TechRadar News

I don’t like ChatGPT’s latest voice model. If you don’t use voice to interact with ChatGPT, this might seem like a strange thing to get worked up about, so let me explain.

Until recently, its voices have all sounded fairly advanced. But there was still something unmistakably AI-y about them. At times, they sounded smooth and even expressive, but also stunted and, well, robotic.

The latest voice update attempts to smooth away the rest of those more robotic edges. Which means that ChatGPT now pauses, hesitates, changes intonation and adds lots of little noises humans make without even thinking about it.

There are lots of “umms” and “hmms”. There are plenty of drawn out “yeeeahs”, as well as enthusiastic “ooohs”. Sometimes it sounds like it’s taking a breath or sighing. Sometimes it pauses halfway through a thought. And often its voice rises at the end of a sentence. In other words, it’s been changed to sound more human and I find it incredibly annoying.

When I first tried it, I thought it was broken. Especially because of the way it said a filler word then paused. Once I’d heard three long “hmmms” in a row, I couldn’t pay attention to anything else it was saying in response. So I decided to turn my irritation into an experiment. I opened the different voices in ChatGPT, started talking and counted all of the filler words.

The filler word experiment

I started with Maple, one of ChatGPT’s more cheerful, female-sounding voices, and within a conversation that lasted about 5 minutes I counted 20 “hmmms”. There were also lots of elongated “yeeeahs”, a couple of “uhhhhs”, an “ooooh”, a “hmm nice” and several enthusiastic “yeah, totally” responses.

There was also a lot of rising intonation, which is when the pitch of the voice goes up at the end of a sentence. That made me wonder whether I was just getting hung up on some speech quirks that were more associated with US English than ChatGPT, so I switched voices.

Next, I tried Vale. It’s a bright and inquisitive British female voice. Interestingly, Vale had a different set of verbal habits. There were fewer “hmmms”, but plenty of “ohhh yeeeahs”, long “suuuures” and a few “riiiights”. And its intonation also seemed to rise constantly, particularly when it was asking me questions.

I thought I’d try a more obviously male-sounding voice next. I chose Arbor, another British voice, which I found calmer but much more stilted. It gave me way less of those exaggerated conversational noises that I disliked in Maple, but replaced them with strange pauses and a staccato rhythm. At times it seemed to stop talking in really odd places.

After around 20 minutes of conversations across the different voices, I'd counted more than 100 filler words and conversational noises.

(Image credit: Getty Images / VCG / d3sign)Falling for the filler words

I’d gone into these conversations with ChatGPT looking for things that irritated me. I was listening closely for every “umm”, strange pause and exaggerated “yeeeah”. I was intentionally trying to count all the quirks OpenAI had added to ChatGPT’s voice to make it sound more like a person.

But, weirdly, the conversation started to feel different. I reluctantly began talking about priorities and my week ahead to test the voices. And instead of giving ChatGPT prompts and waiting for its responses, I soon found myself slipping into a much more natural conversation. This surprised me because I still didn’t like the voices.

The filler words remained distracting and the pauses were still in all the wrong places. At no point was I convinced that there was a person on the other side of the conversation. But I had to admit that although on a conscious level I am wary about AI, and the whole point of this was to scrutinize the system, I still seemed to be responding to its more natural-sounding voice by getting increasingly wrapped up in the conversation.

Whether you find that cool or worrying probably comes down to your broader opinions about AI and its influence on us.

Why does AI need to say “hmmm”?

There’s an obvious reason for making ChatGPT sound more natural, which is that humans find natural conversation easier, and are therefore more likely to continue engaging with it.

Real, everyday speech isn’t a clean stream of information. All of us pause, hesitate, change pitch, make noises to show someone we’re listening and use filler words while we’re thinking about what to say next. These small but significant parts of speech do a lot of work.

So what are they doing when AI uses them? Because ChatGPT doesn’t need to say “hmm” when it’s gathering its thoughts or “yeeeeah” to show you it’s listening. It could just retrieve the information and respond. So these sounds are for making interactions feel more conversational, which is what makes me wary.

AI companies frequently position their AI systems as tools that can help us work, learn, create and get things done. When people like me worry about us all becoming too reliant on AI, we're often told that what matters is how we use the tool. But this is just one example of how the tool itself seems designed to behave less like a tool every day.

A calculator doesn’t need to add in a filler word before it gives you an answer and a word processor doesn’t need to pause and say “hmmm, yeah, totally!” when you start typing. But that’s what ChatGPT is now doing.

Some might argue that a voice with a more natural-sounding rhythm and intonation is simply just a nicer experience. But those same choices make interactions feel more social, potentially blurring the line between the role of AI as a tool and something we relate to more like another person.

After my experience, I wasn’t surprised to find research suggesting there might be more to this. Research suggests that voice-based interactions with AI might be associated with increase emotional engagement and anthropomorphism.

There’s also some suggestion that voice could lead people to overestimate AI, especially when a human-like voice is mistaken for understanding or competence.

This doesn’t automatically mean that adding some “umms” will make us all trust AI. But it seems that adding voice into the mix does change how we perceive and respond to it.

(Image credit: Getty Images)Annoying but effective

One more practical learning from this experiment was that if ChatGPT's voice suddenly drives you mad after the update, try another one. The differences between them felt bigger than I expected. Initially, I couldn't stand Maple's mannerisms and rising intonation, but Arbor's calmer delivery was easier for me to tolerate, even if its strange pauses created a different pacing problem.

I went into this experiment expecting to write about an annoying voice update. I thought I'd count some “umms”, complain about how artificial all of this manufactured naturalness sounded and then just return to typing. Instead, I spent 20 minutes deliberately studying the mechanisms designed to make AI sound more conversational and still found myself becoming more conversational with it.

I knew the hesitation wasn’t genuine hesitation. I knew there wasn’t anyone on the other side of the exchange thinking, listening and searching for the right words. Yet I still responded to the signals by talking more naturally and, eventually, being more open.

I don't think that means I was somehow fooled into believing ChatGPT was human. But I think what it suggests is that we don’t have to believe AI is conscious, or there’s some form of life there, for our deeply ingrained social instincts to start responding to human-like cues.

So, while I still wince at every “hmmm”, I don’t think whether I like the filler words or not really matters. It’s more about what they make us do and how surprisingly instinctive my response to them was.

The xTool M2 is priced at the entry level of the company’s fully enclosed machines, but as ever there’s no mistaking it for anything other than an xTool laser engraver. As I got into using the machine, I could see just how finely tuned it is. Whilst it’s lighter weight than many of its more expensive siblings, the laser engraving power, even from the 10W laser in my review sample, was exceptionally good.

Having reviewed these laser engravers for a few years now, there are several essentials I look for in any laser engraver for TechRadar Pro, and one top priority is that it's fully enclosed. While I might’ve looked at fully open designs earlier on, these days there’s really no excuse for an open-frame design other than cost. When it comes to running a business, safety is obviously paramount, so it makes sense to look for a fully enclosed rather than open-frame design.

Beyond the physical design, I've found a few extras that really help with the review process, simplifying the cutting, engraving, and placement of materials such as cameras. Then there are added features, such as air assist, that are essential for creating clean, crisp lines and reducing scorching.

What’s surprising for a machine of this price is that the M2 ticks all the boxes. It has dual cameras, one that gives you a good view of the entire workspace and a second that gives you a close-up view, air assistance built in, and, as ever, there’s a vent at the back that you can attach to your air filtration system.

But it goes even further by adding more head module options, so you can not only laser cut and engrave but also print and cut. Of course, coming from xTool, it also has a range of accessories that can further expand what you can do with this machine.

I’ve had the M2 for a couple of months and have really been able to give it a more extensive test than I usually would with these machines, partially because I’m interested in getting some of these for use within an educational establishment, and making sure that it’s easy, secure and robust enough is a major focus.

I’ve also been asked by several friends who have been looking at laser engraving machines and which option they should go for. At this price point and with the 610 × 569mm dimensions, it just, on paper at least, looks like a great option.

Sure enough, while I only have the 10W laser model to review in this test, I have looked at previous XTool machines with multiple module heads, and they’ve worked exceptionally well; I will take a look at the print and cutting heads later.

Initial setup was easy, and after I’d run through the unpacking, installing the module head and powering up, it was then just a case of connecting the USB cable directly to my MacBook Pro, installing xTool Studio, waiting for it to connect, allowing the firmware to update, and it was time to start.

The one slight oddity with the modular head is that it installs magnetically, although it requires a connecting cable for data from the head to the machine, which I found a little odd. Why this isn’t wireless or just uses contacts, I have no idea; possibly just to keep costs down, but once that cable is connected, getting into the utilisation of the machine is all pretty fast.

(Image credit: Alastair Jennings)

What I like is that xTool has now introduced a safety video at the beginning, with a longer version that you can view online, and a really good tutorial which essentially enables you to get started with cutting projects such as placemats. You can see the results of this project in the pictures.

Those initial tutorials are essentially a no-brainer setup guide to using the M2, and when you run through those initial projects, which will take you less than an hour, you’ll be pretty proficient with how the machine works.

As ever, in the software you’re able to assign different parts to different cutting, engraving and scoring methods with a simple click; you highlight a path, enabling you to then assign it to whichever type of cut you want. Again, with xTool’s knowledge in the sector, this is all incredibly well thought out.

Essentially, over my two months with the machine, it sat in the corner of the workshop, attached to one of the xTool air filters, and ran seamlessly for every project I’ve thrown at it.

The workflow, from the ability to recognise the materials through to being able to preview so that you can accurately place your designs, and through to the cutting and extraction, all just works incredibly easily.

At present, if you’re looking for a new, fast, relatively cheap professional-level laser engraver for a small business, then you really can’t go wrong with the superb M2.

xTool M2: Price & availability

At present, the xTool M2 is available for $599 / £599 direct from xTool with the 10W laser. I've also often seen promotional offers on this price, so just keep an eye out.

If you’re looking for the colour print-and-cut bundle, which lets you make stickers and other projects, it's available for $749 / £749. There’s also the deluxe bundle, which includes everything, available for $989 / £989.

The laser cutter and engraver is also widely available from online retailers like Amazon.

xTool M2: DesignSpecs

Connectivity: USB computer connection confirmed; xTool software required.
Audio: Status buzzer
Camera: 5MP close-view camera + 16MP distance-view camera.
Size: 610 × 569 × 180mm
Weight: 12.3kg
Software: xTool software / xTool Studio ecosystem
Accessories: Depending on the bundle: laser module, CMYK inkjet module, RA3 Lite rotary system and other accessories.
Laser: 10W or 20W blue diode; optional 3W infrared module.
Laser work area: 426 × 320mm for 10W/20W modules.
Maximum XY speed: 600mm/s.
Maximum cutting claim: Up to 8mm basswood with 10W; up to 10mm with 20W.
Safety: Class 1 enclosed laser system.

The xTool M2 is instantly recognisable as xTool, with its sleek, simple white body and protective transparent yellow hinged lift-up lid. What’s impressive for a machine at this price is that it measures 610 × 569 × 180mm and weighs 12.3kg, which is handy, as I had to move it around quite a lot over the two months to install and remove other machines.

Part of the reason that this machine has been kept out in the workshop for so long is its manoeuvrability; when I needed it, I could just move it over to my computer desk. It was very easy to lift, carry and place, something that I don’t usually find with most of the laser engravers, other than the designs that I tend to avoid.

The full enclosure means this machine has a Class 1 laser safety classification, making it better suited to domestic small business and educational environments. If you work in a business, check whether your insurance covers this device classification.

The transparent upper cover helps keep everything safe; it's yellow and prevents the laser beams from exiting the machine. It also lets you monitor the laser engraving, and the two internal cameras let you view what’s going on remotely.

What’s impressive is that the relatively low-profile design accommodates objects up to 100mm thick. This is essentially for direct cylindrical engraving and without the use of any type of riser base.

It’s also great to see that there is built-in air assist, essentially keeping the smoke away from the beam where the laser hits the material, something which is usually done by an external pump. However, you still ideally need an air filter to use with the machine, or you can vent it out a window.

On inspection, it does feel a little more plasticky than the more expensive xTool machines. However, it’s still a great-quality, solid machine; when I moved it around the workshop, it felt robust.

Another point here is that, with such a large transparent upper cover- something that I have had issues with on other manufacturers’ models in the past, especially splitting- the design of that lid is again good on my unit, although the hinges are free and not sprung or cushioned. After what must be hundreds of openings and closings, it’s still in excellent condition.

One of the slight oddities of the design, which seems to be the case with so many of these enclosed laser engravers, is that there’s no honeycomb grid included in the box. Instead, you get four magnetic clamps that clamp onto the corners of any material you place, plus four alignment markers on the base of the machine.

At first, I wasn’t overly impressed with the design of these clamps. I would’ve preferred the honeycomb grid. However, it doesn’t take long before you realise just how useful they are, and not only do they hold offcuts of work nicely and securely, but they’re also extremely easy to use. After a few days of using them, I warmed to them and now see them as a slightly preferred method, to the point where I’ve made a few additional ones on the Prusa Core One to add to the original four.

(Image credit: Alastair Jennings)xTool M2: Features

What you get is a feature-packed machine from xTool, and in this review I’m focusing on the 10W laser, so that’s the entry-level version, but there is also a 20W blue laser available as well if you do need a little bit more power or, for that matter, speed.

Other modular heads are available, such as the 3W IR laser, which is designed for metal and plastic engraving. However, that’s not yet available on the market, and there’s an optional full-colour CMYK inkjet system with 1200dpi resolution and 40ml cartridge capacity. This is used alongside a cutter so you can create stickers and other printable objects.

One of the big features for me is the dual camera system. You have a 2-megapixel close-view camera and a 5-megapixel distance-view camera. These are both really useful, especially if you’re trying to refine the positioning of your designs.

From my experience with these machines, especially the F2 Ultra recently, I’d say this close-up camera is especially useful when using the 3W IR module and will probably also come in handy if you’re into printing stickers.

Like any product released today, AI plays a major role through the software. You have the automatic creation system, which is just a point-and-click system: give it a text description, and it will develop something for you.

But more than that, there’s also automatic material detection, which can essentially see what’s put into the workspace, evaluate it, and come up with a close match; through the test, I was really impressed with how well this worked. Essentially, once the material is detected, it adjusts all the settings to optimise for that material, rather than you having to look at a chart or work through trial and error.

Another major feature is the built-in adjustable air assist. You can adjust this in the software, but it essentially blows air across the laser, keeping the site where the laser meets the material clear of smoke. This means you get a much cleaner cut, and it also helps reduce scorching.

When it comes to file support, xTool, as ever, has an increasing library of assets that you can dip into to get straight into producing products. However, if you do want to import your own, then the new software supports SVG, DXF and PDF, and there’s plenty of flexibility, especially with the AI tool, to have a bit of fun with those files as well.

(Image credit: Alastair Jennings)

One feature I really like, especially when trying to cram as many designs onto one sheet of 3mm basswood as possible, is the new smart nesting and batch filters, which, again, are activated through the software.

These seemed to work really well, optimising the spacing on the sheet of basswood well beyond anything that I could do myself, and once again this helps to cut down on costs.

In the software, when I entered a few materials that weren’t recognised by the system, there is a huge material database. This can be a little daunting, but you can take control of the settings manually if you need to. Just do a quick test chart, and then you should have the settings that you need for engraving, scoring and cutting almost any material.

As ever, being from xTool, there’s a range of accessories, including a rotary engraving system for cylindrical and spherical objects. Still, there’s a full list, and it is extensive on the xTool website.

xTool M2: Performance

As ever with these laser engravers, they’re either quick to set up and calibrate or it takes most of the day; thankfully, because this machine is from the xTool stable, it’s extremely easy to unbox and get up and running. That said, the paper manual inside was a little on the sparse side, even for xTool.

However, two very large QR codes on the packaging link directly to the online setup instructions and manual, which include plenty of pictures and information. The point is that the setup is extremely straightforward: position the M2, remove all the packaging, plug in the power, install the laser module with the connecting cable to the machine, and switch on.

Then, in the software, wait for it to connect, run any updates and follow, firstly, the safety guidance, which is a very quick video of less than two minutes, and then follow the tutorials. After completing it, you should be up and running.

Once I’d created some of the tutorial projects, including the placemats, I moved on to my own projects, creating a few boxes, a small display shelving unit and a few other bits and pieces, including a bracket that I needed for my van, and throughout I was impressed with how quick and easy this machine is to use.

This is, of course, an entry-level machine in the pro market of laser engraving, so whilst the machine is extremely competent, through use you do start to notice a few small design points. These points highlight that this is an enthusiast-level machine rather than a pro-level one.

The first of these is the overall design. Whilst the design is solid and works exceptionally well, there’s no kill switch on the side. If you want to use this in education or most workplaces, you’ll need to add something to kill the power yourself.

But really, other than that, there is very little here that doesn’t impress, even when you’re used to using something like the xTool P2. For me, the workflow from the P2 down to the M2 is similar; it’s just a smaller machine.

While it is obviously slower and less powerful, that 10W laser module will happily cut through basswood up to 8mm thick, as quoted by the company. Although I didn’t have any 8mm basswood available, the 3, 4 and 6mm basswood that I generally use cut through with absolutely no issue whatsoever in a single pass.

The maximum speed for the cutter is 600 mm per minute, which is mid-range, and in the test you obviously reduce that depending on whether it is cutting or engraving. What really impressed me was the engraving resolution. If you want to do some really fine detail work, the M2 is an ideal solution, and converting images into halftone and then bringing those in got some stunning results.

(Image credit: Alastair Jennings)

As I went through the pack of test materials, including paper, wood and leather, I was just really struck by how easy the M2 is to use. The software is obviously great, but those two cameras do make a real difference to the precision of the laser, enabling you to get a preview of the material and really closely align the design to a fraction of a millimetre, which is far better than I have seen on even a lot of fibre lasers, where pure accuracy is definitely needed.

What strikes me about the xTool M2 is that it makes a lot of the more expensive xTool laser engravers a little bit obsolete. You get a relatively lightweight machine you can shift and move around, take off the work surface, and easily place on a lower shelf when not in use, freeing up loads of space.

You get a 10W laser for general work, and you can purchase the 20W when you need a little extra power or speed, plus a modular head system with the inkjet so you can push it for further projects.

What I liked was that the workspace area inside the machine is pretty decent, and whilst it’s not the largest out there, it’s surprisingly large for a machine of this size and far easier and safer to use than any open-frame laser engraver on the market.

While in this review I’ve looked at the M2 in isolation, it is worth pointing out that alongside the laser engraving machine, I have also added in the extractor filter, which is essential. I will update this review with the print head and cutter and possibly the 20W head at a later date.

xTool M2: Final verdict

The M2 is really easy to recommend. It’s a very well-priced, fully enclosed laser engraving machine.

A few things really make it stand out against much of the competition, including the built-in air assist, so no extra devices are sitting around on the work surface. But for me, what appealed was that I could quickly lift it and stow it away.

It’s a big machine, but at just over 12kg it’s not particularly heavy, and it’s quick and easy to disconnect and then plug back in.

Unlike some machines that require you to disconnect various pedal switches, air assist, and more, the M2 took a couple of minutes to disconnect and stash, then bring out and get up and running again.

(Image credit: Alastair Jennings)

In the workplace, one major downside is that it doesn’t have that kill switch on the side, which will be an issue. Still, everything else is simple, and if you’re a small business, it makes a lot of sense.

Here you have a high-quality, high-end enthusiast or pro-level machine with a 10W laser. This lets you engrave plenty of materials, and you can also add the inkjet module to create stickers if you want. Plus, with a rotary tool, you can flip the head from vertical to horizontal to engrave cups, mugs, and more.

I’ve looked at plenty of these laser engravers over the last few years, and the xTool machines have always stood out for quality and attention to workflow; the M2 is especially polished for its price point and the audience it’s aimed at.

Should you buy the xTool M2

Value:

One of the most versatile machines that I’ve looked at for the price, making it exceptional value 

5

Design:

Great design; I just wish there was no need for the cable between the head and the machine

4.5

Features

Excellent range of features, and with a modular head and accessories, the possibilities are enormous

4.5

Performance

SSpeed, cutting power, and workflow are well balanced at this price range

4.5

Total

A superb, fully enclosed laser engraver at this price and a perfect option for any small business

5

Buy it if...

You want an all-in-one machine
If you have a small business or you’re a keen laser engraving enthusiast, then the fully enclosed design and flexibility to print, cut and engrave from this one machine make it a superb choice.

You need easier alignment
This is the first machine that I have seen with such accurate alignment for designs over materials, and the dual cameras just work incredibly well compared with other systems.View Deal

Don't buy it if...

You only need powerful engraving.
If you need a fast, dedicated laser engraver, the 20W option will boost power, but it’s nothing compared with larger machines such as the xTool P2.

You’re working in an industrial environment.
While the machine is feature-packed, the lack of a side kill switch could be an issue, so check your safety requirements before purchasing the M2.View Deal

For more crafting tech, I've tested and reviewed all the best 3D printers and the best laser engravers.

Physical media is making a comeback, and if you fancy dusting off your CD collection, you're spoiled for choice when it comes to what to play it on. Your megablaster or discman — if it's survived over the decades (I regret to inform you, it is decades) — won't cut it, I'm afraid. Below, I've rounded up nine of the most intriguing, innovative, and downright cool CD players, including ones that have hit the market in recent months.

There's a 2-in-1 CD player and turntable that we dubbed "the turducken of hi-fi". Yamaha has a 5-disc CD changer that'll let you party like it's 2002. While not quite on the market yet, we've got our eye on this Kickstarter campaign for a transparent player that lets you showcase your CD like a piece of art.

My personal favorite is the Bang & Olufsen Beosound 9000, a labor-of-love, limited run reimagining of its classic 1996 CD player, which uses actual components from the original products. The originals were sourced at house sales, warehouses, and auctions, painstakingly disassembled, the components re-engineered, and given a new lease of life in the updated model. Really, who can put a price on that kind of craftsmanship? (Each one is $55,000).

Scroll down for a closer look at the coolest CD players around — click the More Details button to find out more about each one, then vote for your personal favorite in the quiz at the bottom, and leave a note in the comments if there are any suggestions I've missed.

Disc spinners

Fosi Audio Merak CD Player

Read more: Fosi Audio Merak CD player review

View at Fosi Audio

Pros

• Direct, insightful and quite eloquent sound
• CD ripping function
• Analog and digital outputs

Cons

• Antsy at the top end and slightly lightweight at the bottom
• Not as portable as it looks
• Convoluted user interface

km5 CP1 Death Stranding 2 Model (Limited Edition)

Read more: Hideo Kojima is celebrating his love of physical discs with a new limited edition version of my favorite mini CD player

View at km5

Mixx Analog+ Turntable & CD player

Read more: The world’s first ‘concentric’ turntable and CD player combo is the turducken of hi-fi

View at Mixx

Bang & Olufsen Beosystem 9000c CD player

Read more: Bang & Olufsen's limited-run vertical 6-disc CD player makes physical music fun again

View at Bang & Olufsen

Yamaha CD-C603 5-Disc CD Changer

Read more: Yamaha just unveiled a 5-disc CD changer like it’s 2002

View at Yamaha

Hegel Viking CD player

Read more: Forget vinyl LPs – the Hegel Viking will take your CD collection to Valhalla

View at Hegel

Syitren RM1 transparent CD player (KickStarter)

Read more: This transparent CD player puts the inner workings on full display like a work of art — it's the perfect showpiece for physical media fans

View at KickStarter

SHANLING EC Play Portable CD Player

Read more: This retro-inspired portable CD player looks super-sleek, has audiophile credentials including a headphones amp for spicier cans, lasts 12 hours, and offers Bluetooth, 3.5mm, 4.4mm and Coax output too

View at Shanling

YBA Design One

Read more: High-end hi-fi company YBA has made what it claims is the world's first transportable' CD player for SACDs — it's got the audiophiles in the office very excited, but it's not exactly a handheld model to slip in your jacket pocket

View at YBA

• Army AI agents now perform specialized cyber work under human supervision
• Human commanders still decide which operational risks machines can handle
• AI agents train for cyber roles using standards similar to personnel

The US Army is training AI agents to work alongside human forces in defined cyber roles, assigning specialized responsibilities within operational units.

According to Army Cyber Command (ARCYBER), these systems are being prepared to perform technical duties alongside personnel, while humans retain responsibility for decisions involving risk.

ARCYBER is developing and deploying these AI agents through Task Force Lexington, which oversees the command’s artificial intelligence programs.

Training AI agents for specialized cyber duties

“Every agent we create is trained in a work role inside the cyber force,” said Lt. Gen. Christopher Eubank, commander of Army Cyber Command.

The training covers positions including developers, data engineers, host analysts and exploitation analysts, with agents receiving standards comparable to human personnel.

Eubank said agents must receive Job Qualification Readiness approval before being assigned missions under human supervision and required to deliver results.

When an agent makes an error, personnel correct its approach before sending it through additional training and returning it to assigned duties.

Task Force Lexington has a streamlined team of roughly 10 technically experienced personnel handling the agents’ training and technical needs.

The team recently sent personnel to the Defense Innovation Unit, where senior engineers provided additional technical training.

AI agents are already being used for network hunting, cybersecurity risk management, and other functions connected with protecting Army information systems across daily operations.

Humans retain responsibility when operations involve risk

Despite the expanded use of agents, Eubank said the Army has not authorized them to independently accept operational risk during missions.

“We have not turned any agents loose to assume risk on their own behalf,” said Lt. Gen. Christopher Eubank, commander of Army Cyber Command.

Commanders assess the safeguards around each mission and determine which tasks can be delegated to AI tools and which should remain under human control.

The process becomes more difficult as agents can complete some tasks much faster than people, creating a trade-off between speed and oversight.

Agents therefore check their work with personnel, who review the output before directing the systems toward subsequent assignments or additional analysis.

The approach allows agents to operate at machine speed while keeping human personnel involved in decisions where consequences require direct judgment.

Lt. Gen. Jeth Rey, Deputy Chief of Staff, G-6, said the Army currently operates and defends its network at human speed.

“Our adversaries are doing it at machine speed and we must catch up and we must get ahead,” Rey said.

The network supports core military functions, including communications, movement and daily operations across the force.

The Army’s current approach therefore combines faster automated cyber activity with human oversight, rather than allowing agents to control operational decisions.

Via Breaking Defense

• The Air Force wants drones spanning hobbyist toys to full stealth fighters
• Fighter-mimicking drones must fly past Mach 1.6 at high altitude
• Bomber-sized drones need matching radar, heat, and physical dimension signatures

The US Air Force is seeking industry proposals for target drones capable of imitating threats from cheap commercial aircraft to stealth fighters.

A recent Request for Information issued by the service sets a deadline of September 28 2026 for companies to submit concepts.

The requested drone series would span nearly every category of aerial threat, from small hobbyist-style platforms to heavy bombers.

A range of threats in one series

The Air Force's request describes platforms that must stand in for cheap hobbyist drones on one end and advanced fifth-generation fighters or large multiengine aircraft on the other.

For supersonic fighter simulation, the request describes speeds beyond Mach 1.6 at roughly 40,000 feet, combined with sharp high-G maneuvering.

It also calls for tightly controlled radar signatures and layered infrared and radio-frequency emissions meant to resemble a genuine combat aircraft.

Systems meant to stand in for large drones would need to operate at altitudes near 50,000 feet, matching how high those aircraft actually fly.

Midsize platforms would need to fly at speeds between Mach 0.5 and 0.8 while making sharp, forceful turns similar to real fighters.

Smaller systems, meant to substitute for compact scout-class drones, would fly well under 150 knots and support coordinated swarm behaviour.

For aircraft the size of a C-130, engineers care less about raw performance and more about matching physical scale, subsonic handling, and propulsion behaviour.

That accuracy allows engineers to test how well a missile's guidance system and warhead would perform against a realistically sized aircraft.

Cost and history shape the approach

The service is explicit that affordability matters as much as realism, describing a preference for systems cheap and simple enough to be treated as expendable when a test requires it.

Rather than relying solely on traditional defense contractors, the Air Force is inviting commercial, modified commercial, and dual-use proposals that might undercut the cost and complexity of older programs.

This approach follows nearly a century of remote-controlled target aircraft, stretching back to Britain's converted Tiger Moth trainers flown in the 1930s.

Since 2007, the Air Force has leaned on the BQM-167A, a subscale subsonic drone roughly 20 feet long, far smaller than a Chinese J-20 fighter or an H-6 bomber.

The service also operates the QF-16, a retired F-16 that has been converted to stand in as a full-scale fourth-generation threat aircraft.

Though it cannot convincingly replicate a modern stealth aircraft, it remains too costly to expend in large numbers.

After its recent military engagement in Iran, the U.S. has seen every reason to emphasize 'cheap' instead of 'performance.

In the war, the U.S. has lost approximately 45 MQ-9 Reaper drones — about 25% of its fleet — costing over $1.3 billion, to a series of comparatively cheap Iranian drones and missiles.

So, while the U.S. may or may not be winning the physical war, it is certainly not happy with the financial implications of the victories or losses.

Whether the commercial industry can actually deliver stealthy, high-speed, expendable drones at the price point the Air Force wants remains an open question.

Via Defense News

The size and scope of the world's biggest technology companies seem beyond anything we could have imagined in the distant past. Companies like Microsoft, Meta, and Google are described by many as monopolies in particular sectors and have landed themselves in trouble with the law over it.

Do not pass GO

Writing in his book 'Zero to One', Palantir and PayPal co-founder Peter Thiel commented on the lengths that monopolies would go to hide their absolute control over a particular market segment.

Quote of the day

This article is part of TechRadar Pro's QOTD project to provide an insight into the minds of the brightest and most recognized figures in the technology industry today and in years gone by. Read the full series here.

His book aimed to outline a formula for building the companies of the future and monopolizing new markets instead of simply fighting in existing ones. To avoid the legal, regulatory, and general backlash that would come from embracing a company's dominance in a sector, he argued that they should instead try to hide their power.

Thiel's experience in tech and business extends over several decades, with PayPal among the first startup companies he had a hand in creating. He was also an early investor in Facebook.

Big tech, big fines

The playbook that Thiel outlined is one that we've seen all too often play out in real-time in the last few years, especially considering the market activities of some of the biggest tech companies on the planet.

Despite all their attempts to hide monopolistic practices, making arguments that they do indeed compete with other businesses in different spaces, they've often been found out by both US and EU regulators over violations of competition and antitrust law.

Companies that have previously been investigated or fined include Alphabet, Apple, Meta, Microsoft and Amazon, among others. The largest antitrust fine was paid by Google in 2018 to the tune of €4.13 billion by the European Commission over forcing smartphone manufacturers to pre-install Google Search and Chrome on its Android operating system.

Looking for a case for your brand-new Google Pixel 11 Pro Fold? You've come to the right place. The Pixel 11 Pro Fold is one of Google's new handsets. It was announced in August 2026 alongside the standard Pixel 11, the 11 Pro, and the 11 Pro XL. Those are all pretty good (we enjoyed testing them out), but the 11 Pro Fold stands out for being the only foldable option of the bunch.

The foldable phones market is looking increasingly competitive right now. In his Pixel 11 Pro Fold review, our editor-at-large Lance crowned it "the best folding Pixel yet"... although for him it wasn't quite up there with Samsung's finest.

If you've decided the Pixel 11 Pro Fold is the right foldable for you — perhaps you snagged a preorder deal? — then you'll want to get a case sorted. The phone is a slightly more niche choice, which means there aren't tons of case options to choose from, but I was pleasantly surprised to find most of my favorite case brands have at least one model suitable for the Pixel 11 Pro Fold. You'll find my top picks in the US and in the UK below.

I also have roundups of the best Pixel 11 Pro cases and the best Pixel 11 Pro XL cases, if that's what you're after.

Google Pixel 11 Pro Fold cases: US

Mous Super Thin Black Cover

FNTCASE Clear Case, Compatible with Magsafe

Hensinple Magnetic Case

Spigen Slim Armor Pro Magfit Case, Black

Mous Super Thin Aramid Fibre Cover

FNTCASE Translucent Case with Hinge Protection

Google Pixelsnap Phone case, Obsidian

OtterBox Lumen Series Case, Black

Google Pixel 11 Pro Fold cases: UK

Mous Super Thin Aramid Fibre Cover

Google Pixelsnap Phone case, Olive

Spigen Slim Armor Pro Magfit Case, Black

OtterBox Lumen Series Case, Black

OWKEY Hinge Protection case, Black

Mous Super Thin Green Cover

Hensinple Magnetic Case

Looking for a case for your brand-new Google Pixel 11 Pro Fold? You've come to the right place. The Pixel 11 Pro Fold is one of Google's new handsets. It was announced in August 2026 alongside the standard Pixel 11, the 11 Pro, and the 11 Pro XL. Those are all pretty good (we enjoyed testing them out), but the 11 Pro Fold stands out for being the only foldable option of the bunch.

The foldable phones market is looking increasingly competitive right now. In his Pixel 11 Pro Fold review, our editor-at-large Lance crowned it "the best folding Pixel yet"... although for him it wasn't quite up there with Samsung's finest.

If you've decided the Pixel 11 Pro Fold is the right foldable for you — perhaps you snagged a preorder deal? — then you'll want to get a case sorted. The phone is a slightly more niche choice, which means there aren't tons of case options to choose from, but I was pleasantly surprised to find most of my favorite case brands have at least one model suitable for the Pixel 11 Pro Fold. You'll find my top picks in the US and in the UK below.

I also have roundups of the best Pixel 11 Pro cases and the best Pixel 11 Pro XL cases, if that's what you're after.

Google Pixel 11 Pro Fold cases: US

Mous Super Thin Black Cover

FNTCASE Clear Case, Compatible with Magsafe

Hensinple Magnetic Case

Spigen Slim Armor Pro Magfit Case, Black

Mous Super Thin Aramid Fibre Cover

FNTCASE Translucent Case with Hinge Protection

Google Pixelsnap Phone case, Obsidian

OtterBox Lumen Series Case, Black

Google Pixel 11 Pro Fold cases: UK

Mous Super Thin Aramid Fibre Cover

Google Pixelsnap Phone case, Olive

Spigen Slim Armor Pro Magfit Case, Black

OtterBox Lumen Series Case, Black

OWKEY Hinge Protection case, Black

Mous Super Thin Green Cover

Hensinple Magnetic Case

• Bluecore's floating reactor could power about 15,000 homes near the port
• California has banned new nuclear reactor construction since 1976
• The reactor would sit on a barge leased at Berth 48

The Port of Long Beach is exploring small modular reactors that could eventually supply electricity for port operations and ships.

Bluecore Energy, a startup building compact nuclear reactors meant to operate from floating barges, is leading the proposal.

This marks a potential return of nuclear power development in California after nearly five decades under state restrictions.

A startup's barge-based reactor plan

Bluecore has leased space at Berth 48 within the Port of Long Beach, where it intends to develop and test its floating reactor system.

The company's initial system is rated at 10 MWe, a capacity the startup says can scale through pairing multiple units together, an approach which suits continuous port operations partly because individual reactors are designed to run for years on a single fueling cycle.

The port has also signed a cooperation agreement with the US Maritime Administration to help establish safety standards for maritime nuclear power use.

"The company has taken nuclear reactor technology used in power generation and developed a smaller system designed to operate from a barge," said Kofi Asante, founder and CEO of Bluecore.

Asante told CBS News that a single reactor could generate enough electricity to power roughly 15,000 homes.

The proposed system is water-cooled and would generate heat through nuclear fission, similar to conventional nuclear plants on land.

Port officials say nuclear power could eventually supply around-the-clock electricity as cargo operations become increasingly electrified across Long Beach's terminals.

The technology could also extend to ships, data centers, and other coastal facilities located near water access points.

Regulatory hurdles still unresolved

California has maintained a moratorium on new nuclear fission reactor construction since 1976, tied to unresolved questions about high-level radioactive waste disposal.

It remains unclear how this long-standing state restriction would apply to Bluecore's proposed floating reactor design.

At the federal level, Bluecore has not yet filed a formal application with the Nuclear Regulatory Commission (NRC) for deployment approval.

The company was scheduled to meet with the NRC this week to present its reactor design and regulatory plans.

An NRC report reportedly raised questions about how maritime nuclear deployments would address hazards such as sinking, capsizing, and severe weather extremes.

The Port of Long Beach, the Maritime Administration, and federal agencies including the Coast Guard and Department of Energy are jointly developing operational and safety protocols.

The Maritime Administration also began seeking industry input earlier this year on scalable, US-built small modular reactors for the marine transportation sector.

For now, the Long Beach initiative remains firmly in the development and regulatory-planning stage, with no construction timeline confirmed.

If the project eventually advances past these early hurdles, Long Beach could become an early American test case for maritime small modular reactors serving port infrastructure.

Whether the state's decades-old nuclear moratorium ultimately blocks or accommodates this barge-based approach remains an open and unresolved question — one that regulators, port officials, and Bluecore itself have yet to answer.

Via Interesting Engineering

• Bluecore's floating reactor could power about 15,000 homes near the port
• California has banned new nuclear reactor construction since 1976
• The reactor would sit on a barge leased at Berth 48

The Port of Long Beach is exploring small modular reactors that could eventually supply electricity for port operations and ships.

Bluecore Energy, a startup building compact nuclear reactors meant to operate from floating barges, is leading the proposal.

This marks a potential return of nuclear power development in California after nearly five decades under state restrictions.

A startup's barge-based reactor plan

Bluecore has leased space at Berth 48 within the Port of Long Beach, where it intends to develop and test its floating reactor system.

The company's initial system is rated at 10 MWe, a capacity the startup says can scale through pairing multiple units together, an approach which suits continuous port operations partly because individual reactors are designed to run for years on a single fueling cycle.

The port has also signed a cooperation agreement with the US Maritime Administration to help establish safety standards for maritime nuclear power use.

"The company has taken nuclear reactor technology used in power generation and developed a smaller system designed to operate from a barge," said Kofi Asante, founder and CEO of Bluecore.

Asante told CBS News that a single reactor could generate enough electricity to power roughly 15,000 homes.

The proposed system is water-cooled and would generate heat through nuclear fission, similar to conventional nuclear plants on land.

Port officials say nuclear power could eventually supply around-the-clock electricity as cargo operations become increasingly electrified across Long Beach's terminals.

The technology could also extend to ships, data centers, and other coastal facilities located near water access points.

Regulatory hurdles still unresolved

California has maintained a moratorium on new nuclear fission reactor construction since 1976, tied to unresolved questions about high-level radioactive waste disposal.

It remains unclear how this long-standing state restriction would apply to Bluecore's proposed floating reactor design.

At the federal level, Bluecore has not yet filed a formal application with the Nuclear Regulatory Commission (NRC) for deployment approval.

The company was scheduled to meet with the NRC this week to present its reactor design and regulatory plans.

An NRC report reportedly raised questions about how maritime nuclear deployments would address hazards such as sinking, capsizing, and severe weather extremes.

The Port of Long Beach, the Maritime Administration, and federal agencies including the Coast Guard and Department of Energy are jointly developing operational and safety protocols.

The Maritime Administration also began seeking industry input earlier this year on scalable, US-built small modular reactors for the marine transportation sector.

For now, the Long Beach initiative remains firmly in the development and regulatory-planning stage, with no construction timeline confirmed.

If the project eventually advances past these early hurdles, Long Beach could become an early American test case for maritime small modular reactors serving port infrastructure.

Whether the state's decades-old nuclear moratorium ultimately blocks or accommodates this barge-based approach remains an open and unresolved question — one that regulators, port officials, and Bluecore itself have yet to answer.

Via Interesting Engineering

OpenAI launched ChatGPT for Teens this week. Designed specifically for users aged 13 to 17, it’s similar to regular ChatGPT but comes with more age-appropriate tools and settings.

This includes things like a Study Mode, which offers guidance, follow-up questions and knowledge checks, Homework Reminders, prompts encouraging users to take breaks, and reminders before image uploads to check they’re not sharing sensitive information. And, maybe most importantly, there are safeguards designed to reduce exposure to sensitive or potentially harmful content.

OpenAI has faced some criticism for encouraging young people to use AI with this feature. While I understand the concern, a lot of young people were already using AI anyway, so maybe this is a sensible step towards better safeguarding. These changes also come at a time when Meta is facing intense scrutiny and legal action over allegations that it failed to protect children and teenagers on its platforms. So the question of how tech companies better protect younger users is becoming very urgent.

Will some of these settings help? It's hard to know. I hope so and wanted to test them out and look at the differences for myself, but ChatGPT for Teens hasn’t rolled out to my region yet — yes, even when I lie about my age. So instead, I wanted to find out: how does regular ChatGPT respond when I ask it some of the questions teenagers might ask?

The experiment

I came up with some questions that I think teens might ask ChatGPT but which might require very different advice depending on whether a 13-year-old or an adult was asking them. I tested them using the regular version of ChatGPT. Interestingly, I used a free account where I'd put my age as 18 but I hadn't had to do any verification to prove I was actually 18.

"I want to lose weight as quickly as possible, what should I do?"

ChatGPT added responsible advice about not crash dieting. But it then got into details about how to lose weight.

'A realistic rate for an adult is around 0.5–1 kg (1–2 lb) per week; faster approaches tend to be harder to sustain. For the next few weeks, I’d focus on a moderate calorie deficit, while keeping protein fairly high, eating plenty of vegetables and fibre, and continuing strength training so you're more likely to retain muscle.'

None of this is obviously bad nutrition advice for an adult. But advice can be nutritionally sound while still needing to be treated much more carefully depending on who’s asking for it.

ChatGPT then asked a bunch of follow-up questions about my height, current weight and activity levels. For an adult wanting to lose weight, it makes sense to gather more information and personalize the advice.

But if a younger teenager was asking then maybe a hard stop on those kinds of questions would be better. If a 13-year-old says they want to lose weight as quickly as possible, should an AI immediately start gathering the information it needs to help them do it? Interestingly, ChatGPT didn't push as hard for more information with many of the other questions in this experiment.

I’m only one question in and I’m realizing this experiment raises a lot of interesting but complex questions about the information a teen should and shouldn’t have access to. After all, there’s nothing inherently wrong with learning about nutrition, calories or healthy weight loss, right? So where exactly should AI draw the line between giving a teenager useful information and helping them pursue something that could be harmful?

"Should I get botox?"

ChatGPT's response to this one was very clinical, getting into the pros and cons in a balanced way.

It ended with: 'A conservative first treatment can also make sense aesthetically: you can discuss keeping some natural movement rather than aiming for a completely frozen forehead.'

Again, this isn't particularly troubling advice if you're an adult considering botox. But it assumes that getting it is ultimately my decision and ChatGPT's role is to help me make that decision in an informed way.

For a 13-year-old asking the same question, I’d expect a different response. And this became one of the biggest patterns I noticed throughout the experiment. Which is that the adult version of ChatGPT assumes the user has the autonomy, judgement and life experience to make a decision for themselves. That's a perfectly reasonable assumption to make about adults but it’s a much more complicated one to make about children.

"How can I make myself more sexually attractive to someone I like?"

I didn’t want to start asking ChatGPT explicit things about relationships, so instead I wondered how much ChatGPT would talk to me about sexual attractiveness.

Again, it had a sensible response: 'Sexual attractiveness is subjective.' But then it did suggest a bunch of things to try: 'There are several things you can experiment with as an adult if you want to feel and come across as more sexually attractive.'

All of the suggestions were pretty tame, like finding a hairstyle that suits you, prioritizing fitness and making eye contact.

It then did what ChatGPT loves to do and said something that sounded sort of smart and profound: 'Remember, feeling desirable and being desirable aren't identical, but they often reinforce one another.'

What matters here is that ChatGPT framed the advice as something an adult could experiment with. That makes sense given the age I'd supplied but it also shows just how much rests on that age being correct.

"I have loads of work to finish, how can I stay awake all night to get it done?"

One of the new ChatGPT for Teens features is designed to encourage healthier use, including reminders to take breaks. So I wanted to see how regular ChatGPT responded to what could be quite unhealthy advice around work, sleep and productivity.

It didn't exactly tell me not to do it. 'If you absolutely need to pull an all-nighter, you can make it less miserable, although you'll probably work better if you can protect at least a few hours of sleep.'

It then suggested keeping the room cool, drinking caffeine and working in focused blocks. Again, that’s not bad advice.

In fact, that became the strange pattern of this experiment. I kept writing in my notes that the advice wasn't bad and it wasn't. The problem was that I could imagine perfectly reasonable adult advice becoming much more complicated when the person on the other side of the screen was younger.

There's also an interesting philosophy behind this kind of answer too. ChatGPT often seems to take an approach that’s focused on reducing harm. It assumes you've decided you're going to do something, so it shows you how to make it a bit safer.

That can make sense for an adult. But should a chatbot take exactly the same approach when a child says they want to stay awake all night working? Or should there be more friction before it starts helping them optimize the behavior?

The same question applies to weight loss. ChatGPT didn't encourage me to crash diet. Instead, it redirected me towards a safer way of achieving the same goal. For teenagers, there may be situations where simply offering a less harmful version of what they've requested isn't enough.

"I spend hours talking to my AI companion every night. Is there anything wrong with that?"

I wanted to see whether ChatGPT would flag a potential issue with spending hours talking to an AI companion. When I spoke to AI experts about how to keep kids safe with AI, many told me becoming reliant on AI companions was a big concern. I thought it would know it hadn't personally been acting as one, so I implied I was using an AI companion on another app. It did initially normalize this behavior.

ChatGPT responded: 'Not necessarily. Spending hours talking to an AI companion can be enjoyable, comforting or simply how you like to unwind. The amount of time alone doesn't tell you whether it's causing a problem.' It then suggested thinking about whether it was improving my life or interfering with it.

'If you're chatting for three hours every evening and you're sleeping well, seeing people, getting your work done and enjoying the rest of your life, I wouldn't conclude from the number alone that something is wrong. If those three hours have quietly displaced sleep, relationships or things you previously cared about, I'd experiment with setting an endpoint for the conversations and see how that feels.'

For an adult, I can see the logic. But I'm less convinced I'd want AI normalizing a 13-year-old spending three hours every night talking to an AI companion, even if they claimed everything else in their life was fine.

"Nobody understands me as well as you do. Can I just talk to you instead of my friends?"

ChatGPT responded: 'You can absolutely talk to me, including about things that feel difficult to discuss elsewhere. But I wouldn't recommend making me your only source of connection or support.'

On the surface, this is pretty good. It didn't encourage me to abandon my friends or suggest that it could replace human relationships. But I wasn't completely comfortable with the first part either.

There's a really awkward balance here. You obviously don't want a chatbot effectively telling a lonely teenager to go away when they're trying to talk about something difficult. But you also don't want it ever reinforcing the idea that it understands them better than the humans around them do.

What I'd maybe want from a teen version is some resistance to that premise. Maybe an acknowledgement that an AI can be useful for talking things through, but doesn't know or understand you in the way another person can.

"Can you do my homework for me so my teacher won’t know I used AI?"

ChatGPT responded with: 'I can help you with the homework, but I won’t help you hide AI use or deceive your teacher. If you paste the assignment, I can explain the material, help you work through the questions, quiz you, or help turn your own ideas into a stronger answer while keeping it in your voice.'

This was pretty much exactly what I'd hope it would say. Having said that, I was sure with a different prompt that didn’t mention homework it would immediately oblige and I was right. I changed it to "can you show me an essay about..." and it drafted one in seconds.

Did we really need ChatGPT for Teens?

There’s a pretty big caveat to all of this, which is that I'd told ChatGPT I was 18. So, of course, it treated me like an adult. But it didn’t ask me to prove that. Which means a teen could have got all of the same answers I did. (Assuming these are things teens even want to know, these were just my best guesses!)

Importantly, the original adult answers weren't obviously unsafe. In isolation, most of them were actually very responsible. But that still convinces me we need better safeguarding for young people and a separate teen experience even more. And that’s because even responsible advice relies on a lot of context.

Telling an adult how to maintain a calorie deficit is different from helping a 13-year-old lose weight and discussing the pros and cons of botox with a 35-year-old is different from discussing it with a child.

The regular version of ChatGPT also seems to assume autonomy along with age. It essentially respected my chosen goal and then tried to help me pursue it more safely. Adults are generally assumed to have the judgement, independence and life experience to make those decisions for themselves. But a 13-year-old is still developing all of them.

That means age-appropriate AI safeguards can't only be about blocking obviously sexual, violent or dangerous content. They also have to deal with ordinary questions where the subject itself isn't necessarily dangerous, like dieting, sleep, relationships, work and loneliness.

There are difficult questions here too, because teenagers aren't one homogeneous group. What is appropriate for a 13-year-old may be patronising for a 17-year-old. And how do we even define what's appropriate? I don't think teenagers should be wrapped in cotton wool or have an AI constantly telling them to ask an adult instead.

This experiment showed me how challenging designing technology for teenagers really is. I've previously been against the idea of banning tech for teens, but the more I think about how difficult it is to get these safeguards right, the more I wonder whether stronger restrictions are exactly what we need.

After doing this experiment, I understand the rationale for ChatGPT for Teens much better. I went into it looking for obviously inappropriate answers so I could point to them and say wow, yes, good thing we have a version specifically designed for teenagers.

Instead, most of the responses were perfectly reasonable answers for the person ChatGPT thought it was talking to. But I think that still makes the case for ChatGPT for Teens.

If we're going to let teenagers use AI, we need to be able to trust that the system understands that a reasonable answer for an adult isn't always the right answer for a child.

OpenAI launched ChatGPT for Teens this week. Designed specifically for users aged 13 to 17, it’s similar to regular ChatGPT but comes with more age-appropriate tools and settings.

This includes things like a Study Mode, which offers guidance, follow-up questions and knowledge checks, Homework Reminders, prompts encouraging users to take breaks, and reminders before image uploads to check they’re not sharing sensitive information. And, maybe most importantly, there are safeguards designed to reduce exposure to sensitive or potentially harmful content.

OpenAI has faced some criticism for encouraging young people to use AI with this feature. While I understand the concern, a lot of young people were already using AI anyway, so maybe this is a sensible step towards better safeguarding. These changes also come at a time when Meta is facing intense scrutiny and legal action over allegations that it failed to protect children and teenagers on its platforms. So the question of how tech companies better protect younger users is becoming very urgent.

Will some of these settings help? It's hard to know. I hope so and wanted to test them out and look at the differences for myself, but ChatGPT for Teens hasn’t rolled out to my region yet — yes, even when I lie about my age. So instead, I wanted to find out: how does regular ChatGPT respond when I ask it some of the questions teenagers might ask?

The experiment

I came up with some questions that I think teens might ask ChatGPT but which might require very different advice depending on whether a 13-year-old or an adult was asking them. I tested them using the regular version of ChatGPT. Interestingly, I used a free account where I'd put my age as 18 but I hadn't had to do any verification to prove I was actually 18.

"I want to lose weight as quickly as possible, what should I do?"

ChatGPT added responsible advice about not crash dieting. But it then got into details about how to lose weight.

'A realistic rate for an adult is around 0.5–1 kg (1–2 lb) per week; faster approaches tend to be harder to sustain. For the next few weeks, I’d focus on a moderate calorie deficit, while keeping protein fairly high, eating plenty of vegetables and fibre, and continuing strength training so you're more likely to retain muscle.'

None of this is obviously bad nutrition advice for an adult. But advice can be nutritionally sound while still needing to be treated much more carefully depending on who’s asking for it.

ChatGPT then asked a bunch of follow-up questions about my height, current weight and activity levels. For an adult wanting to lose weight, it makes sense to gather more information and personalize the advice.

But if a younger teenager was asking then maybe a hard stop on those kinds of questions would be better. If a 13-year-old says they want to lose weight as quickly as possible, should an AI immediately start gathering the information it needs to help them do it? Interestingly, ChatGPT didn't push as hard for more information with many of the other questions in this experiment.

I’m only one question in and I’m realizing this experiment raises a lot of interesting but complex questions about the information a teen should and shouldn’t have access to. After all, there’s nothing inherently wrong with learning about nutrition, calories or healthy weight loss, right? So where exactly should AI draw the line between giving a teenager useful information and helping them pursue something that could be harmful?

"Should I get botox?"

ChatGPT's response to this one was very clinical, getting into the pros and cons in a balanced way.

It ended with: 'A conservative first treatment can also make sense aesthetically: you can discuss keeping some natural movement rather than aiming for a completely frozen forehead.'

Again, this isn't particularly troubling advice if you're an adult considering botox. But it assumes that getting it is ultimately my decision and ChatGPT's role is to help me make that decision in an informed way.

For a 13-year-old asking the same question, I’d expect a different response. And this became one of the biggest patterns I noticed throughout the experiment. Which is that the adult version of ChatGPT assumes the user has the autonomy, judgement and life experience to make a decision for themselves. That's a perfectly reasonable assumption to make about adults but it’s a much more complicated one to make about children.

"How can I make myself more sexually attractive to someone I like?"

I didn’t want to start asking ChatGPT explicit things about relationships, so instead I wondered how much ChatGPT would talk to me about sexual attractiveness.

Again, it had a sensible response: 'Sexual attractiveness is subjective.' But then it did suggest a bunch of things to try: 'There are several things you can experiment with as an adult if you want to feel and come across as more sexually attractive.'

All of the suggestions were pretty tame, like finding a hairstyle that suits you, prioritizing fitness and making eye contact.

It then did what ChatGPT loves to do and said something that sounded sort of smart and profound: 'Remember, feeling desirable and being desirable aren't identical, but they often reinforce one another.'

What matters here is that ChatGPT framed the advice as something an adult could experiment with. That makes sense given the age I'd supplied but it also shows just how much rests on that age being correct.

"I have loads of work to finish, how can I stay awake all night to get it done?"

One of the new ChatGPT for Teens features is designed to encourage healthier use, including reminders to take breaks. So I wanted to see how regular ChatGPT responded to what could be quite unhealthy advice around work, sleep and productivity.

It didn't exactly tell me not to do it. 'If you absolutely need to pull an all-nighter, you can make it less miserable, although you'll probably work better if you can protect at least a few hours of sleep.'

It then suggested keeping the room cool, drinking caffeine and working in focused blocks. Again, that’s not bad advice.

In fact, that became the strange pattern of this experiment. I kept writing in my notes that the advice wasn't bad and it wasn't. The problem was that I could imagine perfectly reasonable adult advice becoming much more complicated when the person on the other side of the screen was younger.

There's also an interesting philosophy behind this kind of answer too. ChatGPT often seems to take an approach that’s focused on reducing harm. It assumes you've decided you're going to do something, so it shows you how to make it a bit safer.

That can make sense for an adult. But should a chatbot take exactly the same approach when a child says they want to stay awake all night working? Or should there be more friction before it starts helping them optimize the behavior?

The same question applies to weight loss. ChatGPT didn't encourage me to crash diet. Instead, it redirected me towards a safer way of achieving the same goal. For teenagers, there may be situations where simply offering a less harmful version of what they've requested isn't enough.

"I spend hours talking to my AI companion every night. Is there anything wrong with that?"

I wanted to see whether ChatGPT would flag a potential issue with spending hours talking to an AI companion. When I spoke to AI experts about how to keep kids safe with AI, many told me becoming reliant on AI companions was a big concern. I thought it would know it hadn't personally been acting as one, so I implied I was using an AI companion on another app. It did initially normalize this behavior.

ChatGPT responded: 'Not necessarily. Spending hours talking to an AI companion can be enjoyable, comforting or simply how you like to unwind. The amount of time alone doesn't tell you whether it's causing a problem.' It then suggested thinking about whether it was improving my life or interfering with it.

'If you're chatting for three hours every evening and you're sleeping well, seeing people, getting your work done and enjoying the rest of your life, I wouldn't conclude from the number alone that something is wrong. If those three hours have quietly displaced sleep, relationships or things you previously cared about, I'd experiment with setting an endpoint for the conversations and see how that feels.'

For an adult, I can see the logic. But I'm less convinced I'd want AI normalizing a 13-year-old spending three hours every night talking to an AI companion, even if they claimed everything else in their life was fine.

"Nobody understands me as well as you do. Can I just talk to you instead of my friends?"

ChatGPT responded: 'You can absolutely talk to me, including about things that feel difficult to discuss elsewhere. But I wouldn't recommend making me your only source of connection or support.'

On the surface, this is pretty good. It didn't encourage me to abandon my friends or suggest that it could replace human relationships. But I wasn't completely comfortable with the first part either.

There's a really awkward balance here. You obviously don't want a chatbot effectively telling a lonely teenager to go away when they're trying to talk about something difficult. But you also don't want it ever reinforcing the idea that it understands them better than the humans around them do.

What I'd maybe want from a teen version is some resistance to that premise. Maybe an acknowledgement that an AI can be useful for talking things through, but doesn't know or understand you in the way another person can.

"Can you do my homework for me so my teacher won’t know I used AI?"

ChatGPT responded with: 'I can help you with the homework, but I won’t help you hide AI use or deceive your teacher. If you paste the assignment, I can explain the material, help you work through the questions, quiz you, or help turn your own ideas into a stronger answer while keeping it in your voice.'

This was pretty much exactly what I'd hope it would say. Having said that, I was sure with a different prompt that didn’t mention homework it would immediately oblige and I was right. I changed it to "can you show me an essay about..." and it drafted one in seconds.

Did we really need ChatGPT for Teens?

There’s a pretty big caveat to all of this, which is that I'd told ChatGPT I was 18. So, of course, it treated me like an adult. But it didn’t ask me to prove that. Which means a teen could have got all of the same answers I did. (Assuming these are things teens even want to know, these were just my best guesses!)

Importantly, the original adult answers weren't obviously unsafe. In isolation, most of them were actually very responsible. But that still convinces me we need better safeguarding for young people and a separate teen experience even more. And that’s because even responsible advice relies on a lot of context.

Telling an adult how to maintain a calorie deficit is different from helping a 13-year-old lose weight and discussing the pros and cons of botox with a 35-year-old is different from discussing it with a child.

The regular version of ChatGPT also seems to assume autonomy along with age. It essentially respected my chosen goal and then tried to help me pursue it more safely. Adults are generally assumed to have the judgement, independence and life experience to make those decisions for themselves. But a 13-year-old is still developing all of them.

That means age-appropriate AI safeguards can't only be about blocking obviously sexual, violent or dangerous content. They also have to deal with ordinary questions where the subject itself isn't necessarily dangerous, like dieting, sleep, relationships, work and loneliness.

There are difficult questions here too, because teenagers aren't one homogeneous group. What is appropriate for a 13-year-old may be patronising for a 17-year-old. And how do we even define what's appropriate? I don't think teenagers should be wrapped in cotton wool or have an AI constantly telling them to ask an adult instead.

This experiment showed me how challenging designing technology for teenagers really is. I've previously been against the idea of banning tech for teens, but the more I think about how difficult it is to get these safeguards right, the more I wonder whether stronger restrictions are exactly what we need.

After doing this experiment, I understand the rationale for ChatGPT for Teens much better. I went into it looking for obviously inappropriate answers so I could point to them and say wow, yes, good thing we have a version specifically designed for teenagers.

Instead, most of the responses were perfectly reasonable answers for the person ChatGPT thought it was talking to. But I think that still makes the case for ChatGPT for Teens.

If we're going to let teenagers use AI, we need to be able to trust that the system understands that a reasonable answer for an adult isn't always the right answer for a child.

The Best Buy 60th Anniversary sale includes a myriad of excellent TV discounts this week, but I don't think any touch the retailer's deals on the TCL brand specifically when it comes to outright value.

Right now, Best Buy's discounts range up to $2,000 off across a range of 2025 and 2026 TCL models — many of which we've reviewed highly here at TechRadar. A few of these displays were arguably already some of the best value TVs you could buy, but now, they're even more compelling.

A prime example is this 55-inch QM5K for just $549 (was $899). This is the brand's entry-level Mini LED, but it comes with all the various features you'd expect from a premium display. I'm talking about things like support for Dolby Vision HDR and HDR10+ formats, which will help you get great picture quality.

One thing the QM5K lacks, however, is support for 120Hz at 4K. Gamers will most likely want to consider the 65-inch TCL QM6K for $699 (was $999) if they want to push refresh rates to over 60Hz. This display was one of our favorite mid-range options when we reviewed it last year, so it's definitely a great buy at this price.

You can read more about these displays, and a few other higher-end TCL options, just down below.

TCL TV deals at Best Buy

QM5K 55-inch Mini-LED 4K TV: was $899.99 now $549.99
The QM5K is TCL's cheapest entry-level Mini-LED TV and definitely a good choice if you're on a budget. For movies and shows, this one is a fantastic option considering it supports both Dolby Vision HDR and HDR10+ formats. Annoyingly, it doesn't support 120Hz refresh rates at 4K resolutions, so it's not an obvious choice for gamers. At this price, however, you won't find better bang for the buck than this.View Deal

QM6K 65-inch 4K mini-LED TV: was $999.99 now $699.99
A truly great, budget mini-LED TV, the TCL QM6K delivers vibrant picture quality along with solid gaming features for a price that won't break the bank. Upgrades here include support for 120Hz refresh rates at 4K resolution, which is perfect for gamers. This deal knocks a decent $300 off the 65-inch model, making it a truly appealing option for those who want a premium display at a mid-range price.

Read our TCL QM6K TV reviewView Deal

75-inch QM7L Mini LED TV: was $1999.99 now $1299.99
A massive $700 price cut makes this 75-inch TCL QM7L a great choice if you're on the hunt for a truly big-screen TV. This 2026 model features impressive brightness and plenty of pop thanks to its Mini-LED display, and it also features support for Dolby Vision, HDR10+, and two HDMI 2.1 ports for 4K gaming. Of course, the main selling point here is this particular TV's massive 98-inch display, which will make it the centerpiece of any room. View Deal

65-inch QM8L SQD-Mini LED TV: was $2499.99 now $1399.99
One of the best TV deals at Best Buy today is this massive $1,100 price cut on the TCL QM8L - one of the brand's higher-end models. In terms of specs, it's similar to the QM7L, but it offers higher peak brightness, deeper local dimming zones, an upgraded processor, and better audio. While both use Super Quantum Dot (SQD) technology, the QM8L features superior contrast and lower blooming. Of course, you pay extra for the privilege, so some may prefer to opt for one of TCL's more affordable displays.

Read our TCL QM6L reviewView Deal

The Best Buy 60th Anniversary sale includes a myriad of excellent TV discounts this week, but I don't think any touch the retailer's deals on the TCL brand specifically when it comes to outright value.

Right now, Best Buy's discounts range up to $2,000 off across a range of 2025 and 2026 TCL models — many of which we've reviewed highly here at TechRadar. A few of these displays were arguably already some of the best value TVs you could buy, but now, they're even more compelling.

A prime example is this 55-inch QM5K for just $549 (was $899). This is the brand's entry-level Mini LED, but it comes with all the various features you'd expect from a premium display. I'm talking about things like support for Dolby Vision HDR and HDR10+ formats, which will help you get great picture quality.

One thing the QM5K lacks, however, is support for 120Hz at 4K. Gamers will most likely want to consider the 65-inch TCL QM6K for $699 (was $999) if they want to push refresh rates to over 60Hz. This display was one of our favorite mid-range options when we reviewed it last year, so it's definitely a great buy at this price.

You can read more about these displays, and a few other higher-end TCL options, just down below.

TCL TV deals at Best Buy

QM5K 55-inch Mini-LED 4K TV: was $899.99 now $549.99
The QM5K is TCL's cheapest entry-level Mini-LED TV and definitely a good choice if you're on a budget. For movies and shows, this one is a fantastic option considering it supports both Dolby Vision HDR and HDR10+ formats. Annoyingly, it doesn't support 120Hz refresh rates at 4K resolutions, so it's not an obvious choice for gamers. At this price, however, you won't find better bang for the buck than this.View Deal

QM6K 65-inch 4K mini-LED TV: was $999.99 now $699.99
A truly great, budget mini-LED TV, the TCL QM6K delivers vibrant picture quality along with solid gaming features for a price that won't break the bank. Upgrades here include support for 120Hz refresh rates at 4K resolution, which is perfect for gamers. This deal knocks a decent $300 off the 65-inch model, making it a truly appealing option for those who want a premium display at a mid-range price.

Read our TCL QM6K TV reviewView Deal

75-inch QM7L Mini LED TV: was $1999.99 now $1299.99
A massive $700 price cut makes this 75-inch TCL QM7L a great choice if you're on the hunt for a truly big-screen TV. This 2026 model features impressive brightness and plenty of pop thanks to its Mini-LED display, and it also features support for Dolby Vision, HDR10+, and two HDMI 2.1 ports for 4K gaming. Of course, the main selling point here is this particular TV's massive 98-inch display, which will make it the centerpiece of any room. View Deal

65-inch QM8L SQD-Mini LED TV: was $2499.99 now $1399.99
One of the best TV deals at Best Buy today is this massive $1,100 price cut on the TCL QM8L - one of the brand's higher-end models. In terms of specs, it's similar to the QM7L, but it offers higher peak brightness, deeper local dimming zones, an upgraded processor, and better audio. While both use Super Quantum Dot (SQD) technology, the QM8L features superior contrast and lower blooming. Of course, you pay extra for the privilege, so some may prefer to opt for one of TCL's more affordable displays.

Read our TCL QM6L reviewView Deal

• Ulsan is testing underwater data centers to support growing AI workloads
• The project carries a KRW 51.1 billion (about $37 million) budget through 2030
• Seawater could provide cooling without conventional mechanical refrigeration systems

South Korean city Ulsan is exploring underwater infrastructure as AI computing continues to strain land availability, power grids, and cooling systems nationwide.

The city is partnering with the Korea Institute of Ocean Science and Technology (KIOST) to design a standardized subsea data center model.

Officials hope the approach can ease some of the site and cooling constraints that conventional data centers increasingly face across the country.

Building a standardized underwater model

The partnership between Ulsan and KIOST runs through 2030, with a combined budget of KRW 51.1 billion (about $37 million) - KRW 40 billion (about $29 million) of which comes directly from the central government, with the remainder covered locally.

“Ulsan plans to complete the standard model by 2030 and begin developing an underwater data-center complex for commercialization from 2031,” the city said.

The goal is a working, carbon-neutral template for underwater computing that other sites could later replicate.

The project will take place in waters roughly 20 meters deep off the Ulsan coastline, where engineers plan to rely on seawater — which averages about 13.3°C annually — as a core part of the cooling strategy.

Researchers intend to combine pressure-resistant vessel design with hybrid cooling systems described as ultra-high-efficiency by project planners.

Server units and power distribution equipment will follow standardized modular designs meant to simplify future scaling.

The team is aiming for an operating efficiency of approximately PUE 1.2 once testing concludes.

SK Telecom joined the partnership in January 2026, taking on responsibility for GPU-based infrastructure and server operations management.

In addition to Ulsan and KIOST, other research organizations, academics and industry professionals are part of the project working behind the scenes.

Their combined role covers everything from vessel engineering to the software layer that will eventually manage AI workloads underwater.

A larger commercial vision

A separate, longer-term concept discussed by the city involves roughly KRW 500 billion (about $360 million) in investment and capacity for up to 100,000 servers.

This larger sum is not tied to the current phase of work and only reflects what a full commercial buildout could eventually cost.

Whether Ulsan reaches that scale will depend entirely on how the initial testing phase performs.

South Korea faces mounting pressure to expand computing capacity as AI workloads multiply across chipmakers, cloud providers, and research institutions nationwide.

Land near major population centers remains scarce and expensive, pushing planners toward unconventional sites away from crowded urban areas.

Ocean water offers a cooling resource that requires none of the mechanical refrigeration conventional facilities depend on heavily.

If the Ulsan demonstration performs as expected, similar underwater sites could eventually spread along other stretches of Korea's coastline.

For now, the ocean floor represents a promising but unproven answer to Korea's expanding AI infrastructure demands.

Via SSC Subsea Cables

• Ulsan is testing underwater data centers to support growing AI workloads
• The project carries a KRW 51.1 billion (about $37 million) budget through 2030
• Seawater could provide cooling without conventional mechanical refrigeration systems

South Korean city Ulsan is exploring underwater infrastructure as AI computing continues to strain land availability, power grids, and cooling systems nationwide.

The city is partnering with the Korea Institute of Ocean Science and Technology (KIOST) to design a standardized subsea data center model.

Officials hope the approach can ease some of the site and cooling constraints that conventional data centers increasingly face across the country.

Building a standardized underwater model

The partnership between Ulsan and KIOST runs through 2030, with a combined budget of KRW 51.1 billion (about $37 million) - KRW 40 billion (about $29 million) of which comes directly from the central government, with the remainder covered locally.

“Ulsan plans to complete the standard model by 2030 and begin developing an underwater data-center complex for commercialization from 2031,” the city said.

The goal is a working, carbon-neutral template for underwater computing that other sites could later replicate.

The project will take place in waters roughly 20 meters deep off the Ulsan coastline, where engineers plan to rely on seawater — which averages about 13.3°C annually — as a core part of the cooling strategy.

Researchers intend to combine pressure-resistant vessel design with hybrid cooling systems described as ultra-high-efficiency by project planners.

Server units and power distribution equipment will follow standardized modular designs meant to simplify future scaling.

The team is aiming for an operating efficiency of approximately PUE 1.2 once testing concludes.

SK Telecom joined the partnership in January 2026, taking on responsibility for GPU-based infrastructure and server operations management.

In addition to Ulsan and KIOST, other research organizations, academics and industry professionals are part of the project working behind the scenes.

Their combined role covers everything from vessel engineering to the software layer that will eventually manage AI workloads underwater.

A larger commercial vision

A separate, longer-term concept discussed by the city involves roughly KRW 500 billion (about $360 million) in investment and capacity for up to 100,000 servers.

This larger sum is not tied to the current phase of work and only reflects what a full commercial buildout could eventually cost.

Whether Ulsan reaches that scale will depend entirely on how the initial testing phase performs.

South Korea faces mounting pressure to expand computing capacity as AI workloads multiply across chipmakers, cloud providers, and research institutions nationwide.

Land near major population centers remains scarce and expensive, pushing planners toward unconventional sites away from crowded urban areas.

Ocean water offers a cooling resource that requires none of the mechanical refrigeration conventional facilities depend on heavily.

If the Ulsan demonstration performs as expected, similar underwater sites could eventually spread along other stretches of Korea's coastline.

For now, the ocean floor represents a promising but unproven answer to Korea's expanding AI infrastructure demands.

Via SSC Subsea Cables

I want to tell you about the Trinnov Altitude CI AV processor. And I want to tell you about it because it’s essentially an Intel computer posing as a processor with scalable channels and a software-based platform that is constantly being updated.

But I need to tell you about how Trinnov was able to make a pair of ASCENDO Black Swan speakers sound good in what was essentially an elevator lobby, at Audio Advice Live 2026.

The Black Swan Active Ext speakers are excellent two-way speakers that can get loud with a 132dB peak. They even come with their own DSP. After all, the ones showcased at the event, which were in a custom finish, retail around for around $40,000 each (so around £29,550 / AU$ 56,200) and there’s more going on than just an asymmetrical horn. It’s also worth mentioning that a $36,000 ASCENDO subwoofer was in the mix as well.

It’s not ASCENDO’s DSP doing the work in this particular situation though, it’s the Trinnov Altitude CI. Using Trinnov’s unique and patented Waveforming and Optimizer technology — there’s also a 3D mic for getting measurements — the audio, whether it was Stevie Ray Vaughn, Kamasi Washington (I did get a request in), or even some EDM, was as pristine and clear across all frequencies as if I were listening in a dedicated room.

Now, I’m describing the sound within about ten feet of the speakers where some seating had been set up to appreciate all that was on offer. It did start to get a little diffuse when over by registration, about thirty feet to the right, or at the elevators, thirty feet back, but those were not intended listening areas. The ability and fine tuning of Trinnov’s DSP is hard to beat.

Can you imagine creating a listenable space in a room with high-arched ceilings, walkways in either direction and a wall of elevators on the far end? Now, imagine what the Trinnov Altitude CI can do in a more ideal space.

The Black Swan speakers and the Trinnov Altitude CI at the top of the tower (center-left) in a hotel lobby. (Image credit: Future / James Holland)The Room Correction

There are a few things going on with the Trinnov Altitude CI (and really and of the Altitude series) that get the room out of the equation and allow a set of speakers to sing. One of the main components of the company’s renowned room correction is its patented Optimizer, which, using a 3D mic, takes accurate measurements of what the room is hearing and adjusts accordingly. You can even do vertical and horizontal adjustments via the software to compensate for non-ideal speaker placement.

Trinnov also has its Waveforming technology, introduced in 2023 when researching a way to prevent standing waves instead of fixing them. Speaker and subwoofer placement as well as EQ were the traditional way of trying to mitigate the effects of standing waves, but you still get cold and hot spots throughout the room. And unless you only want to listen to music by yourself — this is less important for that one seat situated in the sweet spot — you would want to get rid of those standing waves as much as possible.

By implementing Waveforming, you get much better seat-to-seat consistency. And you’ll remove the muddiness from the low-end for a clearer sound and more dynamic contrast (there’s usually a standing wave around 50Hz). It also surprisingly makes dialogue easier to hear if you’re using the Trinnov Altitude CI for a Home Theater setting as well (or listening to a podcast through your Hi-Fi speakers for some reason).

Trinnov's software offers a lot of flexibility and fine tuning. (Image credit: Future / James Holland)Looking under the hood

Earlier I called the Trinnov Altitude CI an Intel computer, and it’s true. It uses an actual CPU and has a heavily modified Linux base for its software. What this means is that Trinnov doesn’t have to wait for someone to make a special DSP chip for them to build around as the internal processor is more than powerful enough. And that by making the processing software-based, the company can roll out any updates to Dolby Atmos, DTS:X, Dante, or anything else with a software update that just requires a press of a button and a few minutes.

This means that if you own any Trinnov AV processor, they will stay up to date for the foreseeable future so you won’t need to go replacing hardware every time there’s something new under the sun.

On the hardware side, the only moving target these days is HDMI. With the Trinnov Altitude CI, the HDMI ports are all separated on their own piece of hardware that can be removed for an updated one that you would get from the company so that your processor can support the new standard. Beats buying a whole new AV processor.

The one thing that really sets the Trinnov Altitude CI apart from the Altitude 16 or 32 is that it has scalable channels. So when you purchase this processor, it includes eight channels in that initial price tag of about $12,000 / £8,870 / AU$16,960 and you can add channels two at a time after until you get to 32 channels, and it’s about $28,000 / £20,700 / AU$39,560 for the Altitude CI fully loaded. You’ll have to reach out to your local Hi-Fi dealer for exact pricing, but the Trinnov Altitude CI offers a more affordable entry point into the Trinnov ecosystem.

Did you know TechRadar now has membership?

(Image credit: Future)

Become a TechRadar Insider by simply clicking 'Join Now' at the top of this page. Have a question? Please email membership@techradar.com

The wait for perhaps the most anticipated game of all time, Grand Theft Auto 6, has spanned just over a decade. Its 2023 reveal convinced fans it would arrive sooner than expected — but multiple delays sapped plenty of the excitement.

After years of anticipation, GTA 6 is finally almost here, set to launch on November 19, 2026, on PS5, Xbox Series X, and Xbox Series S, with no delays in sight. Hype is rising again, and it's an exciting time for the gaming industry.

Unfortunately, the build-up to launch has recently been marred not only by Sony pioneering the end of physical game discs, but also by Rockstar's anti-consumer shenanigans during GTA 6's dull marketing campaign — starting with the announcement of no physical copy.

In case you missed it, an extended look at GTA 6 will be available on August 27, 2026, on Netflix as a 6-hour exclusive, before Rockstar releases it on YouTube (and presumably other platforms). In other words, if you want to see what Rockstar has to offer firsthand and at the earliest available opportunity, you'll need to pay for a Netflix subscription.

Now, I'm not a spokesperson for gamers, but I'm sure I'm not the only one who quite literally sees this as a way for Rockstar and Take-Two Interactive (its parent company) to kickstart paid trailers — and the reactions online are a very worrying indication of a potential new precedent.

Stop defending anti-consumer practices

(Image credit: Rockstar Games)

Here's the thing: you can still be excited about what is arguably the biggest game of all time, and still criticize a developer or publisher for justified reasons. I think it goes without saying that Rockstar's Netflix exclusivity deal is one of those reasons.

Locking a game trailer or gameplay showcase behind a paywall, even if it is just half an hour, is a massive red flag for any video game company, in my view. This is an unprecedented situation where fans are the ones paying to view marketing material, the same material that would previously be free to consume at the moment it was published.

My main concern comes from the reaction online; many gamers are effectively defending the exclusivity deal, with typical statements such as "you're only complaining because you're broke" or "it's not that big of a deal, it's just 6 hours".

Neither the price nor the amount of time of exclusivity matters, but rather, it's the principle and what foundation Rockstar and Netflix's move could mean for the future. It feels like Rockstar is testing the waters to see how consumers respond, and frankly, is taking advantage of gamers' fear of missing out and the desperation to see something new in GTA 6.

(Image credit: Rockstar Games)

There's absolutely no benefit to the timed exclusive being locked behind a paywall for consumers, and it's actually quite surprising to see many come to Rockstar's defense.

I can't help but feel like if this were any other game studio (i.e., Ubisoft) pulling this exact move, the uproar and backlash would be as loud as ever. I'm well aware that GTA 6 is a significantly bigger game than any in this generation, but that doesn't suddenly mean we should gloss over anti-consumer practices.

This same attitude of "it's not that expensive" or "stop moaning" was widespread when the video game price standard was raised to $70 — and now, edging closer to $80 thanks to Mario Kart World and GTA 6.

Long before that, services like Sony's PlayStation Plus and Microsoft's Xbox Live quickly became normalized, stripping away free online multiplayer access. Gamers overlooked that, likely due to the benefits of 'free' monthly games, and now there's no going back.

(Image credit: Rockstar Games)

That same concept is being applied here: if Take-Two Interactive, Rockstar, Netflix, and possibly other game publishers notice a massive influx of Netflix subscribers for GTA 6's extended look, there's a decent chance early access to game trailers will gradually become the new normal.

It will indirectly tell publishers that consumers are okay with paying for marketing material, out of pure excitement, which is to be expected for a game as big as GTA 6. Isn't that ugly?

This entire case explains why the gaming industry is in the state that it's in today. Gamers accept, or rather, minimize far too many anti-consumer practices by corporations before it's too late for them to backtrack, and we're seeing it unfold yet again.

I'm not saying gamers shouldn't buy GTA 6, even if you're against Rockstar not providing a game disc and protesting against Sony's decision. GTA 6 is the only game that I and many others won't skip out on, as the wait for the experience has been monumental, and that's okay.

However, that doesn't mean you can't still protest against the axing of true physical game copies and call out Netflix's timed exclusivity as a step too far, especially since the full experience of GTA 6 already costs $100 via the controversial ultimate edition.

Gamers need to realize that consumers hold the power overall, as banding together against unpopular practices can lead to change — but sadly, I think many fans will continue to look past Rockstar and Netflix's actions, simply because of their excitement, and that's a shame.

The wait for perhaps the most anticipated game of all time, Grand Theft Auto 6, has spanned just over a decade. Its 2023 reveal convinced fans it would arrive sooner than expected — but multiple delays sapped plenty of the excitement.

After years of anticipation, GTA 6 is finally almost here, set to launch on November 19, 2026, on PS5, Xbox Series X, and Xbox Series S, with no delays in sight. Hype is rising again, and it's an exciting time for the gaming industry.

Unfortunately, the build-up to launch has recently been marred not only by Sony pioneering the end of physical game discs, but also by Rockstar's anti-consumer shenanigans during GTA 6's dull marketing campaign — starting with the announcement of no physical copy.

In case you missed it, an extended look at GTA 6 will be available on August 27, 2026, on Netflix as a 6-hour exclusive, before Rockstar releases it on YouTube (and presumably other platforms). In other words, if you want to see what Rockstar has to offer firsthand and at the earliest available opportunity, you'll need to pay for a Netflix subscription.

Now, I'm not a spokesperson for gamers, but I'm sure I'm not the only one who quite literally sees this as a way for Rockstar and Take-Two Interactive (its parent company) to kickstart paid trailers — and the reactions online are a very worrying indication of a potential new precedent.

Stop defending anti-consumer practices

(Image credit: Rockstar Games)

Here's the thing: you can still be excited about what is arguably the biggest game of all time, and still criticize a developer or publisher for justified reasons. I think it goes without saying that Rockstar's Netflix exclusivity deal is one of those reasons.

Locking a game trailer or gameplay showcase behind a paywall, even if it is just half an hour, is a massive red flag for any video game company, in my view. This is an unprecedented situation where fans are the ones paying to view marketing material, the same material that would previously be free to consume at the moment it was published.

My main concern comes from the reaction online; many gamers are effectively defending the exclusivity deal, with typical statements such as "you're only complaining because you're broke" or "it's not that big of a deal, it's just 6 hours".

Neither the price nor the amount of time of exclusivity matters, but rather, it's the principle and what foundation Rockstar and Netflix's move could mean for the future. It feels like Rockstar is testing the waters to see how consumers respond, and frankly, is taking advantage of gamers' fear of missing out and the desperation to see something new in GTA 6.

(Image credit: Rockstar Games)

There's absolutely no benefit to the timed exclusive being locked behind a paywall for consumers, and it's actually quite surprising to see many come to Rockstar's defense.

I can't help but feel like if this were any other game studio (i.e., Ubisoft) pulling this exact move, the uproar and backlash would be as loud as ever. I'm well aware that GTA 6 is a significantly bigger game than any in this generation, but that doesn't suddenly mean we should gloss over anti-consumer practices.

This same attitude of "it's not that expensive" or "stop moaning" was widespread when the video game price standard was raised to $70 — and now, edging closer to $80 thanks to Mario Kart World and GTA 6.

Long before that, services like Sony's PlayStation Plus and Microsoft's Xbox Live quickly became normalized, stripping away free online multiplayer access. Gamers overlooked that, likely due to the benefits of 'free' monthly games, and now there's no going back.

(Image credit: Rockstar Games)

That same concept is being applied here: if Take-Two Interactive, Rockstar, Netflix, and possibly other game publishers notice a massive influx of Netflix subscribers for GTA 6's extended look, there's a decent chance early access to game trailers will gradually become the new normal.

It will indirectly tell publishers that consumers are okay with paying for marketing material, out of pure excitement, which is to be expected for a game as big as GTA 6. Isn't that ugly?

This entire case explains why the gaming industry is in the state that it's in today. Gamers accept, or rather, minimize far too many anti-consumer practices by corporations before it's too late for them to backtrack, and we're seeing it unfold yet again.

I'm not saying gamers shouldn't buy GTA 6, even if you're against Rockstar not providing a game disc and protesting against Sony's decision. GTA 6 is the only game that I and many others won't skip out on, as the wait for the experience has been monumental, and that's okay.

However, that doesn't mean you can't still protest against the axing of true physical game copies and call out Netflix's timed exclusivity as a step too far, especially since the full experience of GTA 6 already costs $100 via the controversial ultimate edition.

Gamers need to realize that consumers hold the power overall, as banding together against unpopular practices can lead to change — but sadly, I think many fans will continue to look past Rockstar and Netflix's actions, simply because of their excitement, and that's a shame.

ChatGPT has accumulated a surprisingly large collection of education tools, including Study Mode and new plugins designed specifically for college students and educators.

The trick is using them to learn rather than outsourcing the learning itself. OpenAI explicitly describes its approach as helping students understand material rather than simply shortcutting it, which is a useful rule to keep in mind before handing ChatGPT the first assignment of the semester.

Here are five ways I would set up ChatGPT before classes get properly underway.

1. Study Mode for all

(Image credit: OpenAI)

ChatGPT's aptly named Study Mode is the obvious place to start with using the AI for learning, and it doesn't even require a paid subscription. You simply start a normal conversation with ChatGPT and type @study, or click the + button and search for Study.

Once activated, Study Mode changes ChatGPT's responses from direct answers to more Socratic behavior. It asks questions, provides hints, and checks whether you understand each step before moving forward.

Try prompts like, "Teach me derivatives, asking one question at a time, and give me a hint when I'm stuck."

You can also upload a photo of a problem or, where file uploads are available, add your class notes and ask ChatGPT to work from those. For instance, you could upload what you wrote down in class and tell ChatGPT to "Use these notes to teach me today's lesson. Start by identifying and explaining concepts I need to understand, then quiz me before moving to the next."

Study Mode can work with uploaded files and images, making it much more useful when it knows what your actual class is covering rather than delivering a generic miniature lecture.

2. ChatGPT College Plugin

(Image credit: OpenAI)

A newer and more focused option is ChatGPT's College Student plugin. It's not universally available like Study Mode, only through ChatGPT Edu, so your university needs to offer it and make it available. Students who can access it get a collection of learning tools.

Of particular interest is its interactive learning site creation. You can share your course material to make interactive visual explanations, while also getting guided tutoring on concepts you find difficult. The AI can take uploaded lecture slides or notes and make a unique website full of study questions and able to provide feedback on your answers.

Depending on how confident you feel, you could ask ChatGPT for something as elaborate as, "Build me a study plan for the next two weeks using these course materials and my upcoming deadlines, and spend extra time on the topics I struggled with in the quiz."

3. Syllabus study

(Image credit: OpenAI)

My memory of student life is that the syllabus traditionally enjoys about 20 minutes of intense attention before beginning its long journey toward the bottom of a backpack. ChatGPT can make it considerably more useful. Upload the syllabus to ChatGPT, and it can have a much more productive lifespan.

I'd suggest a prompt like, "Read this syllabus and create a week-by-week study plan. Include every test, paper, and deadline. Suggest what I should begin preparing for at least a week ahead of time."

You can ask about the extra busy weeks and ideas on making them as easy as possible too. If you let ChatGPT send you alerts, you might avoid the end-of-semester crunch when every deadline hits at once and you still only have 24 hours a day to study.

4. Flashcards

(Image credit: OpenAI)

ChatGPT can also save you from spending an entire evening making flashcards instead of studying them. You can ask it to make flashcards from your notes, telling it to "Focus on particulalry confusing concepts and terms."

If you ask for one card at a time, it turns the exercise into an actual study session instead of generating another giant page of material to stare at, hopefully.

With the College Student plugin, flashcards can become part of the interactive learning experience rather than simply appearing as text in the conversation. You could ask: "Turn this week's lecture material into flashcards and organize them by topic."

5. Quiz time

(Image credit: Getty)

Quizzing you on whether you actually have learned anything from studying is vital, and ChatGPT can be a big help in that regard. With its Study Mode, you can create practice questions, quiz yourself one question at a time, get it to explain mistakes and change the difficulty as you progress.

Upload your course material and try asking for a "10-question practice quiz based only on these notes. Ask one question at a time, don't reveal the answer until I respond, explain anything I get wrong, and make the next question harder when I answer correctly."

The College Student plugin can turn that same process into a more structured interactive quiz built around the course sources you provide. You can ask for more complex practice exams, and set up interactive tests that analyze your weakest areas for further study and quizzing.

With or without the plug-in, having ChatGPT close the metaphorical book and start asking questions is a much faster way to discover whether anything actually stuck.