Feed aggregator

• xAI’s Grok 3 chatbot has added a voice mode with multiple personalities
• One personality is called “unhinged” and will scream and insult you
• Grok also has personalities for NSFW roleplay, crazy conspiracies, and an “Unlicensed Therapist” mode

xAI’s Grok may be about to start remembering your conversations as part of a broader slate of updates rolling out, all of which seek to match ChatGPT, Google Gemini, and other rivals. Elon Musk’s company tends to pitch Grok as a plucky upstart in a world of staid AI tools; it also seems to be aiming for parity on features like memory, voice, and image editing.

As spotted by one user on X, it appears that Grok will get a new "Personalise with Memories" switch in settings. This would be a big deal if it works and mark a shift from momentary utility to long-term reliability. Grok's reported memory system, which is still in development but already appearing in the web app, will allow Grok to reference previous chats.

This means if you’ve been working with it on something like planning a vacation, writing a screenplay, or just keeping track of the name of that documentary you wanted to watch, Grok could say, “Hey, didn’t we already talk about this?”

Grok’s memory is expected to be user-controlled as well, which means you’ll be able to manage what the AI remembers and delete specific memories or everything Grok has remembered all at once. That’s increasingly the standard among AI competitors, and it’ll likely be essential for trust, especially as more people start using these tools for work, personal planning, and remembering which child prefers which bedtime story.

This should put Grok more or less on par with what OpenAI has done with ChatGPT’s memory rollout, albeit on a much shorter timeline. The breakneck pace is part of the pitch for Grok, even when it doesn't quite work yet. Some users have reported already seeing the memory feature available, but it's not available to everyone yet, and the exact rollout schedule is unclear.

Remember Grok

Of course, giving memory to a chatbot is a bit like giving a goldfish a planner, meaning it’s only useful if it knows what to do with it. Even so, xAI seems to be layering memory into Grok Web in tandem with a handful of other upgrades that lean toward making it feel more like an actual assistant and less like a snarky trivia machine.

This memory update is starting to appear as a range of other Grok upgrades loom on the horizon. Grok 3.5 is expected any day now, with Grok 4 slotted for the end of the year.

There’s also a new vision feature in development for Grok’s voice mode, allowing users to point their phones at things and hear a description and analysis of what's around them.

It's another feature that ChatGPT and Gemini users will find familiar, and Grok’s vision tool is still being tested. Upgrades are also coming to the recently released image editing feature that lets users upload a picture, select a style, and ask Grok to modify it.

It’s part of the ongoing competition among AI chatbots to make AI models artistically versatile. Combine that with the upcoming Google Drive integration, and Grok starts to look a little more serious as a competitor.

Also on the horizon is Grok Workspaces, a kind of digital whiteboard for collaborating with Grok on a more significant project. These updates suggest that xAI is pivoting to make Grok seem less like a novelty and more like a necessity. xAI clearly sees Grok’s future as being more useful than just a set of sarcastic and mean voice responses.

Still, even as Grok gains these long-awaited features, questions remain about whether it can match the depth and polish of its more established counterparts. It’s one thing to bolt a memory system onto a chatbot. It’s another thing entirely to make that memory meaningful.

Whether Grok becomes your go-to assistant or stays a curious toy used only when some aspect goes viral depends on how well xAI can connect all these new capabilities into something cohesive, intuitive, and a little less chaotic. But for now, at least, it finally remembers your name.

You might also like

• ChatGPT's memory upgrade might just be the biggest AI improvement we see all year
• I tried Grok’s new AI image editing features – they’re fun but won’t replace Photoshop any time soon
• Grok 3’s voice mode is unhinged, and that’s the point

After almost a decade of tweaking the same formula, Overwatch 2 devs have cooked up something wildly new for season 16.

Americans in at least nine states qualify for automatic IRS tax filing extensions, according to the agency.

(Image credit: Kayla Bartkowski)

Measles is an extremely contagious disease. It's also extremely preventable. There's a vaccine. It's highly effective.

For decades it has made measles outbreaks in the U.S. relatively rare, and measles deaths rarer still. But the U.S. has now seen more than 700 measles cases this year, and 3 deaths so far with active outbreaks across six states.

The federal response is under scrutiny because Health and Human Services Secretary, Robert F. Kennedy, Jr. has made a career spreading false information about vaccines.

What are this administration's views on vaccines, and what do they mean for what is already one of the worst U.S. measles outbreaks this century.

Kennedy publicly promised he would support vaccines. Dr. Peter Marks, who was forced out as the nation's top vaccine regulator says his department isn't doing enough.

For sponsor-free episodes of Consider This, sign up for Consider This+ via Apple Podcasts or at plus.npr.org.

Email us at considerthis@npr.org.

(Image credit: Susan Walsh)

• A new leak claims to detail the specs and pricing of AMD's upcoming Radeon RX 9060 XT GPU
• The price could be as low as $269 in the US for the 8GB model
• Performance is claimed to sit between the Nvidia RTX 4060 Ti and AMD Radeon RX 7700 XT

We might finally have a proper idea of what AMD's long-awaited Radeon RX 9060 XT GPU is going to be capable of - and it's looking like it could be worthy of induction into the best graphics card hall of fame.

Thanks to leaked information shared by Moore's Law Is Dead (MLID) on YouTube, we now know that the RX 9060 XT could be shipping with 32 compute units, a 128-bit memory bus with both 8GB and 16GB VRAM models available, and a TDP between 150W and 200W.

Now, that's not a whole lot to go off at this point, but MLID's leak also notes that performance is 'above RTX 4060 Ti, but likely below RX 7700 XT'. Okay, so that doesn't sound that impressive, but here's the thing: the leak also includes pricing details, and it's looking very attractive on that front.

The 8GB model will reportedly cost between $269 and $299, while the 16GB version is claimed to be priced at $329-379. It's unclear whether these are potential price ranges AMD itself is still contemplating, or projected price ranges for AIB partner cards. For reference, the RTX 4060 Ti - which it supposedly outperforms - originally retailed at $399, and that was almost two full years' worth of GPU price inflation ago.

Battle of the budgets

Even if AMD hits us with the high end of these price estimates, it'll be a seriously competitively priced desktop GPU. Although Nvidia has been dominating the high-performance end of the market with its RTX 5000 cards, there's no denying that AMD's biggest rival has been fumbling its budget game lately, leaving Intel (isn't that weird) as one of the best options in the affordable PC gaming space thanks to its excellent Arc B580 card.

If Team Red can deliver RXT 4060 Ti-beating performance at a $269 price point, well, it's game set and match for the budget GPU market - at least, for the time being. It's worth noting, though, that MLID's source does say 'original plan as of a month ago' regarding these projected prices, likely in reference to the recent tech pricing carnage caused by Donald Trump's tariff war. So... watch this space, I guess.

(Image credit: AMD)

Another interesting point raised by the leaker was that the RX 9060 XT might lack hardware encoders - it's possible that to keep the price down, AMD has opted to include only the free-to-implement AV1 video encoder, not the (arguably better) H.264 and H.265 encode functions, since those require a licencing fee.

The leaker claims there are no plans for a 'non-XT' RX 9060 card, which tracks given the two different VRAM versions and the suggested $269 base price. Additionally, AMD might be considering a Radeon RX 9050 XT, presumably priced somewhere in the $200-250 range.

You may also like...

• Microsoft is digging its own grave with Windows 11, and it has to stop
• Windows 11 is getting a very handy change to the taskbar, as Microsoft takes a leaf from Apple’s Mac playbook
• Microsoft looks to be making a big change to how you install and log in to Windows 11 – and I’m not happy about it at all

One UI 7 was designed to showcase Samsung's AI ambitions, but a bug now appears to be getting in the way of momentum.

Apple Intelligence has not had the best year so far, but if you think Apple is giving up, you're wrong. It has big plans and is moving forward with new model training strategies that could vastly improve its AI performance. However, the changes do involve a closer look at your data – if you opt-in.

In a new technical paper from Apple's Machine Learning Research, "Understanding Aggregate Trends for Apple Intelligence Using Differential Privacy," Apple outlined new plans for combining data analytics with user data and synthetic data generation to better train the models behind many of Apple Intelligence features.

Some real data

Up to now, Apple's been training its models on purely synthetic data, which tries to mimic what real data might be like, but there are limitations. In Genmoji's, for instance, Apple's use of synthetic data doesn't always point to how real users engage with the system. From the paper:

"For example, understanding how our models perform when a user requests Genmoji that contain multiple entities (like “dinosaur in a cowboy hat”) helps us improve the responses to those kinds of requests."

Essentially, if users opt-in, the system can poll the device to see if it has seen a data segment. However, your phone doesn't respond with the data; instead, it sends back a noisy and anonymized signal, which is apparently enough for Apple's model to learn.

The process is somewhat different for models that work with longer texts like Writing tools and Summarizations. In this case, Apple uses synthetic models, and then they send a representation of these synthetic models to users who have opted into data analytics.

On the device, the system then performs a comparison that seems to compare these representations against samples of recent emails.

"These most-frequently selected synthetic embeddings can then be used to generate training or testing data, or we can run additional curation steps to further refine the dataset."

A better result

It's complicated stuff. The key, though, is that Apple applies differential privacy to all the user data, which is the process of adding noise that makes it impossible to connect that data to a real user.

Still, none of this works if you don't opt into Apple's Data Analytics, which usually happens when you first set up your iPhone, iPad, or MacBook.

Doing so does not put your data or privacy at risk, but that training should lead to better models and, hopefully, a better Apple Intelligence experience on your iPhone and other Apple devices.

It might also mean smarter and more sensible rewrites and summaries.

You might also like

• 3 things Apple needs to do at WWDC 2025 to save Apple Intelligence
• How to turn off Mail categories on iPhone, or customize them to your needs
• Apple officially delays the AI-infused Siri

While Xfinity is a great provider in Vancouver, there are other internet providers in the area that might be a better fit for you. Here's what our experts could gather.

Apple is giving out special Global Close Your Rings Day pins on April 24. Here's how to get one before they're gone.

A viral "true crime" story was actually made up, generated by A.I. Reporter Henry Larson explores the ethical questions raised by this new frontier of content.

The latest leak suggests Apple has put its novel facial recognition technology into the long-rumored folding iPhone.

• GenAI can hallucinate open source package names, experts warn
• It doesn't always hallucinate a different name
• Cybercriminals can use the names to register malware

Security researchers have warned of a new method by which Generative AI (GenAI) can be abused in cybercrime, known as 'slopsquatting'.

It starts with the fact that different GenAI tools, such as Chat-GPT, Copilot, and others, hallucinate. In the context of AI, “hallucination” is when the AI simply makes things up. It can make up a quote that a person never said, an event that never happened, or - in software development - an open-source software package that was never created.

Now, according to Sarah Gooding from Socket, many software developers rely heavily on GenAI when writing code. The tool could write the lines itself, or it could suggest the developer different packages to download and include in the product.

Monitor your credit score with TransUnion starting at $29.95/month

TransUnion is a credit monitoring service that helps you stay on top of your financial health. With real-time alerts, credit score tracking, and identity theft protection, it ensures you never miss important changes. You'll benefit from a customizable online interface with clear insights into your credit profile. Businesses also benefit from TransUnion’s advanced risk assessment tools.

Preferred partner (What does this mean?)View Deal

Hallucinating malware

The report adds the AI doesn’t always hallucinate a different name or a different package - some things repeat.

“When re-running the same hallucination-triggering prompt ten times, 43% of hallucinated packages were repeated every time, while 39% never reappeared at all,” it says.

“Overall, 58% of hallucinated packages were repeated more than once across ten runs, indicating that a majority of hallucinations are not just random noise, but repeatable artifacts of how the models respond to certain prompts.”

This is purely theoretical at this point, but apparently, cybercriminals could map out the different packages AI is hallucinating and - register them on open-source platforms.

Therefore, when a developer gets a suggestion and visits GitHub, PyPI, or similar - they will find the package and happily install it, without knowing that it’s malicious.

Luckily enough, there are no confirmed cases of slopsquatting in the wild at press time, but it’s safe to say it is only a matter of time. Given that the hallucinated names can be mapped out, we can assume security researchers will discover them eventually.

The best way to protect against these attacks is to be careful when accepting suggestions from anyone, living or otherwise.

You might also like

• What are AI Hallucinations? When AI goes wrong
• Take a look at our guide to the best authenticator app
• We've rounded up the best password managers

As an avid credit card user and professional credit card reviewer heres how I get enough value from the Delta Reserve to justify its cost

Monday's WNBA Draft is expected to be the latest jewel in the crown of Bueckers, the 6-foot guard who barely a week ago led her UConn Huskies to their first national championship in a decade.

(Image credit: Roy Rochlin)

The government announced it is freezing more than $2.2 billion, hours after the university refused to make changes it said would "dictate what private universities can teach."

(Image credit: Brian Snyder)

The university refused to make changes it said would "dictate what private universities can teach" and "whom they can admit and hire," among other things.

(Image credit: Brian Snyder)

The Trump administration is undertaking shifts in U.S. foreign policy and that has meant big shifts at the State Department, which is in charge of that policy. The changes have veteran diplomats worried.

And the gutted aid agency USAID has been absorbed into the State Department. We'll see what the loss of USAID funding has meant for the search for truth about Syria's civil war.

(Image credit: Kevin Dietsch)

• Change Healthcare suffered a huge cyberattack in 2024
• Disruptions from that attack cost organizations millions
• The firm is now chasing repayment of loans to cover losses from these disruptions

UnitedHealth Group is “aggressively” going after small healthcare organizations that borrowed money following a huge cyberattack on its subsidy Change Healthcare.

The attack is said to have affected almost 190 million Americans, and was the largest US healthcare data breach ever, and was incredibly disruptive, with systems only fully restored 9 months later, costing over $2 billion to recover from.

After the attack, interest-free loans were offered by Change to help medical practices with short-term cash flow needs. The firm is now demanding these funds be “immediately” repaid, with some organizations asked to repay hundreds of thousands of dollars in just a few days.

Monitor your credit score with TransUnion starting at $29.95/month

TransUnion is a credit monitoring service that helps you stay on top of your financial health. With real-time alerts, credit score tracking, and identity theft protection, it ensures you never miss important changes. You'll benefit from a customizable online interface with clear insights into your credit profile. Businesses also benefit from TransUnion’s advanced risk assessment tools.

Preferred partner (What does this mean?)View Deal

Lost revenue

Optum, UnitedHealth’s financial arm, has now confirmed it will withhold separate funds until these loans are repaid.

Doctors with their own private practices used these loans to cover losses from the disruption following the cyber incident, which cost some hundreds of thousands - and some reportedly used personal savings to keep practices afloat.

It’s worth noting UnitedHealth has a net worth of over $470 billion (at the time of writing), and CEO Andrew Witty made over $23 million in compensation in 2023.

Optum has collected over $4.5 billion of the $9 billion debt, but since many practices lost so much in downtime thanks to the disruption, many will struggle to repay the money owed in the just 5 day timeframe Optum have imposed, with one doctor describing it as a “shakedown.”

UnitedHealth paid the ransomware attackers $22 million in cryptocurrency to recover its data - but the operation was still shut down in its entirety, and Change never got its data back. Medical data is, of course, extremely sensitive, and put anyone exposed at risk of identity theft or fraud.

Via CNBC

You might also like

• Take a look at our picks for the best malware removal software around
• Check out our choice for best antivirus software
• Cl0p resurgence drives ransomware attacks to new highs in 2025

Spring Hill is serviced by some large ISP like Spectrum, AT&T Fiber and T-Mobile. Find out which broadband option is right for your home.

With AI agents becoming an increasingly common sight in businesses everywhere, Google Cloud has become the latest major company to ramp up its efforts in the space.

At its Google Cloud Next 25 event, the company unveiled several upgrades to its Agentspace platform to make agent discovery and adoption easier.

Just to give things an extra boost, Google Cloud also announced a new partnership with Nvidia designed at making its offerings even more intuitive.

Google Agentspace expansion

Following the initial launch of Google Agentspace in December 2024, the updates were mainly aimed at making creating and deploying AI agents easier

This includes giving employees access to Agentspace search and analysis tools directly from the search box in Google Chrome. The multimodal search capabilities can help track down exactly the data needed within your business, or give customers access to the answers they need.

The search results can cover content from the web, or from your business' most commonly-used apps and software, including the likes of Google Workspace, Microsoft 365, and apps like Jira, Salesforce, or ServiceNow.

(Image credit: Google Cloud)

Elsewhere, workers can also use a new Agent Gallery to find and deploy new agents quickly as well as creating their own agents with the new no-code Agent Designer platform - or launch some of Google's latest own-brand offerings, Idea Generation agent and Deep Research agent.

In order to make sure all these new agents co-exist effectively, Google Cloud has also launched a new interoperability protocol called Agent2Agent, which it says, "will allow AI agents to communicate with each other, securely exchange information, and coordinate actions on top of various enterprise platforms or applications."

Built on existing standards to allow easier integration, the company has already signed up more than 50 partners for the launch, including enterprise heavyweights such as Salesforce, PayPal, Box, Atlassian and more.

Finally, to harness the power of some of the most powerful computing hardware around today, Google Cloud and Nvidia have signed a collaboration bring the former's AI models to Nvidia Blackwell HGX and DGX platforms, as well as Nvidia Confidential Computing.

“By bringing our Gemini models on premises with Nvidia Blackwell’s breakthrough performance and confidential computing capabilities, we’re enabling enterprises to unlock the full potential of agentic AI,” said Sachin Gupta, vice president and general manager of infrastructure and solutions at Google Cloud.

“This collaboration helps ensure customers can innovate securely without compromising on performance or operational ease.”

You might also like

• Developers are pretty optimistic about AI agents
• AI agents are proving remarkably popular - but firms still face a host of challenges
• These are the best collaboration tools for businesses