Feed aggregator

Large language models have an awkward history with telling the truth, especially if they can't provide a real answer. Hallucinations have been a hazard for AI chatbots since the technology debuted a few years ago. But ChatGPT 5 seems to be going for a new, more humble approach to not knowing answers; admitting it.

Though most AI chatbot responses are accurate, it's impossible to interact with an AI chatbot for long before it provides a partial or complete fabrication as an answer. The AI displays just as much confidence in its answers regardless of their accuracy. AI hallucinations have plagued users and even led to embarrassing moments for the developers during demonstrations.

OpenAI had hinted that the new version of ChatGPT would be willing to plead ignorance over making up an answer, and a viral X post by Kol Tregaskes has drawn attention to the groundbreaking concept of ChatGPT saying, “I don’t know – and I can’t reliably find out.”

GPT-5 says 'I don't know'.Love this, thank you. pic.twitter.com/k6SNFKqZbgAugust 18, 2025

Technically, hallucinations are baked into how these models work. They’re not retrieving facts from a database, even if it looks that way; they're predicting the next most likely word based on patterns in language. When you ask about something obscure or complicated, the AI is guessing the right words to answer it, not doing a classic search engine hunt. Hence, the appearance of entirely made-up sources, statistics, or quotes.

But GPT-5’s ability to stop and say, “I don’t know,” reflects an evolution in how AI models deal with their limitations in terms of their responses, at least. A candid admission of ignorance replaces fictional filler. It may seem anticlimactic, but it's more significant for making the AI seem more trustworthy.

Clarity over hallucinations

Trust is crucial for AI chatbots. Why would you use them if you don't trust the answers? ChatGPT and other AI chatbots have warnings built into them about not relying too much on their answers because of hallucinations, but there are always stories of people ignoring that warning and getting into hot water. If the AI just says it can't answer a question, people might be more inclined to trust the answers it does provide.

Of course, there's still a risk that users will interpret the model's self-doubt as failure. The phrase “I don’t know” might come off as a bug, not a feature, if you don't realize the alternative is a hallucination, not the correct answer. Admitting uncertainty isn't how the all-knowing AI some imagine ChatGPT would behave.

But it's arguably the most human thing ChatGPT could do in this instance. OpenAI's proclaimed goal is artificial general intelligence, AI that can perform any intellectual task a human can. But one of the ironies of AGI is that mimicking human thinking includes uncertainties as well as capabilities.

Sometimes, the smartest thing you can do is to say you don't know something. You can't learn if you refuse to admit there are things you don't know. And, at least it avoids the spectacle of an AI telling you to eat rocks for your health.

You might also like

• GPT‑5 Pro is brilliant, but it’s still nowhere near real AGI, says one of the professors who coined the term
• OpenAI's CEO says he's scared of GPT-5
• AI that seems conscious is coming – and that’s a huge problem, says Microsoft AI's CEO

California and Texas, the country's two most populous states, are getting closer to redrawing their congressional districts in a political fight sparked by President Trump.

(Image credit: Tayfun Coskun/Anadolu via Getty Images; Brandon Bell)

President Trump says he wants to see for himself how National Guard soldiers and police officers are cracking down on what he has described as rampant crime in Washington, D.C.

(Image credit: Kevin Dietsch)

• Apple started with almost no Swift examples and achieved surprising results
• StarChat-Beta was pushed into uncharted territory without clear guidance
• Nearly one million working SwiftUI programs emerged after repeated iterations

Apple researchers recently revealed an experiment in which an AI model was trained to generate user interface code in SwiftUI, even though almost no SwiftUI examples were present in the original data.

The study began with StarChat-Beta, an open source model designed for coding. Its training sources, including TheStack and other collections, contained almost no Swift code.

This absence meant the model did not have the advantage of existing examples to guide its responses, which made the results surprising when a stronger system eventually emerged.

Creating a loop of self-improvement

The team’s solution was to create a feedback cycle. They gave StarChat-Beta a set of interface descriptions and asked it to generate SwiftUI programs from those prompts.

Each generated program was compiled to ensure it actually ran. Interfaces that worked were then compared with the original descriptions using another model, GPT-4V, which judged whether the output matched the request.

Only those that passed both stages remained in the dataset. This cycle was repeated five times, and with every round, the cleaner dataset was fed back into the next model.

By the end of the process, the researchers had nearly one million working SwiftUI samples and a model they called UICoder.

The model was then measured against both automated tests and human evaluation, where results showed it not only performed better than its base model, but also achieved a compilation success rate higher than GPT-4.

One of the striking aspects of the study is that Swift code had been almost entirely excluded from the initial training data.

According to the team, this happened by accident when TheStack dataset was created, leaving only scattered examples found on web pages.

This oversight rules out the idea that UICoder merely recycled code it had already seen - instead, its improvement came from the iterative cycle of generating, filtering, and retraining on its own outputs.

While the results centered on SwiftUI, the researchers suggested the approach “would likely generalize to other languages and UI toolkits.”

If so, this could open paths for more models to be trained in specialized domains where training data is limited.

The prospect raises questions about reliability, sustainability, and whether synthetic datasets can continue to scale without introducing hidden flaws.

UICoder was also trained under carefully controlled conditions, and its success in wider settings is not guaranteed.

Via 9to5mac

You might also like

• The end of laptops at work? Desktop as a service is now cheaper and easier to run
• We've listed the best AI video editors right now
• These are the best AI writers available

CNET rigorously tested more than 50 cordless vacuums to find you the best performance and value for your money and we have some new picks we recommend.

Vivo Vision has many of the same design elements as Apple's VR/AR, but is only available in China, for now.

Current measurements of AI's impact aren't telling the full story. Google has offered a new method it hopes to standardize.

• AWS CEO Matt Garman says younger workers are the most engaged with AI
• They’re also cheap to train up with skills for the future
• Workers need to be prepared to learn, and for change

Amazon Web Services (AWS) CEO Matt Garman has called the idea of firing junior workers because AI can do their jobs “the dumbest thing I’ve ever heard”.

Speaking in a YouTube interview with Matthew Berman, Garman backed up his position by arguing junior staff are usually the least expensive, but also the most engaged with AI tools, therefore they’re best set up to feel the boost of AI in their workflows - rather than be replaced by it.

Taking a longer-term look at the scenario, Garman added without junior employees gaining experience, companies will have no skilled workforce in the future.

AWS CEO doesn’t see AI replacing human roles

Consequentially, Garman advocated for the continued hiring of graduates to teach them core software and problem-solving skills to futureproof against future workplace challenges, like today’s skills shortages.

He still encourages the use of AI tools, but to assist rather than replace junior talent.

When asked about how much of AWS’s code is now AI-written (we know that at least a third of Google, Meta and Microsoft’s code is now AI-generated), Garman described the notion as a “silly metric,” but acknowledged over 80% of the company’s developers now use AI in their workflows across code generation and more.

Garman indicated “curious” workers will get the most out of AI – combining their own desire to learn more with Amazon training initiatives.

He added workers should “be flexible, be willing to learn, and be willing to accept that their job may be a little bit different,” noting that this is a time of transition.

Broadly positive for the future of jobs, Garman’s thoughts align with Amazon CEO Andy Jassy’s thoughts on AI’s impact on the workforce.

In June 2025, ex-AWS CEO Jassy acknowledged that AI will ultimately end up replacing some human workers, but it would free up other jobs to avoid a net loss of roles.

You might also like

• Prepare for the future and start using the best online learning platforms now
• Amazon says it expects to cut human workers and replace them with AI
• These are the best job sites and best recruitment platforms

• Apple Watch fans are due to get the blood oxygen feature back on certain watches, after a lawsuit by health tech company Masimo
• The feature now involves a slight workaround, in which the measurement can only be displayed on an iPhone
• Masimo is suing the U.S. Customs Department for allowing the workaround to go ahead

Apple Watch users in the US are finally about to get a health feature back, as the watchOS 26 public beta launched with access to the blood oxygen feature – sort of.

The feature was removed for US fans after health technology company Masimo sued Apple, claiming it willfully violated Masimo's intellectual property by way of its patented blood oxygen feature. As a result, the feature was temporarily disabled on the Apple Watch Series 9, Apple Watch Series 10, and Apple Watch Ultra 2.

However, a workaround meant that the Apple Watch was able to add the feature back in via its new watchOS 26 software, and it's already arrived on some watches participating in the Apple public beta program.

The workaround is that while the Apple Watch can record blood oxygen, it can only present that information in the Health app on an iPhone.

Masimo doesn't like this and has filed another lawsuit – this time targeting US Customs, rather than Apple itself. Masimo is suing the US government for (according to the lawsuit via this BGR report), "unlawfully let[ting] Apple Inc. reactivate a blood-oxygen tracking feature on Apple Watches that infringes patents for the technology".

Will I get blood oxygen tracking on my Apple Watch?

We don't know. At the moment, if you're signed up to the public beta program, you should already have the feature.

Whether it will survive until the wider rollout in September, or whether the US government will feel pressured by this suit to take action and ban the workaround, remains to be seen.

You might also like...

• Garmin watches could finally get a key feature Apple's had for years – here's why it matters
• I review fitness trackers for a living – here's the best one I've tried in 2025
• The Apple Watch may get a major redesign – but not in time for the Apple Watch 11

In an exclusive interview with CNET, Honor's President of Product Fei Fang reveals how the V5's AI model will allow for more speed, accuracy and privacy.

• Apple TV+ has announced an immediate price increase of 30%
• Existing subscribers will be impacted within 30 days of their next renewal date
• Currently, Apple TV+ is the only major streaming service without an ad-supported tier

Apple TV+ has shocked subscribers today (August 21) with an immediate price hike that will see subscription costs jumping from $9.99 to $12.99 in the US.

The 30% increase will be applied immediately for new subscribers, with existing ones getting more leeway. They will see the higher price within 30 days of their next renewal date.

So if you are an existing customer, you might want to make the most of the best Apple TV+ shows and best Apple TV+ movies before the change happens.

Apple TV+ is home to acclaimed shows like Severance, which was recently renewed for season 3. We have also seen significant hype for Breaking Bad creator Vince Gilligan and his new series, Pluribus, which has a creepy number you can call. But the buzz around these two shows might not be enough to convince subscribers to stay.

Everything we know about Apple TV+'s price hike

Severance is Apple TV+'s most popular show. (Image credit: Apple TV Plus)

It's not just the US that is affected by the change. Apple TV+'s increase will also affect global customers, and we have an outline of the price changes below, which appear to vary by territory.

Country

Old price

New price

US

$9.99

$12.99

UK

£8.99

£9.99

Australia

$12.99

$15.99

Unsurprisingly, subscribers have reacted badly to the news, and some have already canceled. This does make me worry about Apple TV+'s future, and it comes as a disappointment after we crowned the service Streaming Service of the Year in 2024.

With many of the best streaming services offering ad-supported tiers as a cheaper option, I wouldn't be surprised if Apple TV+ is forced to follow suit.

According to Variety, "In justifying the price hike, Apple noted that it has launched a slew of TV shows and movies and that the service doesn’t include any ads. Indeed, Apple TV+ remains the only major streaming service that does not offer a price-reduced ad-supported plan."

But this decision hasn't gone down well. Here are some of the annoyed responses after Apple's announcement.

Apple TV is great quality but the price isn’t justifiable… no huge back catalog. this is just greed. If they had somewhat of a bigger back catalog then I’d be more open to paying but sorry gonna cancel.August 21, 2025

I may have to cancel this when I finish the shows I like. They have much less content than the other services. Apple TV Plus Raises Price on Monthly Plan - CNET https://t.co/0Y3oyEdHapAugust 21, 2025

I was just thinking I should cancel apple tv. This will do it for me.August 21, 2025

Right now, Apple TV+ seems to have no intentions of adding ads, but further backlash may change things in the future.

I like a lot of the shows on there, personally, but I'm not sure I'll be keeping my subscription if prices continue to go up.

You might also like

• When will Foundation season 3 episode 7 be released on Apple TV+?
• Murderbot season 2: everything we know so far about the hit Apple TV+ show's return
• Apple TV+ just dropped a teaser for Pluribus, the show from the creator of Breaking Bad, and we're left with many questions

Do you embrace being a Virgo or Pisces? Netflix's new astrology collection is for you.

Millions of audio recordings of hundreds of bird species have revealed that artificial light is making the birds wake up earlier and go to bed later.

(Image credit: Rasit Aydogan)

The new AI model may help NASA and NOAA better forecast auroras.

GPT-5 is now accessible across all platforms, getting a truly universal rollout in Microsoft.

• Blackview Active 12 Pro rugged tablet has a built-in projector and 108MP camera
• Tablet includes detachable kickstand, plus 400 lumen LED camping light on rear
• Tablet’s huge 30,000mAh battery supports extended use in the field

Blackview has launched the Active 12 Pro, a rugged Android tablet which combines an 11-inch screen with unusual extras, including a built-in projector, camping light, and a 108-megapixel camera sensor.

PCWatch reviewed the device, and called it one of the boldest tablets Blackview has ever produced.

The Active 12 Pro runs Android 15 on a MediaTek Dimensity 7300 processor with up to 16GB of RAM and 1TB of storage.

Big battery

We’ve seen a good number of rugged tablets with integrated projectors in recent years, including the 8849 TANK Pad, and the one in this tablet is capable of displaying a full HD image up to 120 inches.

With 200 lumens of brightness, autofocus, and keystone correction, it’s designed for entertainment and small presentations.

The tablet has a huge 30,000mAh battery, which in PCWatch’s testing supported extended video playback without rapid drain. Unlike many portable projectors, the image remained bright and usable in a dimly lit room.

The rugged tablet is IP68 and IP69K rated for water and dust resistance and meets MIL-STD-810H standards for durability.

At over 1.5kg it is far heavier than most 11-inch devices, but the tradeoff is a chassis that can withstand drops, pressure, and temperature extremes.

A detachable kickstand and optional straps give it flexibility for outdoor or field use.

The display itself is a 1920x1200 IPS panel with a 90Hz refresh rate. Alongside the 108-megapixel Samsung rear camera, a 50-megapixel front camera is included for video calls and photography.

The rear is also home to a large LED camping light with up to 400 lumens of brightness, further adding to the Active 12 Pro’s outdoor credentials.

In PCWatch’s performance benchmarks, the tablet handled demanding mobile games at medium settings without slowdown, aided by advanced cooling with heat pipes and a built-in fan.

Despite its slightly unusual design, the device delivered a credible mix of performance and versatility.

Pricing for the Active 12 Pro starts at around $768 for the 12GB/256GB model or $826 for the 16GB/1TB version on AliExpress, with discounts available.

While it is perhaps not aimed at casual users, the Active 12 Pro shows how rugged tablets are evolving into multi-purpose devices.

You may also like

• These are the best rugged phones for tough environments
• And these the best rugged tablets for when you need larger screens
• This outlandish rugged tablet has a laser projector and waterproof speakers

The Fed chair will speak Friday at an annual economic symposium in Jackson Hole, Wyo. The speech comes as the central bank is under mounting pressure from the White House to lower interest rates.

(Image credit: Kent Nishimura)

• Pixel 10 will not have physical SIM cards slots, Google confirms
• Buyers can transfer eSIMs from Apple to Android and vice-versa
• However eSIM adoption is uneven and can pose a risk for unprepared travelers

Google has confirmed eSIMs will replace physical SIM card slots in its new Pixel 10 smartphone range.

Reports from Android Authority claim Google's implementation may also support bi-directional transfers, meaning users can transfer an eSIM from an Apple device to an Android device or vice versa.

For now, this change appears to apply only to US models, but come after Apple made a similar move with iPhone 14, and now Google is following suit, further indicating eSIMs are revolutionizing the industry in a significant way.

What does this mean for travelers?

eSIMs for international travel have already become the go-to solution for many and now Google's latest move suggests the beginning of a new era.

If major firms like Apple and Google are committing to eSIM-only devices and other tech giants like NordVPN and IPVanish are entering the eSIM market, then it is safe to say that eSIMs have become the norm.

On iPhones, the process could be handled through the “Transfer to Android” option in settings, while Pixel 10 devices may rely on the Android Switch app, either by scanning a QR code or initiating a manual connection session.

Android Authority also reports that the feature may only be supported in iOS 26, expected to launch alongside the iPhone 17 in September 2025. However, these details remain speculative for now, and we will have to wait for the official rollout to see what holds true.

“Going eSIM-only makes it faster and easier to connect to a network. You can download a plan in minutes without fiddling with a tiny SIM card, which is a huge win for convenience", says Sarah McGarr CEO of Sim Local.

For travelers, this shift means getting comfortable with digital connectivity is no longer optional. As physical SIM slots disappear, checking for compatability, understanding how to activate, transfer, and manage eSIMs will be essential for staying connected seamlessly across borders.

"Check the best local eSIM plans, pick the best deal, and be online in minutes. Google’s eSIM move will only make more people expect this kind of convenience, and it’s up to the industry to make sure travelers can always find easy and affordable ways to stay connected", adds McGarr.

However, the biggest challenge to this revolution is that adoption is uneven. The ability to use an eSIM doesn’t just depend on owning a compatible device, it also hinges on whether local carriers support the technology and what plans they offer.

Travelers need to be prepared and do their research on carrier compatibility before flying and keep a global eSIM app as backup.

"The ability to use an eSIM is not so much country-dependent as it is device-dependent and network-dependent. That means a traveler could land somewhere only to find that their mobile network doesn’t yet support eSIM, or that their plan options are limited. In those cases, people may end up paying more in roaming fees or have fewer choices than they expected", concludes McGarr.

You might also like

• UK travelers are on track to waste over £350 million in roaming fees over the summer - I spoke with Holafly eSIM on the ultimate travel hack to beat the charges
• We've covered the best eSIMs for Europe
• Heading to the States? Check out our roundup of the best eSIMs for USA

• King of Meat launches on October 7, 2025, for PlayStation 5, Xbox Series X and Series S, and PC
• The four-player co-op platformer will have a Standard and Deluxe Edition, which are now available for pre-order
• A King of Meat Technical Test will also be available to play this weekend

Amazon Games and Glowmade have announced that King of Meat will officially launch on October 7, 2025, for PlayStation 5, Xbox Series X, Xbox Series S, and PC via Steam.

King of Meat is a four-player co-op party platformer featuring hack-and-slash mechanics and over 100 dungeons filled with puzzles, challenges, and hordes of monsters to defeat.

The online game, which also got a brand new gameplay trailer, was announced last year, but pre-orders for the Standard and Deluxe Editions are now available.

The King of Meat: Standard Edition will cost £23.95 (around $32.24) while the Deluxe Edition will cost £39.95 ($53.78), and both will include the Venerable Defender Legendary Costume, a samurai-style outfit with matching sword, shield, and stickers, as well as five days of early access to King of Meat from October 2.

The King of Meat: Deluxe Edition also features the Tapeboard Betsy and Emerson Shard Character Sets, "for players to enter the dungeons at launch with style and flair."

Ahead of release, Glowmade has also announced that a King of Meat Technical Test will be available to play this weekend, on August 23 at 10 AM PT / 6 PM BST until August 24 at 10 AM PT / 6PM BST.

The test will be playable for free on PC, PS5, Xbox Series X, and Xbox Series S, giving players an early look at the game and an opportunity to provide feedback to the developers.

Fans can also tune in to a partnered Twitch channel for 30 minutes during the Technical Test to claim their Beta Rewards Top, which will carry over into the live game.

King of Meat is also expected to come to Nintendo Switch, although there is no release date for that version at this time.

Glowmade has also confirmed earlier this year that King of Meat will offer both quality and performance options across PC and consoles, including PS5 Pro, that players can customize.

You might also like...

• Everything announced at Gamescom Opening Night Live 2025 from the Sekiro anime to the new Ghost of Yotei mode
• New Resident Evil Requiem gameplay teaser solidifies it as my most anticipated game of 2026 so far
• The Hollow Knight: Silksong appearance at Opening Night Live 2025 was disappointing, but thankfully there's more information on the way

• Security researchers found that a popular free Chrome VPN extension has been stealthily taking screenshots of every website its users visit
• FreeVPN.One boasts over 100,000 downloads and Google's "Verified" and "Featured" badges
• Once limited to VPN functionality, the extension's later updates were designed not for privacy, but to evade detection

A popular Chrome VPN extension has been found to spy on its over 100,000 users.

Koi Security published an extensive report, warning users that instead of protecting their data as the best VPN apps should do, FreeVPN.One takes screenshots of every single website they visit.

FreeVPN.One is a free VPN extension that managed to obtain two badges from the Google Chrome Store, supposedly to further assure users of its security. Unfortunately, those badges might have been mistakenly awarded.

A trustworthy VPN extension turned into a privacy nightmare

(Image credit: Google)

FreeVPN.One looks legitimate at a glance. With over 100,000 downloads and Google's endorsement, it seems like a secure alternative to paid VPN services. However, according to Koi Security, there's a sinister reality hiding just underneath the surface.

The extension was marketed as a free, unlimited VPN service, and according to researchers, for a while, it did just that. However, future updates introduced a worrying development.

Koi Security reports that just seconds after any page loads, a background trigger grabs a screenshot of it. This means that every website you visit with the extension enabled is captured, including private photos, banking sites, medical records, and all kinds of other sensitive information.

The screenshots are then sent to an external source, with scripts running to ensure the quality of data capture.

Even private pictures were sent to FreeVPN.One’s backend, as per Koi Security findings (Image credit: Koi Security)

Another feature in FreeVPN.One, called "AI Threat Detection," also takes screenshots and uploads them for server-side analysis, but at least this one discloses that it does that, whereas the rest of the app does not.

A few of the most recent updates to FreeVPN.One appears to have made things worse for user privacy. The spyware, screenshots, and location tracking allegedly started in July of this year. Later that same month, Koi Security reported, the developer upped the security to evade detection, all the while the screenshots continued being collected.

Koi Security reached out to the developer, who claims that screenshots are only taken if the domain appears suspicious. However, even websites like Google Photos are recorded, which doesn't line up with that claim. The dev also says that the screenshots are not stored anywhere, but this is a difficult claim to verify. They eventually allegedly stopped responding to Koi Security's requests for comment.

How to stay safe?

This situation outlines a larger problem with certain free VPNs, as well as the ease with which certain developers can upload malicious apps into web stores and gain user trust. Just recently, a VexTrio Viper, a group of cybercriminals, successfully shared a long list of malicious apps through popular app stores. This included a free VPN, ad blocker, and even an online dating service.

FreeVPN.One scored Google Play Store's "Verified" badge, which should have meant it was definitely secure, but it sounds rather like "it's best avoided."

If you used the FreeVPN.One extension, we recommend that you uninstall it immediately. Consider downloading one of the best antivirus programs to give your PC a thorough scan, too.

Next, it's time to change your passwords to just about every website you may have accessed while using the extension. Better to be safe than sorry.

This highlights just how important it is not to take chances when using a VPN service, which, by default, often has pretty hefty permissions when it comes to monitoring what you do. If you aren't willing to get a premium service, turn to some of the best free VPN services to ensure they have been thoroughly tested in order to protect your data.

You might also like

• "Scammers scamming other scammers" – This free Android VPN was caught using Windscribe-stolen servers
• To pay or not to pay? Nearly 1 in 4 TechRadar readers say they use free VPNs despite the risks
• This fake VPN could have been spying on you all this time