Technology

Users in the test group are unimpressed by the video ads, which play automatically. Customers say they will be eager to toss their Roku devices if the ads become permanent.

The taxi service will initially operate with Tesla employee drivers only, use current-model vehicles, and won’t allow for driverless rides.

• Apple’s first foldable device could have impressive battery life
• That’s according to a new rumor that details Apple’s latest efforts
• It would compete with Samsung’s best foldable on battery life

Apple’s first foldable device appears to be on the way – we got another rumor seemingly confirming that just today – and now a new detail has emerged that could help Apple compete with the likes of the Samsung Galaxy Z Fold 7.

According to a Chinese leaker known as Phone Chip Expert (via MacRumors), Apple has been pulling out all the stops to increase the efficiency of its foldable display. That could give it an edge against the best foldable phones in what is a highly competitive market.

To achieve this, Phone Chip Expert claims that Apple has reduced the display driver integrated circuit from 28nm to 16nm. That’s a massive reduction and could significantly impact battery life.

Apple likes to promise that its devices come with “all day battery life,” and when you’re dealing with a large screen like on a foldable product, battery life is key. That makes Phone Chip Expert’s claim seem believable at first glance.

All day battery life

(Image credit: Future)

A move like this makes a lot of sense. After all, a foldable iPhone or MacBook will have a much larger display than its non-foldable siblings, and all that screen real estate can get awfully power-hungry. That’s something we’ve seen in other foldable devices like the Asus Zenbook 17 Fold, which have failed to impress us with their mediocre battery life.

Apple is thought to be building both a foldable iPhone and a MacBook-iPad hybrid with a foldable display. In the case of the former, it’s likely to directly compete with the upcoming Samsung Galaxy Z Fold 7, which is rumored to get the same battery capacity as the Galaxy Z Fold 6.

That said, Samsung is apparently working on making its foldable screen more efficient, so it will be fascinating to see which device comes out on top here.

At the moment, most of this is confined to the land of rumors, but it is worth noting that Phone Chip Expert has a reasonably solid track record regarding Apple rumors.

They correctly predicted that Apple would build its own servers to run Apple Intelligence and that the iPhone 15 would continue to use the A15 Bionic chip instead of the newer A16 chip, among other claims.

That doesn’t mean that this latest rumor is nailed on to happen, though. Various sources have claimed that Apple’s first foldable device could arrive in either 2026 or 2027, so we’ll have to wait until then to judge the efficiency and battery life of Apple’s first foldable product.

You might also like

• Apple’s foldable iPhone specs hinted at in fresh leak – here are 5 things to expect
• Apple expert just tipped a load of iPhone 17 upgrades: here are 7 things we’ve learned
• New rumors predict a foldable iPhone will launch next year – and cost almost twice as much as the iPhone 16 Pro Max

Celebrate this major milestone with a thoughtful, practical gift that will help them get ready to take on the world.

• HP Threat Insights Report reveals new malware campaigns
• Victims have their data exfiltrated by a remote access trojan
• Attackers have been observed using fake CAPTCHA verification pages

New research has claimed victims are increasingly infecting themselves with malware thanks to a surge in fake CAPTCHA verification tests - taking advantage of a growing ‘click tolerance’ as users are increasingly accustomed to ‘jumping through hoops to authenticate themselves online.’

This isn’t the first report to flag this attack, with security researchers identifying fake CAPTCHA pages spreading infostealer malware in late 2024, but HP’s latest Threat Insights Report now warns this is on the rise.

Users were commonly directed to attacker-controlled websites, and then pushed to complete convincing but fake authentication challenges.

More campaigns identified

These false CAPTCHAs usually trick users into running malicious PowerShell commands on their device that install a Lumma Stealer remote access trojan - a popular infostealer capable of exfiltrating a wide range of sensitive information, like browser details, email credentials, client data, and even cryptocurrency wallets.

Fake CAPTCHA spreading wasn’t the only threat uncovered, with attackers also able to access end-users webcams and microphones in concerning attacks spread via social engineering attacks, primarily using open source RAT and XenoRat to control devices, exfiltrate data, and log keystrokes.

Alongside this, attackers were observed delivering malicious JavaScript code “inside Scalable Vector Graphic (SVG) images to evade detection”. These images are opened “by default” in browsers, and the embedded code is executed, “offering redundancy and monetization opportunities for the attacker” thanks to the remote access tools.

"A common thread across these campaigns is the use of obfuscation and anti-analysis techniques to slow down investigations," said Patrick Schläpfer, Principal Threat Researcher in the HP Security Lab.

“Even simple but effective defence evasion techniques can delay the detection and response of security operations teams, making it harder to contain an intrusion. By using methods like direct system calls, attackers make it tougher for security tools to catch malicious activity, giving them more time to operate undetected – and compromise victims endpoints."

You might also like

• Take a look at our picks for the best AI tools around
• Check out our choice for best antivirus software
• The FCC is creating a security council to bolster US defenses against cyberattacks

WIRED talked to actual federal auditors about how government auditing works—and how DOGE is doing the opposite.

Here are hints and the answer for today's Wordle No. 1,369 for March 19.

Hints and answers for Connections for March 19, #647.

Here are hints and answers for the NYT Strands puzzle No. 381 for March 19.

Hints and answers for the NYT Connections: Sports Edition puzzle, No. 177, for March 19.

Nvidia has unveiled a dozen new GPUs at its GTC 2025 event, the company's biggest launch in two years.

Based on the Blackwell architecture, the newcomers use the RTX Pro moniker to differentiate themselves from the previous generations (Ada Lovelace, Ampere, and Turing) and, from their consumer breathens.

The flagship models are three RTX Pro 6000 Blackwell GPU variants with 96GB ECC GDDR7 memory and up to 4000 AI TOPS performance - twice the amount of memory in its former performance champion, the RTX 6000 and a staggering 4x the RTX 5090, the best GPU on the consumer market has to offer.

Nvidia GPU launches

Alongside the standard Workstation Edition, Nvidia also introduced the Blackwell Max-Q Workstation Edition and the Server Edition.

The latter is the successor to the L40 Data Center GPU series, bringing some much-needed consistency to the GPU nomenclature.

As for the Max-Q Workstation Edition, it remains a bit of a mystery. Nvidia launched Max-Q technology back in 2017 and this is usually associated with laptop GPUs trying to achieve maximum efficiency.

However, workstation PCs rarely aim for optimal energy consumption except perhaps in power-constrained environments like small form factor mini PCs.

Three other professional desktop GPUs were also introduced: the Pro 5000, Pro 4500 and Pro 4000, which will be available starting May 2025.

Given past product launch cycles, I expect more models focusing on the entry-level and mainstream parts of the market, to be launched by the end of 2025.

Six new laptop GPUs were also launched, all of them carrying the RTX Pro naming convention and, confusingly enough, some having the same name as their desktop counterparts.

The RTX Pro 5000 Blackwell is the new laptop flagship GPU with 24GB ECC GDDR7 memory; other models include the 4000, 3000, 2000, 1000, and 500 series and should directly replace their respective “Ada Generation” part.

All these parts will be available from OEM partners in mobile workstations starting in June 2025.

We’ll strive to update this article when further details of the cards (including pricing) are published.

Nvidia’s GTC Keynote also saw the formal launch of DGX Spark, formerly known as Project Digits, the DGX station and Blackwell Ultra (or GB300), Nvidia’s most powerful GPU ever.

You might also like

• These are the best business laptops on the market
• Check out our list of the best rugged laptops
• We curated the best mobile workstations right now

A surprising announcement at Nvidia GTC 2025 was the launch of the DGX station, a powerful supercomputer-class workstation PC that looks a lot like a traditional tower computer but with an Arm-based CPU inside.

This is not the first workstation Nvidia launched; it famously partnered with AMD to launch the precursor to the 2025 DGX Station called the DGX Station A100.

That one didn't have a Nvidia Arm CPU and needed separate PCIe AI accelerators (A100); the 2025 iteration doesn't. It also carried a price of more than $100,000 at launch.

Nvidia DGX Station

Nvidia confirmed that Asus, Boxx, Dell, HP, Lambda and Supermicro will sell their own versions of the DGX Station: the big name missing out is Lenovo.

The GB300 Grace Blackwell Ultra Desktop Superchip that powers it delivers up to “20 PFlops of AI performance” which is likely to be measured using FP4 with sparsity.

That would also infer that it is half the performance of GB200 Grace Blackwell Superchip, so something’s not clear here and I wonder whether there’s more than one version of the GB300.

Nvidia hasn’t said how many or what type of (Arm) CPU cores the GB300 uses; ditto for the GPU subsystem.

Its predecessor, GH200, had 72 Arm Neoverse 2 CPU cores clocked at 3.1GHz, up to 144GB HBMe memory and 480GB LPDDR5x memory.

What we do know is that it has 784GB of unified system memory, which one can assume means HBM (288GB HBM3e) plus what Nvidia calls Fast Memory (496GB LPDDR5x most probably).

The original DGX Station (Image credit: Nvidia) Fastest NIC in a computer

Nvidia also disclosed the DGX Station will use its proprietary ConnectX-8 SuperNIC, a network technology that can deliver up to a staggering data center-class 800Gb/s connectivity.

A close-up of the opened chassis shows the workstation has three forward-facing 120mm fans, a motherboard with three PCIe slots, an Nvidia-branded soldered chip (perhaps the SuperNic), and two large uncovered dies.

One of which is the Grace GPU and the other with eight distinct tiles, the Blackwell GPU.

No details about expansion or storage capabilities, the PSU capacity, the cooling solution used or the price have been revealed.

Additionally, we do not know whether you will be able to plug in accelerator cards like the H200 NVL (or a theoretical B300 NVL) to significantly improve the performance of the DGX Station.

The DGX station is expected to compete with the likes of the Camino Grando, an EPYC-powered tower workstation that packs two AMD CPUs and up to eight GPUs.

We’ll strive to update this article when further details of this workstation PC (including pricing and availability) are published.

Nvidia’s GTC Keynote also saw the formal launch of DGX Spark, formerly known as Project Digits, 12 new professional GPUs and Blackwell Ultra (or GB300), Nvidia’s most powerful GPU ever.

You might also like

• These are the best business laptops on the market
• Check out our list of the best rugged laptops
• We curated the best mobile workstations right now

Henderson has plenty of fast internet options, but choosing the right one can be confusing. Don’t worry -- CNET experts are here to help.

A local fiber internet provider is the best choice for most Burlington residents, but other companies also offer affordable plans and fast speeds.

We're at that stage in the month where we're having to prepare ourselves for the next wave of movies that Hulu will be removing from its catalog.

But it looks like April is shaping up to be frightfully different compared to Hulu's previous lists. It's almost impossible to say but, I think this is the shortest 'leaving' list I've ever seen – and that's across all the best streaming services, not just Hulu.

A mere nine titles are getting the chop from Hulu this month, including seven movies and two documentaries, so it's a pleasure for me to say that Hulu's best movies and best TV shows have gained an extra life with their time on the platform. Have I heard of any of them? Never in my life – that's when you know you're not missing out on much.

But what's more exciting are the brand new movies and shows that are being added to Hulu in April – indeed, we're most excited for the arrival of the sixth and final season of The Handmaid's Tale, even if we must say our final goodbyes.

Everything leaving Hulu in April 2025

Leaving on April 6

Agnes (movie)

Leaving on April 13

She Will (movie)

Leaving on April 16

Toni Morrison: The Pieces I Am (documentary)

Leaving on April 20

Totally Under Control (documentary)

Leaving on April 24

The Good Neighbor (movie)

Leaving on April 27

Resurrection (movie)

Leaving on April 30

After Everything (movie)
Code Name Banshee (movie)
Stars Fell Again (movie)

You might also like

• More YouTubers are getting their own TV shows – here are 3 with over 75% on Rotten Tomatoes
• 'There's a reason why we do it': The Wheel of Time showrunner responds to fans who are still upset over the Prime Video show's plot alterations
• Reacher season 3 becomes Prime Video’s biggest returning show thanks to Hollywood’s biggest heavyweight

• Nvidia unveils Blackwell Ultra at GTC 2025
• New flagship GPU offers more power and efficiency than previous generations
• There's also a new Nvidia RTX PRO 6000 Blackwell Server Edition designed for enterprise workloads

Nvidia has taken the wraps off its latest Blackwell Ultra flagship GPU hardware as it looks to assert its place as the global leader in AI computing.

Revealed at its Nvidia GTC event in San Jose, the new Blackwell Ultra hardware offers more power and efficiency in the data center than previous generations as the company looks to establish itself as the backbone of future AI development and deployment.

Declaring it to be "built for the age of AI reasoning", Nvidia says Blackwell Ultra can help democratize AI adoption across the world, making it possible for more organizations to enjoy the benefits such compute can bring.

Nvidia Blackwell Ultra

Nvidia noted that the rapid growth of AI use cases around the world in the past few years has led for a huge demand for compute, as businesses and consumers alike clamour for more.

“AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance,” said Jensen Huang, founder and CEO of NVIDIA.

“We designed Blackwell Ultra for this moment — it’s a single versatile platform that can easily and efficiently do pretraining, post-training and reasoning AI inference.”

The increase in reasoning models in particular has led for a boom in requirements for a full-stack offering that is cost-effective but also gets the job done.

Built on the initial Blackwell architecture unveiled at GTC 2024, Blackwell Ultra will offer 1.5x more FP4 inference, and are set to be available ind evices built by Nvidia partners in the second half of 2025.

It will be present in a host of new offerings from Nvidia, including the upgraded GB300 NVL72 rack, which it says offers improved energy efficiency and serviceability, with bandwidth speeds of up to 130Tb/s.

Nvidia RTX PRO 6000 Blackwell Server Edition

However that wasn't all when it comes to new Blackwell hardware, as the company also revealed the Nvidia RTX PRO 6000 Blackwell Server Edition, designed for enterprise workloads such as multimodal AI inference, immersive content creation and scientific computing.

Nvidia says the new release offers huge advances on the previous-generation Ada Lovelace architecture L40S GPU, providing up to 5x higher LLM inference throughput for agentic AI applications, nearly 7x faster genomics sequencing, 3.3x speedups for text-to-video generation, nearly 2x faster inference for recommender systems and over 2x speedups for rendering.

(Image credit: Nvidia)

Each RTX PRO 6000 can also be partitioned into as many as four fully isolated instances with 24GB each to run simultaneous AI and graphics workloads, with 96GB of ultrafast GDDR7 memory and support for multi-instance GPU.

They can also be configured in high-density accelerated computing platforms for distributed inference workloads — or used to deliver virtual workstations using Nvidia vGPU software, for graphics-intensive applications or to power AI development.

"With the RTX PRO 6000 Blackwell Server Edition, enterprises across various sectors —including architecture, automotive, cloud services, financial services, game development, healthcare, manufacturing, media and entertainment and retail — can enable breakthrough performance for workloads such as multimodal generative AI, data analytics, engineering simulation, and visual computing," noted Nvidia's Sandeep Gupte.

Nvidia has taken the world a step closer to smart, humanoid robots with the launch of its latest advanced AI model.

At its Nvidia GTC 2025 event, the company revealed Isaac GROOT N1, which it says is, "the world’s first open Humanoid Robot foundation model", alongside several other important development tools.

Nvidia says its tools, which are available now, will make developing smarter and more functional robots easier than ever, along with allowing them to have more humanoid reasoning and skills - which doesn't sound terrifying at all.

Isaac GROOT N1

“The age of generalist robotics is here,” said Jensen Huang, founder and CEO of NVIDIA. “With NVIDIA Isaac GR00T N1 and new data-generation and robot-learning frameworks, robotics developers everywhere will open the next frontier in the age of AI.”

The company says its robotics work can help fill a shortfall of more than 50 million caused by a global labor shortage.

Nvidia says Isaac GROOT N1, which can be trained on real or synthetic data, can "easily" master tasks such as grasping, moving objects with either a single or multiple arms, and moving items from one arm to the other - but can also carry out multi-step tasks which combine a number of general skills.

The model is built across a dual-system architecture inspired by the principles of human cognition, with “System 1” is a fast-thinking action model, mirroring human reflexes or intuition, whereas “System 2” is a slow-thinking model for "deliberate, methodical decision-making."

Powered by a vision language model, System 2 is able to consider and analyze its environment, and the instructions it was given, to plan actions - which are then translated by System 1 into precise, continuous robot movements.

Among the other tools being released are a range of simulation frameworks and blueprints such as the NVIDIA Isaac GR00T Blueprint for generating synthetic data, which help generate large, detailed synthetic data sets needed for robot development which would be prohibitively expensive to gather in real life.

There is also Newton, an open source physics engine, created alongside Google DeepMind and Disney Research, which Nvidia says is purpose-built for developing robots.

Huang was joined on stage by Star Wars-inspired BDX droids during his GTC keynote, showing the possibilities of the technology in theme parks or other entertainment locations.

Nvidia first launched Project GROOT ("Generalist Robot 00 Technology") at GTC 2024, primarily focusing on industrial use cases, which could learn and become smarter by watching human behaviour, understanding natural language and emulating movements, allowing them to quickly learn coordination, dexterity and other skills in order to navigate, adapt and interact with the real world.

When No One Sees Us, The White Lotus, and Celtics City are just a few of the shows you need to be watching on Max this month.

Amazon is turning off the ability to process voice requests locally. It's a seemingly major privacy pivot and one that some Alexa users might not appreciate. However, this change affects exactly three Echo devices and only if you actively enabled Do Not Send Voice Recordings in the Alexa app settings.

Right. It's potentially not that big of a deal and, to be fair, the level of artificial intelligence Alexa+ is promising, let alone the models it'll be using, all but precludes local processing. It's pretty much what Daniel Rausch, Amazon's VP of Alexa and Echo, told us when he explained that these queries would be encrypted, sent to the cloud, and then processed by Amazon's and partner Antrhopic's AI models at servers far, far away.

That's what's happening, but let's unpack the general freakout.

After Amazon sent an email to customers, actually only those it seems who own an Echo Dot 4, Echo Show 10 (3rd Gen), and Echo Show 15, that the option to have Alexa voice queries processed on device would end on March 28, some in the media cried foul.

They had a point: Amazon didn't have the best track record when it comes to protecting your privacy. In 2019, there were reports of Amazon employees listening to customer recordings. Later, there were concerns that Amazon might hold onto recordings of, say, you yelling at Alexa because it didn't play the right song.

Amazon has since cleaned up its data act with encryption and, with this latest update, promises to delete your recordings from its servers.

A change for the few

(Image credit: Future)

This latest change, though, sounded like a step back because it takes away a consumer control, one that some might've been using to keep their voice data off Amazon's servers.

However, the vast majority of Echo devices out there aren't even capable of on-device voice processing, which is why most of them didn't even have this control.

A few years ago, Amazon published a technical paper on its efforts to bring "On-device speech processing" to Echo devices. They were doing so to put "processing on the edge," and reduce latency and bandwidth consumption.

Turns out it wasn't easy – Amazon described it as a massive undertaking. The goal was to put automatic speech recognition, whisper detection, and speech identification locally on a tiny, relatively low-powered smart speaker system. Quite a trick, considering that in the cloud, each process ran "on separate server nodes with their own powerful processors."

The paper goes into significant detail, but suffice it to say that Amazon developers used a lot of compression to get Alexa's relatively small AI models to work on local hardware.

It was always the cloud

In the end, the on-device audio processing was only available on those three Echo models, but there is a wrinkle here.

The specific feature Amazon is disabling, "Do Not Send Voice Recordings," never precluded your prompts from being handled in the Amazon cloud.

The processing power that these few Echos had was not to handle the full Alexa query locally. Instead, the silicon was used to recognize the wake word ("Alexa"), record the voice prompt, use voice recognition to make a text transcription of the prompt, and send that text to Amazon's cloud, where the AI acts on it and sends a response.

The local audio is then deleted.

Big models need cloud-based power

(Image credit: Amazon)

Granted, this is likely how everyone would want their Echo and Alexa experience to work. Amazon gets the text it needs but not the audio.

But that's not how the Alexa experience works for most Echo owners. I don't know how many people own those particular Echo models, but there are almost two dozen different Echo devices, and this affects just three of them.

Even if those are the most popular Echos, the change only affects people who dug into Alexa settings to enable "Do Not Send Voice Recordings." Most consumers are not making those kinds of adjustments.

This brings us back to why Amazon is doing this. Alexa+ is a far smarter and more powerful AI with generative, conversational capabilities. Its ability to understand your intentions may hinge not only on what you say, but your tone of voice.

It's true that even though your voice data will be encrypted in transit, it surely has to be decrypted in the cloud for Alexa's various models to interpret and act on it. Amazon is promising safety and security, and to be fair, when you talk to ChatGPT Voice and Gemini Live, their cloud systems are listening to your voice, too.

When we asked Amazon about the change, here's what they told us:

“The Alexa experience is designed to protect our customers’ privacy and keep their data secure, and that’s not changing. We’re focusing on the privacy tools and controls that our customers use most and work well with generative AI experiences that rely on the processing power of Amazon’s secure cloud. Customers can continue to choose from a robust set of tools and controls, including the option to not save their voice recordings at all. We’ll continue learning from customer feedback, and building privacy features on their behalf.”

For as long as the most impactful models remain too big for local hardware, this will be the reality of our Generative AI experience. Amazon is simply falling into line in preparation for Alexa+.

It's not great news, but also not the disaster and privacy and data safety nightmare it's made out to be.

You may also like

• Amazon’s new AI-powered Alexa+ is a lot more fun to talk to than ChatGPT or Siri, and I like it
• All your burning Alexa+ questions answered – by one of the people who built it
• Alexa Plus subscription pricing confirmed – and it's good ...

• Write-once memory card offers tamper-proof, long-term data storage
• TeamGroup’s 256GB D500R WORM SD card can store up to 374 CDs’ worth of data
• Built-in protection features guard against power loss, damage and tampering

CDs and DVDs are (or rather were) great for storage, but they can be affected by disc rot, a form of physical deterioration that affects optical discs, causing them to become unreadable over time due to corrosion or damage to the reflective layer. If you have a lot of important data stored on discs and are worried about losing it, TeamGroup has a solution to your fears – the D500R ISD WORM SD card.

Shown off at Embedded World 2025 (where it won the Embedded Vision category's top honor and the 2025 Community Choice Award), the D500R WORM card uses Write Once, Read Many technology for non-erasable, tamper-proof, long-term data retention storage.

The D500R ISD uses MLC NAND Flash, which offers better durability and steady performance, and promises read and write speeds of up to 70MB/s and 65MB/s. It supports capacities from 8GB to 256GB. That largest card is big enough to store data from around 374 CDs, so it can likely back up all your discs easily.

Seals the data in place

As you can tell from the description, the card can be read many times, but only written to once. Via a combination of hardware design and firmware control, the D500R ISD seals the data in place and ensures it can’t be deleted, overwritten, or updated. This makes it useful for users who require lasting data protection without the risk of accidental changes or even malicious tampering from ransomware.

The card includes features like Power Fail Management, which helps preserve data in the event of an unexpected power cut, and Bad Block Management, which detects and isolates faulty storage sectors, helping extend the product’s lifespan.

It also has a built-in S.M.A.R.T system that tracks usage and sends alerts if an attempt is made to alter or delete the data in some way.

Built to withstand demanding environments, it meets multiple military-grade shock and vibration standards and is rated IP58 for dust and water resistance.

TeamGroup suggests it could be useful for law enforcement footage, medical records, or financial archives, but it could be used for backing up family photos and videos or company data.

While it’s clearly a great solution for long-term archival storage, it’s probably worth noting that the D500R WORM card is a lot easier to lose than a box of CDs or DVDs, so you might want to think about where you’ll keep it. There’s also no word on pricing yet, but it's fair to assume it won’t be cheap.

You might also like

• These are the largest microSD cards you can buy right now
• Lexar's Armor SD steel cards are water resistant and can survive a 16-feet drop
• This is the fastest 2TB memory card ever launched and I can't wait to test it