Technology

Employees at the Cybersecurity and Infrastructure Security Agency tell WIRED they’re struggling to protect the US while the administration dismisses their colleagues and poisons their partnerships.

NymVPN is officially entering the VPN market with a big ambition – being the world's most secure VPN app.

NymVPN's Alpha version originally broke through in November 2023. Now, it aims to offer a fully decentralized virtual private network (VPN). Looking to protect against AI-driven surveillance and censorship, the service uses new mixnet technology to protect your metadata alongside the content of your communications.

As of today, March 13, 2025, NymVPN is available for iOS, Android, macOS, Linux, and desktop. TechRadar will join Nym’s in-house security expert, activist, and whistleblower, Chelsea Manning, and Nym CEO Harry Halpin, at an event in London tonight, where NymVPN will be officially unveiled. So, stay tuned for more exclusive information.

An "untraceable VPN"

"Nym was created with the core principle that secure, private, and uncensored access to the Internet is as fundamental and essential as access to clean water and electricity," said Halpin. "NymVPN marks the split between Big Tech’s exploitative control of the internet with a new vision – one where privacy is a right, not a privilege."

Halpin explains the technology behind NymVPN was built with the support of the original ‘cypherpunks’ and the world’s top cryptographers to defeat even government-level mass surveillance.

Early adopters of the NymVPN alpha and beta versions then helped to shape the service in countries plagued by strict censorship, while others leveraged its security features to prevent attacks on their cryptocurrency transactions.

NymVPN sets itself apart from some other best VPN apps on the market by seeking to be a truly "untraceable VPN" that protects users from metadata surveillance. Metadata refers to all the data attached to your communications that's not the content, such as device type, activity type, timestamps, and more.

This is crucial, said the provider, as the rise of AI-driven tracking has proven that hostile actors – those being cybercriminals or intelligence agencies – can "generate more value from metadata than analyzing specific user activity."

Based on the idea of mix networks proposed by the cryptographer David Chaum in the 80s, Chelsea Manning came up with the Mixnet concept independently while in prison for disclosing classified documents to non-profit media organization WikiLeaks. (Image credit: Nym Technology)

Nonetheless, NymVPN offers two modes to provide flexibility between protection and performance. These are:

• Fast (WireGuard). Better suited for causal browsing, streaming and sharing, this mode uses the AmneziaWG protocol, a highly censorship-resistant version of WireGuard developed by the team behind Amnezia VPN. Here, your internet connection is rerouted through two independent servers for maximum privacy and security.
• Anonymous (mixnet). What truly sets NymVPN apart from the competition. This mode claims to secure your metadata by using its noise-generating mixnet. This tech employs several network strategies to confuse data surveillance efforts (data fragmentation, dummy data packets, timing delays, and data packet shuffling), while rerouting your connections over five independent servers.

Anonymous by design

Not only does NymVPN claim to be better at keeping you anonymous when browsing the web or carrying out any other activities on the internet.

The provider also promises to never collect any of your personal information, going further than existing no-log VPN solutions in several ways.

For starters, the decentralized Nym network is built to prevent any single identity from linking the activities with a user. All the servers claim to be operated independently, so it should be impossible for Nym to trace traffic records.

You don't need to give up your email, name, or other personal details to sign up either. NymVPN has found a way to leverage the blockchain technology (the same tech used by cryptocurrencies, for example) to increase users' privacy and security.

So-called “zk-nyms” are zero-knowledge proofs that enable users to verify payment and account usage without revealing their identities.

Part of these anonymous proof of payments are then turned into NYM tokens to reward the independent contributors running the nodes worldwide. At the time of writing, you can choose among servers in 51 countries worldwide.

You can choose between a dark and light appearance, directly within the NymVPN app settings (Image credit: Future)

All in all, Chelsea Manning said: "Even in democratic nations, people are faced with unrestricted data collection, hyper-narrow algorithmic feeds, and normalized censorship tactics. NymVPN takes an infrastructure-based privacy approach to try and combat this increasingly uncertain and splintered internet."

You can now download NymVPN across all the major platforms, with pricing starting from a $5.49 monthly fee – even better as you can try it out for 30 days at no cost.

When property taxes and homeowners insurance premiums increase, you'll have a bigger bill.

Great data automation software creates value. It enables your business to work faster, more efficiently and more cost-effectively. At the other end of the scale, though, is software that bottlenecks progress and eats up resources.

Cost-cutting continues to dominate the agenda for Operations leaders. It’s therefore important to make sure that the data management, reconciliation or automation software you’re using falls into the former category and not the latter.

We explored some of the costs of inefficient software in our article on the true expense of supposedly ‘free’ recon tools. But even if you have paid for your on-premise system – do you really know how much it’s costing you?

Answering that question is difficult.

Calculating the total cost of ownership (TCO) for on-premise systems is easier said than done, because many of the costs are indirect, hidden, or part of the finance allocation process.

Licenses, maintenance and hardware are just the tip – there’s a lot more lurking below the surface. Today, we’re going to help you uncover them all, so you can build a transparent picture of the cost of managing data.

Here’s what else you need to factor into your calculations.

Unhelpful upgrades

With a cloud computing platform you effortlessly move from version to version to access new features and performance improvements. But software upgrades and patches are the only way of delivering innovation in an on-premise world – and they cost a lot of money. They are often mandatory and short notice; shorter, at least, than the time it would take you to shop around for an alternate system.

You often have no choice but to pay to move to the new version (we’ll explore why sticking with what you’ve got is anything but cost-effective later). But it may not bring any additional features and benefits that you need. It may introduce capabilities outside of your use cases, for instance, or perhaps support for a new asset class that you don’t trade.

One client told us that they recently finished upgrading their ten-year old on-premise system to the latest version. It took them three years and an estimated $5-$10m and they got no value or functional benefit out of it.

Another client – a smaller firm – budgets $1m every year for upgrades to their on-premise system.

I know from my own experience how costly and time consuming these types of migrations can be – absorbing focus that could be better spent elsewhere

Those costs go beyond just the licensing costs for the new system. Changing from one software version to another often means your IT team has to completely rebuild all your processes. Then they and the users have to test them to ensure they still work correctly and none of your integrations have been broken by the upgrade.

Again, it’s worth pointing out that you may not see any benefits from being on the new system.

This is change-the-bank money that you’re having to spend, but it’s not delivering any change. It’s the cost of business-as-usual in an on-premise world.

End-of-life – and beyond

But what do you do when you either can’t afford to upgrade, or rightly refuse to do so because of the unnecessary cost? In that situation, your software ends up being unsupported.

The vendor may offer you a premium support package, or you have to hire in extra IT resources just to keep the system working. You’ve just traded the cost of upgrading for the higher cost of keeping your existing software running.

By this point you have a zombie system – dead as far as the vendor is concerned, and with no spark of innovation left. It shambles along, doing as much of the job as it can manage, while you run around after it, picking up all the bits that have fallen off.

Are you factoring in the cost of life support for sunset systems into your on-premise TCO?

Inagility

The slow and laborious patch management and upgrade cycle means that innovation is a sluggish process in an on-premise world. Time-to-market for change can stretch to months or even years. Firms rely on their IT teams to install software updates, rebuild processes, release them into production environments, and so on.

This slow time-to-value makes Operations very inagile. This is an acute problem in today’s world. Regulatory change, for instance, is relentless and happening on multiple fronts. How many risks is this introducing to the business? How many opportunities are missed? These are all costs that must be considered.

There is a way around this for Operations teams, but it’s one that just introduces more cost and risk into the business…

Endless manual work

On-premise reconciliation systems exist to help you automate your data. But they often have the opposite effect. The lack of agility explored above forces Ops teams to resort to solving problems manually.

There’s often an army of ‘Human APIs’ surrounding on-premise systems – people whose job it is to manually plug automation gaps. They perform tasks such as:

• Extracting data from unstructured formats like emails and PDFs
• Keying data into systems in the correct schema
• Copying data into a spreadsheet to reconcile
• Manually matching (often thousands of) items where the system hasn’t spotted the connection

All this means that a lot of your resources in Operations are invested into low-value work. Your data experts are handling tedious and repetitive tasks instead of performing valuable activities such as root cause analysis, fixing data at source, or providing strategic insights for the business.

And that’s to say nothing of the cost of operational risk that manual work introduces to the business, which can result in direct costs such as regulatory fines or errors leading to trading losses.

Conclusion

The upfront license fees for on-premise software are just the start of a long tail of costs that snake their way through your operations. Your IT team is responsible not only for hardware and maintenance, but often for actually operating the system.

This cost only grows if the system is sunset and requires life support. Upgrading is costly and usually mandatory – regardless of how much value, if any, you get out of it. Long change management processes cost you in agility and slow time-to-value. Hundreds or thousands of workers are required to fill the functionality gaps, introducing risk to your business.

These systems are usually well-entrenched in your architecture, making them difficult to remove. But that’s in a world where you’re replacing one on-premise tool with another. Replatforming to an agile cloud platform is an entirely different affair; gradual, measured and low-risk.

We've listed the best IT asset management software.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

Whether you've just bought the best Apple Watch on the market, like the Apple Watch Ultra 2, or you've got your hands on the budget Apple Watch SE 2 model, opening your Apple Watch and putting it on for the first time is a pretty great experience.

You might have an idea of what your Apple Watch can do, or how you want to use it, but what about some of the settings and tweaks you can use to truly personalize your Apple Watch experience?

For as long as I've been a tech user I've been enthralled by customizing gadgets, whether it's the theme on Windows XP or the color of the menus on my Sony Ericsson K850i.

Having worn an Apple Watch for 10 years, I know a thing or two about settings you should definitely tweak on day one. Here are my go-tos.

1. Your Apple Watch face

(Image credit: Future)

If you're new to Apple Watch, the very first thing you should do on your new one is change the watch face. You won't find third-party options like you might on Wear OS or Garmin, but there are plenty of great Apple options to choose from and they're very customizable.

To change your watch face simply long press on the center of your Apple Watch display until it vibrates. Then simply swipe through the options until you see one you like.

If you tap Edit you can tweak aspects of the face including colors, and the Complications (small nuggets of glanceable information) that show up.

2. Silent mode

Your Apple Watch is designed to sit pretty discreetly on your wrist, and one of the joys of using one is that you can get notifications without too much disturbance, either for yourself or others.

Obviously, if you have accessibility needs, you might want to leave the Apple Watch notification sounds on. For me, however, this is the first toggle I switch off. The watch gently vibrates whenever I get a notification (if I have them turned on at all), and I definitely don't need that piercing ding telling everyone in the vicinity that I've just got a notification.

Simply press your Apple Watch's side button and select the bell icon on the left to silence your Apple Watch.

3. Return to clock

(Image credit: Future)

Apple Watch is pretty easy to navigate, and moving between apps and the watch face is fairly straightforward. However, another setting I like to tweak is the Return to Clock setting. This takes your Apple Watch back to that smashing watch face you've just customized automatically once you put your wrist down.

By default, it's set to a two-minute delay, but I like to select Always, so that I can drop my wrist at any time and know my Apple Watch is displaying the time, rather than sensitive information or an app where I might accidentally pause a workout or change something unknowingly.

Simply press the Digital Crown, scroll to Settings, select General, and then Return to Clock to toggle this.

4. Siri

Siri isn't particularly helpful these days, and lags behind rival virtual assistants from the likes of Google and Android, not to mention more advanced AI offerings that overshadow the delayed Apple Intelligence upgrade it badly needs.

To that end, I always turn Siri off on my Apple Watch to some degree. In the Siri section of the Settings app on your Apple Watch, you can tweak how and when Siri responds to your inputs.

I actually prefer to select Listen for to "Hey Siri," that way it's much more difficult to trigger accidentally. I also tend to leave Raise to Speak on, and turn off Press Digital Crown.

I can't count the amount of times I used to accidentally activate Siri during workouts by pressing the Digital Crown, so this works well for me. I can access Siri by very deliberately raising my wrist, but never trigger it accidentally with a button press.

However, you might want to consider the opposite, leaving the Crown on and turning off Raise to Speak. Or you can turn Siri off on your Watch altogether to avoid it getting in your way.

5. Screenshots

A more recent addition to my Settings repertoire is screenshots. You can take a screenshot on Apple Watch just like iPhone, by pressing both buttons on the side at the same time.

I can't think of any scenario whereby taking a screenshot on the Apple Watch would be necessary. However, as with Siri on the Digital Crown, it's very easy to do accidentally, especially during workouts. As such, I turn screenshots off in the General section of Settings, to avoid my Photos library getting clocked up with random snaps of my watch face.

You may also like

• New figures claim the smartwatch market just shrunk for the first time ever, and the Apple Watch Ultra 3 is to blame
• In 10 years, this $99 Amazfit Active 2 is the closest I've come to ditching my Apple Watch
• I've spent 10 years with Apple Watch, and after 10 minutes with the Galaxy Watch Ultra, I can't believe Samsung skipped this feature

The mRNA technology behind coronavirus vaccines is now being used to create bespoke vaccines for cancer patients.

Consumers today expect faster, more personalized services in the insurance industry. To meet these demands, insurers are increasingly adopting AI-powered solutions to enhance efficiency and customer experience. Research from McKinsey reveals that Generative AI alone is projected to contribute up to $4.4 trillion to the global economy annually, with insurers expecting gains in productivity, premium growth and underwriting accuracy. As a result, a third of major insurers already have Gen AI use cases in production, whilst most are advancing their data capabilities.

From AI chatbots managing claims in real time to data-powered insights enabling personalized policies, AI is revolutionizing insurance. Insurtech startups are leading this transformation, integrating automation and blockchain to improve efficiency and security. However, this digital shift also raises concerns about cybersecurity and algorithmic bias. Striking the right balance between innovation, fairness and security will be key to fostering trust and ensuring AI-driven insurance solutions help drive long-term success.

Speeding up claims resolution with AI

AI is changing how insurance companies engage with customers. With AI chatbots, insurers are now able to anticipate customer enquiries and manage claims proactively. Digital assistants can provide real-time support, streamlining the resolution process and reducing wait times, which enhances overall user experience.

For instance, after a vehicle collision, policyholders can upload images via their smartphones, enabling AI algorithms to instantly assess damage and estimate repair costs. This accelerates claim approvals, improves transparency and reduces disputes.

Personalizing policies with data-driven insights

Modern consumers expect personalized services that align with their lifestyles and behaviors. As such, insurers are now using data-driven insights to create customized policies. In particular, insurance companies are leveraging geolocation, edge computing and IoT devices to collect data on individual risk profiles and then transform this raw data into actionable insights, unlocking AI’s full potential. For example, real-time driving behavior analysis allows insurers to offer lower premiums to cautious drivers whilst adjusting rates for riskier behaviors.

However, whilst the use of AI in insurance can help differentiate between low- and high-risk drivers, insurers must ensure these systems do not lead to unfair discrimination or exclusion. AI models should be carefully designed to avoid biases, especially those that may unintentionally factor in race or ethnicity, to promote fairness and equity in insurance practices.

Mitigating AI-enabled fraud and strengthening cybersecurity

Whilst AI enhances efficiency, it also introduces challenges, particularly in fraud detection. Fraudsters are increasingly using deepfake technology and synthetic identity fraud, making traditional detection methods insufficient. To counter these threats, insurers are deploying AI-driven fraud detection systems that analyze behavioral patterns and detect malicious activity in real time. This way, insurance companies can identify vulnerabilities and prevent potential data breaches before they occur, which significantly improves cybersecurity resilience.

Beyond fraud, the shift towards digital-first insurance models raises cybersecurity concerns. With access to vast amounts of sensitive customer data, insurers are prime targets for cyberattacks. Implementing advanced encryption, biometric authentication and AI-powered threat monitoring is no longer optional – it is essential.

Insurers must also address consumer skepticism about AI’s role in insurance by prioritizing transparency. Clearly communicating the benefits of AI-powered personalization whilst implementing strong data protection measures is essential. In particular, compliance with data privacy laws is key to securing AI-driven insurance models and fostering consumer trust. Additionally, establishing AI ethics committees and consumer education initiatives can further promote trust and acceptance of AI in the insurance industry.

Leading the future with responsible AI integration

The insurance industry must continue evolving to harness the full potential of AI. From intelligent automation to data-driven personalization and advanced fraud detection, AI is transforming operations, enhancing customer experiences and driving competitive advantage. Those who strategically integrate AI will lead the future, whilst others risk falling behind.

However, success requires more than just innovation – it demands responsibility. Insurers must strike a careful balance between leveraging AI’s capabilities and ensuring transparency, fairness and security. Adaptability will also be key, as AI models must continuously evolve to address emerging risks and evolving market demands. By embracing AI with a customer-centric approach, insurers can build a more secure, efficient and trustworthy industry – one that goes above and beyond to meet consumers’ evolving expectations in today’s digital era.

We've showcased the best identity theft protection.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

If you're looking for fast and reliable internet, both cable and fiber can deliver the goods. But which one should you choose?

iRobot just scrapped and replaced almost its entire range of robot vacuums, but amongst the shiny new fleet there's one model that has caught my eye in particular: the Roomba 205 DustCompactor Combo. This bot is designed especially for smaller homes, and as the owner of such a compact dwelling, I am particularly keen to get my hands on one.

The fact I live all on one floor means robot vacuums are, in some ways, a great fit for my home. Apartment living often translates to 'not much space,' and that's certainly the case for me. Many of today's best robot vacuums come with gigantic docks that can do everything from emptying the dust to refilling the water tanks to washing and drying the mop pads. While they certainly make for a very hands-off experience, I lose half of my kitchen every time I test one.

The other option is a charge-only dock. These tend to be nice and petite but require me to take care of all maintenance – including bin emptying. The bins you'll find onboard robot vacuums are necessarily small, so if you have a lot of dust, you'll do that with irritating regularity. If testing one of these coincides with me having a canine visitor, bin emptying becomes a chore; this definitely isn't an ideal option if you're looking for the best robot vacuum for pet hair.

(Image credit: iRobot)

iRobot's solution is to introduce dust compacting into the 205 Combo's onboard bin, which can hold more dust and thus requires less frequent manual intervention. The dock is the basic, charge-only one that's small enough to fit unobtrusively in almost any home, including mine.

Inside the small onboard bin is a paddle that twists left and right to squish any dust and debris that has been sucked up into a smaller space. The paddle then returns to the center of the main inlet, ready for the next influx of dirt.

The 205 Combo is designed to last up to 60 days without emptying, although that will depend on how much dirt and dust you're sucking up, of course.

(Image credit: iRobot)

iRobot says its compacting technology is an 'industry-first'. Another model on the market – the Narwal Freo X Plus – does a similar thing, although I guess it might not use the same mechanism.

Get low

Another notable feature here is that iRobot has removed the navigation puck from the top of the Roomba 205 DustCompactor Combo, making it shallow enough to sneak under furniture that might otherwise have been off-limits. Perhaps iRobot decided that people with small homes must also have low furniture; I don't know.

This is particularly interesting because this robovac uses LiDAR for navigation. In fact, the introduction of 'ClearView' LiDAR navigation is one of the best updates to feature in the new Roomba range.

Previously, the brand used an older form of robot vacuum navigation, which was far slower, less reliable, and unable to operate in the dark.

(Image credit: iRobot)

iRobot says the LiDAR technology has simply been shifted into the front of the 205 DustCompactor model to give it a more streamlined silhouette. The thing is, robots that use LiDAR need that raised puck to be able to 'see' a nice wide area around them. All three of the other new Roombas have one.

I asked an iRobot representative about this, who explained that although the 205 would have a narrower field of view, this wouldn't impact navigation accuracy for most people. She reassured me that she was testing two of the new Roomba bots, one with the puck and one without, and they were both behaving the same way.

If that holds true, this could be the perfect robot vacuum for me. We have a Roomba 205 DustCompactor on its way and report back as soon as we've had a chance to put it through its paces.

The Roomba 205 DustCompactor is due to go on sale from 18 March, with a list price of $469 / €449 (around £380 / AU$780).

You might also like...

• Dyson vacuums have one big problem and I don't understand why
• Tiny feet, bionic arms, and bots that can deliver a sandwich – 6 weird and wonderful robot vacuum innovations from CES 2025
• Browse the best budget robot vacuums 2025

Save 74% on 2-year plans and get 3 free months with our NordVPN discount codes.

Save up to 50% on business and personal memberships with a 1Password promo code.

Here are the answers for The New York Times Mini Crossword for March 13.

Though I experiment with AI chatbots professionally, I've found many ways to incorporate them into my regular life, including as a parent to a lively baby and uncle to several nieces and nephews.

ChatGPT has been a surprisingly big help. It's not a babysitter, but it can be a real boon for trying new ways to entertain kids or just keep juggling all of the many simultaneous parenting roles without dripping any balls. It may even be able to provide the rarest of all commodities as a parent – free time.

Bedtime stories

(Image credit: Images created by DALL-E 3)

Every parent or caregiver has been here: it’s bedtime, and you're exhausted. You've somehow read every book you've ever bought for your kid all in one day, and you want to give them something new, but even your imagination feels depleted. If only you had a ready story, maybe something about a robot who loves pancakes?

ChatGPT can take vague ideas or random words and make something surprisingly entertaining from them. Try asking it: “Can you tell me a bedtime story about a dragon who loves tacos and befriends a unicorn who dislikes rainbows?” or "Create a bedtime story about a gentle giant who learns to make friends despite everyone being afraid of his size."

Jukebox lyricist

(Image credit: Images created by DALL-E 3)

My son loves music, and I love introducing him to new songs or making up new ones. However, Improv lyrics don't always cut it, especially if I'm entertaining his older cousins.

Rather than listen to "Baby Shark" and "Wheels on the Bus" on repeat, I use ChatGPT to bring new life to old tunes. I've asked the AI everything from "Write me a song about a dancing giraffe named Gerald who loves ice cream to the tune of Twinkle, Twinkle, LIttle Star," to "Give me lyrics to a song about a puppy who dreams of going to space to the tune of Hush Little Baby." And had pretty solid results, as you can see below.

"Oh little pup, looking at stars above
Dreaming of rockets, and moonbeams you love
Wearing your helmet, wagging your tail
Puppy astronaut ready to sail!"

Snack time

(Image credit: Images created by DALL-E 3)

Snack time with babies and toddlers can be a minefield with attempts to make things that are healthy but that they'll like. Trying to do it while also keeping an eye on the children is even more challenging.

Luckily, you can combine the two by asking ChatGPT for ideas. in the vein of "What are healthy snack ideas for children aged one, three, and five that they can also help make." The banana sushi rolls (banana slices rolled in granola), apple slices with peanut butter and cinnamon, and make-your-own berry yogurt parfaits have all been massive hits.

DIY games

(Image credit: Images created by DALL-E 3)

On rainy days, keeping kids entertained indoors can feel like taming wild animals. They were even bored with Bluey, which is a real warning sign. I turned to ChatGPT and asked for a game that didn't require screens, glitter, or immense patience.

ChatGPT delivered beautifully. "Come up with a simple game for young kids that doesn't require much setup or clean up," I asked. The answer was "Sock Toss Basketball," which employs laundry baskets and balled-up socks as a fun activity.

You could also try something like "What's a quick indoor activity for kids using only paper and markers?" and get a suggestion of making indoor treasure maps that the kids can populate with their favorite toys and treats.

Better boo-boos

(Image credit: Images created by DALL-E 3)

Kids get scrapes and sniffles; that's just inevitable. Keeping their spirits up and making the road to recovery bearable (for both of you) can be tricky, but I've found ChatGPT to have a lot of ideas.

Ask it, "How can I get my young child to take their medicine without complaining," and you'll get ideas about making a whole production for the "magic potion" or suggestions on incorporating it into the bedtime rituals they already like, maybe even embedding it in a bedtime story.

The same goes for applying a bandage or even sunscreen. You have a lot on your plate as a parent. It's not lazy to look for ideas from parenting forums or books, and ChatGPT is just another helpful tool that you can call on for very specific advice any time.

You might also like

• I tried Deep Research on ChatGPT, and it’s like a super smart but slightly absent-minded librarian from a children’s book
• How ChatGPT helps me keep my kids entertained over the holiday period and into 2025
• ChatGPT 4.5 understands subtext, but it doesn't feel like an enormous leap from ChatGPT-4o

The web's inventor says companies worked together decades ago to create an open internet. That isn't happening with generative AI.

Hints and answers for the NYT Connections: Sports Edition puzzle, No. 171, for March 13.

• Gemini Robotics is a new model
• It focuses on the physical world and will be used by robots
• It's visual, interactive, and general

Google Gemini is good at many things that happen inside a screen, including generative text and images. Still, the latest model, Google Robotics, is a vision language action model that moves the generative AI into the physical world and could substantially speed up the humanoid robot revolution race.

Gemini Robotics, which Google's DeepMind unveiled on Wednesday, improves Gemini's abilities in three key areas:

• Dexterity
• Interactivity
• Generalization

Each of these three aspects significantly impacts the success of robotics in the workplace and unknown environments.

Generalization allows a robot to take Gemini's vast knowledge about the world and things, apply it to new situations, and accomplish tasks on which it's never been trained. In one video, researchers show a pair of robot arms controlled by Gemini Robotics, a table-top basketball game, and ask it to "slam dunk the basketball."

Even though the robot hadn't seen the game before, it picked up the small orange ball and stuffed it through the plastic net.

Google Gemini Robotics also makes robots more interactive and able to respond not only to changing verbal assignments but also to unpredictable conditions.

In another video, researchers asked the robot to put grapes in a bowl with bananas, but then they moved the bowl around while the robot arm adjusted and still managed to put the grapes in a bowl.

Google also demonstrated the robot's dextrous capabilities, which let it tackle things like playing tic-tac-toe on a wooden board, erasing a whiteboard, and folding paper into origami.

Instead of hours of training on each task, the robots respond to near-constant natural language instructions and perform the tasks without guidance. It's impressive to watch.

Naturally, adding AI to robotics is not new.

Last year, OpenAI partnered up with Figure AI to develop a humanoid robot that can work out tasks based on verbal instructions. As with Gemini Robotics, Figure 01's visual language model works with the OpenAI speech model to engage in back-and-forth conversations about tasks and changing priorities.

In the demo, the humanoid robot stands before dishes and a drainer. It's asked about what it sees, which it lists, but then the interlocutor changes tasks and asks for something to eat. Without missing a beat, the robot picks up an Apple and hands it to him.

(Image credit: Google)

While most of what Google showed in the videos was disembodied robot arms and hands working through a wide range of physical tasks, there are grander plans. Google is partnering with Apptroniks to add the new model to its Apollo humanoid Robot.

Google will connect the dots with additional programming, a new advanced visual language model called Gemini Robotics-ER (embodied reasoning).

Gemini Robotics-ER will enhance robotics spatial reasoning and should help robot developers connect the models to existing controllers.

Again, this should improve on-the-fly reasoning and make it possible for the robots to quickly figure out how to grasp and use unfamiliar objects. Google calls Gemini Rotbotics ER an end-to-end solution and claims it "can perform all the steps necessary to control a robot right out of the box, including perception, state estimation, spatial understanding, planning and code generation."

Google is providing Gemini robotics -ER model to several business- and research-focused robotics firms, including Boston Dynamics (makers of Atlas), Agile Robots, and Agility Robots.

All-in-all, it's a potential boon for humanoid robotics developers. However, since most of these robots are designed for factories or still in the laboratory, it may be some time before you have a Gemini-enhanced robot in your home.

You might also like

• Apple might be building a humanoid robot, and I truly hope ...
• Nvidia unveils a fleet manager for warehouse robots
• Robot abuse is still the path to the future – even if I hate it

The 2025 iteration of the Mac Studio marks a new era in personal computing for Apple as it pushes the boundaries of what can be achieved, using what is essentially a bigger version of Intel’s now-defunct next unit of computing (NUC).

I pitted its full-fat, top-end version, the M3 Ultra with a 16TB SSD and 512GB of RAM against the competition to get a rough idea of how good a bargain it is. Unsurprisingly, it easily competes against some of the best workstation PCs in the world.

There’s a trio of ThreadRipper rigs, seven Xeon workstations (across three generations) and for good measure, Nvidia’s magnificent Digits and three of Apple’s very own PCs, the Mac Mini (M4 Pro), the Mac Pro (M2 Ultra) and a less capable version of the Mac Studio (M4 Max).

The entire table is at the end of this article. Note that some cells are empty because I simply don’t have the information at hand (e.g. size of some workstations, performance of Digits etc).

Credit where credit is due, Apple managed to deliver something I was not expecting it to do anytime soon: 512GB of unified memory in a computer.

That stroke of genius meant moving the comparison to Intel Xeon and AMD ThreadRipper Pro systems, far more expensive professional platforms compared to AMD Ryzen and Intel Core ones.

As well as zeroing on the workstation vertical, it also firmly signals Apple’s ambitions to be a major player in the red-hot local AI hardware market.

(Worth checking out my article published in June 2023 where I suggested that the Mac Studio would replace the Mac Pro (and make PC vendors nervous).

I analysed the prices of these systems (using the most up to date data I could find) and there’s no two way of putting it.

A $14,000 bargain if you’re after memory

The Apple Mac Studio Ultra, even at more than $14,000 is a bargain if you’re looking for as much system memory as possible, at $28 per GB.

That’s a bit less than Puget Systems and almost a third of what you’d pay for a maxed-out Mac Mini (or half the Apple Mac Pro with its puny 192GB RAM).

Most importantly, it is far, far cheaper than the likes of Dell or HP; turns out Windows workstations with half a terabyte RAM, don’t come cheap.

Remember that Apple uses unified memory which has a much higher bandwidth (800GB/s) than DDR5 memory (even across eight memory modules).

It delivers the type of performance you’d expect from an integrated GPU; as Servethehome’s Patrick Kennedy puts it, “The cost is high in the context of a workstation, but it is cheap when you compare it to NVIDIA’s cost per GB of memory”.

Remember as well that I have configured this workstation with a 16TB onboard storage; no other major player has access to that sort of storage unless you resort to RAID-0 across multiple much smaller SSDs.

There’s also the fact that it comes with 10GbE LAN by default as well as Thunderbolt 5, a data transfer technology not available to any other workstation at the time of writing, which, in due time, will pave the way for innovative external peripherals.

I also looked at 3 other composite metrics, all looking at processor performance (based on the popular Geekbench 6*) and not surprisingly, Apple excelled in all three: GeekBench point per dollar, Geekbench point per core and Geekbench point per volume.

The M4 devices (Mac Mini and Mac Studio) had the cheapest performance per $, at about $0.20, that’s almost 1/8th of the most expensive one (the HP G4).

It also had the highest performance per core (1623, 5X faster than the slowest processor, an Intel Xeon) and the highest performance per unit volume (more than 140x compared to the worst entry).

GPU details remain a mystery

The big unknown remains the official GPU metrics as it is difficult to gauge Apple’s graphics performance across different systems-on-chip as Apple doesn’t divulge additional data beyond core numbers.

For the sake of simplicity, I chose the Nvidia RTX 2000 Ada Generation as the default card - where possible - as it was the cheapest professional graphics card that came with 16GB of RAM. Only one entry used the more expensive RTX 4000 Ada Generation with 20GB of RAM.

Tech publication, TechPowerUp, estimates that the RTX 2000 AG should be slightly slower than an RTX 3060.

The Apple M3 Ultra scores just under 130,000 marks on the popular OpenCL Geekbench tests, that’s 50% faster than the RTX 2000 AG, so for a more balanced comparison, readers may want to substitute a more powerful (and expensive) card like the NVIDIA RTX 5000 Ada Generation.

* Geekbench 6 entries are sadly not averaged. So I picked what I considered to be the median on the first page of each CPU. Your mileage may vary.

You might also like

• Future exascale supercomputers will be built using these HPE Lego-like cabinets able to hold nearly 100,000 AMD EPYC Cores
• First AMD EPYC 9965 benchmarks are in; Turin server CPU sets flurry of new world records across wide range of software, but will that be enough to stop Xeon and Arm's forthcoming onslaught?
• Want to put 24TB of RAM in your PC? This motherboard can, and will even accommodate 384 AMD EPYC Turin cores and almost 1 petabyte storage

Fight with giant mechs or your bare hands in these games.

• Disney has released the first trailer for the live-action Lilo & Stitch
• It's our first peek at the entire cast, including Maia Kealoha as Lilo
• The film will make its theatrical debut on May 23, 2025.

We can finally say, ‘Aloha!’ to our first real look at the live-action Lilo & Stitch, thanks to the official trailer. And let’s just say as someone who loves Stitch – including the original, ‘Stitch has a Glitch,’ and the TV series – Lilo & Stitch: The Series – it’s a journey.

It’s pretty packed, displaying all sorts of emotions as we’ve come to expect from the main character, yet the two-minute and 24-second trailer does fly by. It also provides a first look at most of the cast, including Maia Kealoha as Lilo, Sydney Agudong as Nani, Billy Magnussen as Pleakley, and Zach Galifianakis as Jumba. Of course, we also hear plenty of Chris Sanders reprising his role as Stitch.

The trailer starts with the apprehension of Stitch – aka a ‘dangerous experiment’ – but it quickly falls off the rails with a glimpse of the iconic alien in a red starship. Then, we flash to Hawaii and see Lilo spotting a shooting star and wishing for a ‘best friend.’ Quick enough after that, we hear ‘fluffy,’ see Nani, the sounds of Elvis come through, and chaos err, fun ensues.

You can watch it for yourself below, and it’s our look at the film and the whole cast ahead of the release on May 23, 2025.

Lilo & Stitch is the latest live-action remake of an iconic Disney film. Snow White, another live-action remake, is set to release on March 21, 2025. It’s been a trend for Disney, which also did a live-action Beauty and the Beast – and you can now build the iconic castle from that franchise out of Lego – and The Lion King.

With Lilo & Stitch, though, the hype is here, and this could wind up being the best live-action remake from Disney as of yet. Especially with the original voice of Stitch – Chris Sanders – reprising his role and doing a lovely job, at least from the trailer so far.

The trailer also provides a first – if only super quick – at Pleakly, and Billy Magnussen, as well as Zach Galifianakis, but no signs of an animated Jumba as of yet. We also see Cobra Bubbles, played by Courtney B. Vance, and in the name of fan service – which this trailer does a lot of – we saw ice cream/a snow cone fall on the ground.

(Image credit: Disney)

All in all, the official trailer has leveled up my excitement for Lilo & Stitch, and if you weren’t sold as of yet, I think this will do the trick. In the meantime, I will head to one of the best streaming services – Disney+ – to rewatch the original, the sequel, and some other Stitch content.

And if you're looking to pass the time in the lead-up to the film, we've seen a trove of new Lilo & Stitch toys and collectibles. Here are a few of our favorites.

From 730 pieces, you'll build your own Lego version of stitch, which has the character enjoying an ice cream cone and rocking a a flower on his head. Also available at Lego for $64.99.View Deal

This high-tech version of Stitch stands over 18 inches tall and produces over 100 sounds and movements.View Deal

At 152 pieces, you'll build a compact version of Stitch in the Lego Brickheadz style.View Deal

Lilo & Stitch will hit theaters on May 23, 2025.

You might also like

• New Disney+ movies: every new film to stream in March 2025, including Moana 2
• Moana 2 review: Disney's anticipated sequel is a real tearjerker, but I wish we'd seen more from my favorite character
• Discovery+ just got a big update to its streaming app that makes it more like Max – here are 5 great new features to try

Try light therapy if you're fighting fatigue from the time change or struggling with seasonal depression. Here's how it works.