home.

Finally, and perhaps the highest hurdle of all, they need to create a really compelling product that people will pay for.

It’s clear that both companies view generative AI as the long-awaited recurring revenue model for the smart home. Alexa and Google Assistant were free for the first decade of their existence, but these new versions require a Prime membership or a Google Home Premium subscription for most of Gemini’s advanced capabilities. (You can use the old Alexa and the natural language-capable Gemini for free.) Many of the flagship features also require a Ring or Nest subscription.

Crucial features for Alexa Plus’ new intelligence, like AI-powered text descriptions from Ring cameras, require a subscription.

Photo by Owen Grove / The Verge

 

We’re a long way from something people will pay extra for. But what would tip the scale? Could a proactive AI assistant that not only controls your home but can also manage it for you provide enough value? Both companies are moving in that direction, and it will be interesting to see what they come up with.

“We’ve had this vision for a long time: can we deliver on the promise of having an AI security guard, an AI pet sitter, or an AI elder care?” says Kattukaran. “These are core needs in the home, and we think these components coming together start to deliver on that vision.”

Google’s new Home Brief, which can summarize all the actions your home has taken, lays the groundwork for a proactive system. Amazon’s new Omnisense platform could help Alexa understand more about a home’s occupants and react appropriately. Both companies’ smart security cameras are becoming virtual eyes for their AI assistants, thanks to AI-generated text descriptions and facial recognition technology.

I’d rather see less reliance on invasive cameras and more use of ambient sensing — millimeter wave, ultrawideband, ultrasound, Soli radar, RF sensing — although that’s far harder to implement. This is an area where companies like Apple, with its focus on privacy, and Home Assistant, with its commitment to local control, could step in with some compelling solutions.

Generative AI might be the key to unlocking a truly intelligent home, but there’s clearly still a long road to get there. And the race starts now.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Thursday 9 October 2025 at 2:54 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

As reported by The Verge, Microsoft sent out emails to Xbox Game Pass subscribers across some countries, including Germany, Ireland, South Korea, Poland, and India, informing them that existing customers with auto-recurring billing enabled can continue renewing their subscriptions at the old rates. However, new customers or those who cancel their subscriptions and try to subscribe again will have to subscribe at new rates, which, in terms of dollar value, stand at $29.99/month in the United States.

However, this good news may be short-lived as it likely relates to Microsoft complying with local regulations in the aforementioned countries, which enforces vendors to inform existing customers beforehand about upcoming price hikes. Kari Perez, the head of Xbox communications, is quoted as saying:

Our recent Game Pass update remains unchanged. Current subscribers in certain countries will continue renewing at their existing price for now, in line with local requirements. We’ll provide advance notice before price adjustments take effect in these countries.

The email also indicates that Microsoft will inform gamers in these regions about a price hike at least 60 days in advance, which means that they can mark themselves safe for at least two months.

Microsoft faced a lot of backlash for suddenly increasing the cost of Xbox Game Pass Ultimate from $19.99/month to $29.99/month last week. It tried to justify this hike by offering new benefits such as EA Play, Ubisoft+ Classics, and Fortnite Crew. However, the community hasn't really responded positively, and it will be an uphill battle for Microsoft trying to convince subscribers to stay. But if you're in the U.S. and want to dabble with Game Pass, you're in luck, as Amazon is still selling the subscription at the old rates.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Wednesday 8 October 2025 at 12:32 pm AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

It's been nearly 10 years since we've had a truly beloved Battlefield game with 2016's Battlefield 1, as 2018's Battlefield V was met with a very mixed reception and 2021's Battlefield 2042 was highly contentious due to major gameplay changes to the series' core.

With Battlefield 6, however, Electronic Arts and developer DICE are looking to change that — and all signs point to a massive success in that endeavor. Following an incredible multiplayer reveal in late July, it ran two explosively popular Open Beta weekends in August that shattered a Call of Duty player count record and drove a tidal wave of excitement for Battlefield 6's October 10 release date and launch times.

Now, the game's arrival is only days away, and ahead of that launch, the Battlefield team has shared some interesting insights from Open Beta and closed Battlefield Labs testing along with tweaks changes it's making to Battlefield 6 for its release. Most of these are relatively minor (though still worth looking through), though there's one big decision from the developers that's getting a lot of attention. And it's not the good kind.

The decision in question is the choice to stick with Open Weapons — a setting that allows every infantry soldier class to use every type of weapon — over Closed Weapons that limit the firearm types available to each class (only Recon players can use sniper rifles, for example).

Many in the community have been vocal about their strong preference for Closed Weapons, arguing that with Open Weapons, the differences between classes don't feel meaningful enough, and that certain class/weapon combinations are too difficult to deal with through counterplay.

With the 24 hours of extra play, the Call of Duty: Black Ops 7 open beta will now be ending on Thursday, October 9, at 9am PT instead. As for why the change, Treyarch said that this was due to fan requests regarding an extended beta.

"The reaction to the Black Ops 7 Beta so far has been awesome to see," said the development team in its official blog today. "Whether you’re calling in Nukes in Multiplayer or going for high rounds in Zombies Survival, there’s plenty to jump into and we've had a ton of fun playing alongside everyone. On behalf of all of us at Treyarch, we’re truly grateful for your feedback so far."

Today also marks as the day the beta is gaining another new map. Dubbed Toshin, the 6vs6 multiplayer map is set in Japan, letting players fight it out in a downtown shopping district lit up by advertisements. "Explore the park and market streets around the wreckage and fight through the transit station in a tightknit urban layout that rewards exploration and quick thinking," said the studio, describing the arena.

This bumps up the available multiplayer maps in the Black Ops 7 open beta to six, while Vandorn Farm remains the sole Zombies experience. Treyarch is also keeping the current 2XP and 2X Weapon XP benefits live through the rest of the beta too, as it wants players to hit level 30 and try out all the weapons, attachments, gear, and overclocks available.

Call of Duty: Black Ops 7 is releasing on November 14, 2025, across PC, Xbox One, Xbox Series X|S, PlayStation 4, and PlayStation 5. It will also be available on Game Pass Ultimate on day one.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Tuesday 7 October 2025 at 4:32 pm AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

Under the terms of the arrangement, OpenAI will deploy hundreds of thousands of AMD’s new Instinct MI450 processors over several years. The size of this rollout is huge, as the AI trendsetter plans to build computer systems that, together, will require around 6 gigawatts of power. The first stage of this plan will begin with around 1 gigawatt and is expected to start running in the second half of 2026.

The agreement carries a financial component rarely seen in chip-supply partnerships. AMD has issued OpenAI a warrant giving it the right to purchase as many as 160 million AMD shares for a nominal price of one cent per share. If fully exercised, the option would translate into around a 10% ownership stake in the semiconductor company. The warrant will not be available to OpenAI all at once but will unlock in stages. These stages depend on how successfully OpenAI rolls out AMD’s chips and also on how AMD’s stock performs.

For AMD, the contract could bring potential revenue in the tens of billions of dollars, which would be a massive boost as the company is constantly trying to challenge Nvidia’s dominance in the AI chip market. After the latest announcement, AMD shares had already jumped more than 20% in early trading, reaching their highest value since early 2024. Meanwhile, with this deal, the ChatGPT maker aims to diversify its hardware supply chain.

However, the timing of this latest deal is noteworthy, as it comes just a week after AMD’s prominent rival Nvidia announced plans to invest up to $100 billion in OpenAI. Under that agreement, Nvidia will provide at least 10 gigawatts of its high-performance AI hardware to support OpenAI’s expanding computational needs. Even the AI giant recently also signed a major five-year, $300 billion cloud computing deal with Oracle. But despite all these efforts and its newly reached $500 billion valuation, the AI company still faces significant financial and legal challenges. For example, its $200-per-month ChatGPT Pro plan is reportedly losing money despite its popularity, and OpenAI is involved in several lawsuits, including a recent one from Elon Musk’s xAI over alleged trade secret theft and another filed in August 2025 by the parents of a teenager who took his own life, claiming ChatGPT was partly responsible.

At the same time, AMD is also facing significant challenges, including financial pressures. Recently, the company agreed to pay 15% of its revenue from chip sales to China to the US government in exchange for export licenses.

Source

OpenAI and star designer Jony Ive are grappling with a series of technical issues with their secretive new artificial intelligence device, as they push to launch a blockbuster tech product next year.

The San Francisco-based startup run by Sam Altman acquired the former Apple design chief’s company io for $6.5 billion in May, but the pair have shared few details on the projects they are building.

Their aim is to create a palm-sized device without a screen that can take audio and visual cues from the physical environment and respond to users’ requests.

People familiar with their plans said OpenAI and Ive had yet to solve critical problems that could delay the device’s release.

Despite having hardware developed by Ive and his team—whose alluring designs of the iMac, iPod, and iPhone helped turn Apple into one of the most valuable companies in the world—obstacles remain in the device’s software and the infrastructure needed to power it.

These include deciding on the assistant’s “personality,” privacy issues, and budgeting for the computing power needed to run OpenAI’s models on a mass consumer device.

“Compute is another huge factor for the delay,” said one person close to Ive. “Amazon has the compute for an Alexa, so does Google [for its Home device], but OpenAI is struggling to get enough compute for ChatGPT, let alone an AI device—they need to fix that first.”

A person close to OpenAI said the teething troubles were simply normal parts of the product development process.

Multiple people familiar with the plans said OpenAI and Ive were working on a device roughly the size of a smartphone that users would communicate with through a camera, microphone and speaker. One person suggested it might have multiple cameras.

The gadget is designed to sit on a desk or table but can also be carried around by the user. The Wall Street Journal previously reported some of the specifications around the device.

One person said the device would be “always on” rather than triggered by a word or prompt. The device’s sensors would gather data throughout the day that would help to build its virtual assistant’s “memory.”

OpenAI overtook Elon Musk’s SpaceX to become the world’s most valuable private company this week, after a deal that valued it at $500 billion. One of the ways the ChatGPT maker is seeking to justify the price tag is a push into hardware.

The goal is to improve the “smart speakers” of the past decade, such as Amazon’s Echo speaker and its Alexa digital assistant, which are generally used for a limited set of functions such as listening to music and setting kitchen timers.

OpenAI and Ive are seeking to build a more powerful and useful machine. But two people familiar with the project said that settling on the device’s “voice” and its mannerisms were a challenge.

One issue is ensuring the device only chimes in when useful, preventing it from talking too much or not knowing when to finish the conversation—an ongoing issue with ChatGPT.

“The concept is that you should have a friend who’s a computer who isn’t your weird AI girlfriend... like [Apple’s digital voice assistant] Siri but better,” said one person who was briefed on the plans. OpenAI was looking for “ways for it to be accessible but not intrusive.”

“Model personality is a hard thing to balance,” said another person close to the project. “It can’t be too sycophantic, not too direct, helpful, but doesn’t keep talking in a feedback loop.”

OpenAI’s device will be entering a difficult market. Friend, an AI companion worn as a pendant around your neck, has been criticized for being “creepy” and having a “snarky” personality. An AI pin made by Humane, a company that Altman personally invested in, has been scrapped.

Still, OpenAI has been on a hiring spree to build its hardware business. Its acquisition of io brought in more than 20 former Apple hardware employees poached by Ive from his alma mater. It has also recruited at least a dozen other Apple device experts this year, according to LinkedIn accounts.

It has similarly poached members of Meta’s staff working on the Big Tech group’s Quest headset and smart glasses.

OpenAI is also working with Chinese contract manufacturers, including Luxshare, to create its first device, according to two people familiar with the development that was first reported by The Information. The people added that the device might be assembled outside of China.

OpenAI and LoveFrom, Ive’s design group, declined to comment.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Tuesday 7 October 2025 at 6:16 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

As reported by Bloomberg’s Mark Gurman, Tim Cook will turn 65 next month, and Apple’s board may already be considering his potential successor. John Ternus, Apple’s Senior Vice President of Hardware Engineering, is reportedly the leading candidate to replace Cook for several reasons.

First, John Ternus has a strong technical background, and as Apple’s hardware engineering chief, he has played a pivotal role in the company’s product development. According to Gurman, “Apple probably needs more of a technologist than a sales or operations person.” Given that Apple has faced several shortcomings in areas such as generative AI and mixed reality, the company’s board is more likely to favor a leader with a hardware and engineering background.

“Apple has had great success designing its own chips, but the company has stumbled in areas such as mixed reality, generative AI, the smart home, and autonomous vehicles. That could lead the board to conclude that a product engineering leader like Ternus is the answer, despite him not being known internally as someone who pursues big bets.”

The second determining factor is age. Many of Apple’s senior executives are now over 60 and approaching retirement. However, at 50, Ternus still has the potential to lead Apple for at least the next decade.

According to Gurman, Apple’s marketing and public relations teams are now trying to give John Ternus more public exposure. He asserts, “Those close to the company see little doubt that Ternus will eventually be CEO.”

At Apple’s September event this year, Ternus was responsible for introducing the iPhone Air — one of Apple’s key products and the foundation for the next generation of iPhone design.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Monday 6 October 2025 at 6:00 pm AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

The Shift button is pretty handy in a variety of use cases, apart from capitalization of characters. Many games use it as the default button for sprint functionality, and it's also used by various pieces of software in shortcut combinations. Heck, Windows uses the Shift + Arrow keys to highlight text character by character, along with Win + Shift + S to take screenshots of select portions of your screen.

Meanwhile, the Caps Lock button placed above it is barely used for anything other than capitalization of characters to "shout" during arguments. Its origin dates back to the keyboard layout in mechanical typewriters, where it made sense to have a dedicated key that could hold down the "Shift" key so typists could type upper-case characters quickly, especially in legal documents.

But in modern Windows computers? I don't believe the Caps Lock has any real usage. Most keyboards today follow ergonomic designs where it's very easy to keep your preferred finger (or thumb) on the Shift key as you type in upper case. All Caps Lock does is annoy me when it turns out that I have accidentally pressed it, and my attempt to enter credentials fails because "Caps Lock is on". At this point, Caps Lock is just a relic that needs to die off.

The built-in Narrator in Windows does use this particular key in some scenarios. This is a fairly niche use-case, which, although important, can use other keys for the same purpose. Microsoft itself lets you switch from Caps Lock to Insert in Narrator settings.

The interesting thing is that Google seems to have realized how useless Caps Lock is for most people. Chromebooks have replaced the Caps Lock key with a Search key. Windows and Mac hardware, though, is still intent on keeping this fossil alive for some reason. As a person who solely uses Windows, that's the only platform I care about. There are ways to remap the key to some other functionality, but it's bizarre to me why the hardware for it is even present in the first place.

Microsoft isn't averse to changing keyboard layouts for Windows devices either. It encouraged OEMs to place a dedicated Copilot key on new Windows 11 PCs instead of the right Ctrl key, and while I don't particularly like the company's latest AI assistant, I wouldn't have thrown a fit if it were the Caps Lock key getting the axe instead.

There are so many better alternatives for the Caps Lock key in Windows. It could be used to take a screenshot, open Task Manager, launch Windows Search, toggle Do Not Disturb, and so much. It's just a shame that Microsoft isn't brave enough to encourage OEMs to ditch a legacy layout.

Caps Lock isn’t a feature. It’s a fossil - the appendix of the keyboard. But Microsoft, for reasons known only to the Registry gods, can’t or does not want to kill it.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Monday 6 October 2025 at 2:53 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

Quick links:

1. Windows 10 and 11
2. Windows Insider Program
3. Updates are available
4. Reviews are in
5. Gaming news

Windows 11 and Windows 10

Here, we talk about everything happening around Microsoft's latest operating system in the Stable channel and preview builds: new features, removed features, controversies, bugs, interesting findings, and more. And, of course, you may find a word or two about older versions.

Windows 11 version 25H2 is now available. On September 30, 2025, this year's feature update for Windows 11 became available. You can get the update (the process should be easy and simple) by heading to Windows Update and toggling on the "Get the latest updates as soon as they are available." Microsoft plans to increase the rollout scope in the coming months. Alternatively, you can update or clean-install the operating system with the updated Media Creation Tool app. Keep in mind that MCT is currently broken on Windows on Arm systems.

Before you update, make sure to check the list of known bugs.

After releasing Windows 11 version 25H2, Microsoft announced a delay for Administrator Protection, but one very useful feature skipped all the preview rings and showed up in release, delighting users with multi-monitor systems.

Note that if you want to get all the new features (here is a list of them), you have to install KB5065789, the latest non-security update for Windows 11 versions 24H2 and 25H2. Microsoft also removed a long-standing upgrade block, which prevented certain users from upgrading to Windows 11 version 24H2 and now version 25H2.

On October 1, StatCounter published its latest findings, revealing some interesting data. For one, Google Chrome has reached a new all-time high and now sits at a whopping 71.86%. On the Windows side, Windows 7 is experiencing a sudden increase, and with less than two weeks until the end of Windows 10 support, the operating system suddenly jumped to 9%. On Steam, however, very few people use Windows 7, and the overwhelming majority is already on Windows 11.

We also published some useful stuff for Windows users. They include an article detailing how to check if your PC has a Trusted Platform Module. Also, we reported about updates for apps like WinToys (our favorite app for tweaking Windows 11), Flyoobe (a useful tool for skipping Windows 11's hardware requirements), ExplorerPatcher, another utility for those unhappy with the stock File Explorer, Start11 (this one for those who hate Windows 11's Start menu), and Rufus.

Microsoft is also working on a new OneDrive client for Windows 11, and you can check it out thanks to a recent leak that revealed the app in its full web-based glory.

To finish this week's Windows section, check out these absolutely amazing Microsoft Limited Edition Crocs, all plastered with Windows XP-era icons, including MSN, Clippy, Internet Explorer, and more.

Windows Insider Program

Here is what Microsoft released for Windows Insiders this week:

Updates are available

This section covers software, firmware, and other notable updates (released and coming soon) delivering new features, security fixes, improvements, patches, and more from Microsoft and third parties.

This week, Microsoft announced a new Microsoft 365 Premium tier, which combines Copilot Pro and Office apps for your whole family. With it, Microsoft increased usage limits on AI features for Microsoft 365 subscribers and unveiled a gorgeous set of new Office icons.

Additionally, Microsoft launched Copilot Agent Mode so that you can automate tasks in Word and Excel. By the way, earlier this week, Excel celebrated its 40th birthday.

Look out for more unsolicited app installs in the coming weeks. Microsoft revealed that Microsoft 365 Companion Apps would soon be force-installed on Windows 11 PCs with Microsoft 365 desktop apps.

Finally, check out a recap of everything new in September 2025 for Teams and Excel.

Firefox received two updates this week. Version 143.0.3 brought a pretty big list of fixes for bugs that affect extensions and their updates, problems with keyboard shortcuts resetting when opening extension settings, long load times on certain connections, security issues, and more. Version 143.0.4 was later released to address performance issues on Google websites.

Opera had big news this week. The company started rolling out Neon, its first agentic AI browser, which offers some interesting features, including Neon Do. You can ask the browser to perform various tasks on your behalf, and it will get things done even when you are away. The bad news is that Opera Neon is not free, and it will set you back $19.99/mo.

Finally, Microsoft released Edge 141 in the Stable channel, bringing some important changes to browsing history, passwords, and more. Also, Microsoft is working on some interesting features for Teams and Edge soon.

Here are other updates and releases you may find interesting:

• Microsoft to remove crucial SQL Server on Linux VM images, forcing temporary manual installs.

• Microsoft unlocks collaboration features by bringing Neat into MDEP.

• Satya Nadella promotes Judson Althoff as CEO of Microsoft's commercial business.

• Microsoft unveils open-source agent framework to unify AI development.

• Microsoft promises incredible cloud productivity for US government agencies with Azure Local.

• Microsoft is endorsing the use of personal Copilot in workplaces, frustrating IT admins.

   

Here are the latest drivers and firmware updates released this week:

• Nvidia 581.42 driver brings optimizations for Battlefield 6 and DLSS 4 for FBC: Firebreak.
• Nvidia App update introduced Project G-Assist for laptops and other improvements.

• Nvidia 581.47 hotfix is out to fix game crashes.

   

• Intel 32.0101.8135 non-WHQL with optimizations for Call of Duty: Black Ops 7 Beta.

Reviews are in

Here is the hardware and software we reviewed this week

Steven Parker published a hands-on review of the Mixx Resonate Pro, waterproof bone conduction headphones with some interesting and useful features, such as 32GB of onboard storage so that you can take your music or podcasts with you while swimming, running, etc.

Robbie Khan reviewed the Xiaomi 15T Pro, a new high-end Android smartphone from Xiaomi, which boasts great design and build quality, a fantastic AMOLED display, and a decent price tag. It is not without quirks, but the 15T Pro still managed to score 8 out of 10.

Also, Robbie reviewed the MX Master 4, the latest flagship mouse from Logitech, which promises some interesting new features and design tweaks to make one of the most popular (and one of the most expensive) mice even better.

On the gaming side

Learn about upcoming game releases, Xbox rumors, new hardware, software updates, freebies, deals, discounts, and more.

Game Pass is becoming more expensive. Again. This week, Microsoft revealed a whopping 50% price increase for Xbox Game Pass Ultimate. It now costs $29.99 per month, which upset quite a lot of Game Pass users. However, to make up for the price increase, Microsoft announced 50 new games for Game Pass, most of which are coming from Ubisoft.

The good news is that you can still stack up some codes at the old price, as Amazon has plenty of them for you to grab.

Nvidia announced new games for GeForce NOW. The latest additions include Battlefield 6, Little Nightmares III, Jurassic World Evolution 3, Vampire: The Masquerade – Bloodlines 2, and more throughout October.

Deals and freebies

The Epic Games Store is giving away the survival game Nightingale, so grab it while the offer lasts (until next Thursday). By the way, the Steam Autumn Sale is still running, so you can get a lot of games with big discounts. As usual, you can find more deals in our weekly Weekend PC Game Deals series.

Other gaming news includes the following:

• Colossal Order announced a free bicycle update for Cities: Skylines II, as its first expansion nears release.

• Battlefield 6's post-launch schedule has two new maps arriving before 2026.

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Monday 6 October 2025 at 2:48 am AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

Paul's Hardware (1.53M subscribers)

October 5, 2025

Video length: 12m 21s

0:00 Welcome to Paul’s Tech News - October 5, 2025 

1:23 Intel Might Make AMD Chips 

3:46 EA's $55 billion acquisition is the biggest leveraged buyout in private equity history 

TECH BRIEFS 

5:55 MSI Confirms “Future CPU” (Zen 6?) Support On AM5 800-Series Mobos 

7:08 Intel Arc GPU Driver Overhead Issues Fixed 

8:28 OpenAI's Stargate project to consume up to 40% of global DRAM output 

9:43 Portable 1TB bamboo SSD has built-in aroma diffuser 

10:12 ASUS addresses stuttering issues, beta patch released for ROG laptops 

10:57 AOL Ends Dial Up Service After 34 Years

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Sunday 5 October 2025 at 5:43 pm AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of September): 4,533

RIP Matrix

Most of us have likely experienced artificial intelligence (AI) voices through personal assistants like Siri or Alexa, with their flat intonation and mechanical delivery giving us the impression that we could easily distinguish between an AI-generated voice and a real person. But scientists now say the average listener can no longer tell the difference between real people and "deepfake" voices.

In a new study published Sept. 24 in the journal PLoS One, researchers showed that when people listen to human voices — alongside AI-generated versions of the same voices — they cannot accurately identify which are real and which are fake.

"AI-generated voices are all around us now. We’ve all spoken to Alexa or Siri, or had our calls taken by automated customer service systems," said lead author of the study Nadine Lavan, senior lecturer in psychology at Queen Mary University of London, in a statement. "Those things don’t quite sound like real human voices, but it was only a matter of time until AI technology began to produce naturalistic, human-sounding speech."

The study suggested that, while generic voices created from scratch were not deemed to be realistic, voice clones trained on the voices of real people — deepfake audio — were found to be just as believable as their real-life counterparts.

The scientists gave study participants samples of 80 different voices (40 AI-generated voices and 40 real human voices) and asked them to label which they thought was real and AI-generated. On average, only 41% of the from-scratch AI voices were misclassified as being human, which suggested it is still possible, in most cases, to tell them apart from real people.

However, for AI voices cloned from humans, the majority (58%) of were misclassified as being human. Only slightly more (62%) of the human voices were classified correctly as being human, leading the researchers to conclude that there was no statistical difference in our capacity to tell the voices of real people apart from their deepfake clones.

The results have potentially profound implications for ethics, copyright and security, Lavan said. Should criminals use AI to clone your voice, it becomes that much easier to bypass voice authentication protocols at the bank or to trick your loved ones into transferring money.

We've already seen several incidents play out. On July 9, for example, Sharon Brightwell was tricked out of $15,000. Brightwell listened to what she thought was her daughter crying down the phone, telling her that she had been in an accident and that she needed money for legal representation to keep her out of jail. "There is nobody that could convince me that it wasn’t her," Brightwell said of the realistic AI fabrication at the time.

Lifelike AI voices can also be used to fabricate statements by, and interviews with, politicians or celebrities. Fake audio might be used to discredit individuals or to incite unrest, sowing social division and conflict. Con artists recently built an AI clone of the voice of Queensland Premier Steven Miles, using his profile to try to get people to invest in a Bitcoin scam, for instance.

The researchers emphasised that the voice clones they used in the study were not even particularly sophisticated. They made them with commercially available software and trained them with as little as four minutes of human speech recordings.

"The process required minimal expertise, only a few minutes of voice recordings, and almost no money," Navan said in the statement. "It just shows how accessible and sophisticated AI voice technology has become."

While deepfakes present a multitude of opportunities for malign actors, it isn’t all bad news; there may be more positive opportunities that come with the power to generate AI voices at scale. "There might be applications for improved accessibility, education, and communication, where bespoke high-quality synthetic voices can enhance user experience," Navan said.

Source

Electronic Arts' (EA) and DICE's long-running, large-scale FPS franchise Battlefield has struggled to hold a place in the limelight for quite a while now, with games like Call of Duty and others overshadowing titles like 2018's Battlefield V and especially 2021's Battlefield 2042. Now, though, the series is poised to make a huge comeback with the latest installment Battlefield 6.

Featuring the return of Battlefield's classic class-based structure, a variety of new and remade maps, creative new gadgets and advanced destruction systems that can reshape the battlefront, a suite of campaign, multiplayer, Halo Forge-like Portal custom games, and battle royale content, and a grounded modern day military aesthetic, it's ticking all the right boxes.

It's not surprising, then, that hype for the title has skyrocketed in the months and weeks leading up to its launch — evidenced by August's Battlefield 6 Open Beta absolutely blowing up, and public interest seemingly even soaring past the likes of Call of Duty: Black Ops 7.

With that in mind, many are naturally wondering when they'll be able to jump into the full experience. Luckily, ahead of the game's release, EA and DICE have publicly announced its official release date and launch times, all of which you'll find below.

"We believe that the ongoing advancement of AI-assisted protein design holds great promise for tackling critical challenges in health and the life sciences, with the potential to deliver overwhelmingly positive impacts on people and society," commented the researchers in their paper published in the journal Science. "As with other emerging technologies, however, it is also crucial to proactively identify and mitigate risks arising from novel capabilities."

Testing defenses

When biotech companies make DNA for researchers, they use Biosecurity Screening Software (BSS) to look for similarities between the new sequence and a database of known threats. And that is both a strength and a weakness because it can only screen against what is listed in the database.

To test this biosecurity gap, the Microsoft researchers used publicly available AI programs to create more than 76,000 synthetic variants of known dangerous proteins, including ricin. They didn't actually produce the proteins; they designed the genetic instructions for their synthesis. Then they ran the sequences through four different screening software tools to see if any could slip through. And they did—in big numbers. A significant percentage of these AI-designed sequences breezed through the checks.

After discovering the flaws, the Microsoft team worked with BSS providers to develop patches. These included updating threat databases and fine-tuning the screening software. The result? The beefed-up screening tools caught 97% of the most dangerous sequences in a second test.

The research serves as a clear warning. Even though the patches increased the detection rate, they were not foolproof, as 3% of potentially dangerous sequences were missed. It's also unclear how the resulting proteins would perform in the real world, as the sequences were computer predictions.

Clearly, more work will be needed to build robust defenses against increasingly sophisticated AI techniques, but this effort will be ongoing. A constant evolutionary arms race is inevitable. Just as vaccines must keep pace with new viral mutations, so too will biosecurity screening tools need continuous updates to counter AI-generated threats.

Source