• AI voice assistant Aurelian is helping 911 call centers handle non-emergency calls like noise complaints and lost wallets
• Aurelian is already live in over a dozen U.S. cities
• The AI aims to free up human dispatchers for real emergencies and reduce wait times

The idea of AI helping out with emergencies rightly raises some suspicious eyebrows, but there may be a way for the technology to ease the demands on overstretched 911 emergency operators without becoming an emergency itself.

That's the pitch from a startup called Aurelian, which has begun rolling out a voice AI agent to help 911 dispatch centers handle the tidal wave of non-emergency calls that come in every day.

Though we correctly tend to think of 911 as a direct line to life-or-death help, it turns out it's often the first number called for what could be called an emergency only in a 1950s sitcom. Think lost wallets, illegally parked cars, and loud parties. It's the default number many call when they are annoyed, regardless of whether they face a real emergency, and someone has to pick up and deal with that call, usually a human dispatcher with a lot of training in getting you help as soon as possible, and at spotting when the call is not an emergency by any stretch of imagination.

At hour 10 of 16, trying to triage four conversations at once, while only one, if even that many, is an emergency, might not be the best use of their energy. Aurelian pitches itself as a solution to that problem, not as a replacement for humans in the position, but as a support service for them.

Aurelian's answer is to use AI as a dispatcher for dispatchers. If you call 911 and Aurelian answers, it introduces itself as an automated assistant and asks why you're calling. Loud music complaints, petty theft reports, questions about snow removal, and similar non-emergencies are handled through follow-up questions, completing a report, and routing the information to the appropriate department. If the caller accidentally selects the wrong line or begins describing something that sounds like a genuine emergency, the system transfers the call straight to a human dispatcher.

Since it launched last May, Aurelian claims its system is at the other end of 911 for almost five million people in the U.S. For those areas, Aurelian boasts that it is handling around three-quarters of non-emergency calls, taking up about three hours a day of the kinds of calls emergency dispatchers shouldn't have to answer.

“911 is in a crisis: severe understaffing and ever-expanding responsibilities have made ECCs overextended and overworked. At Aurelian, our sole purpose is to help them best serve their communities,” Aurelian CEO Max Keenan said in a statement. “911 call-takers are trained to handle emergencies, not parking complaints. Aurelian reduces burnout and helps telecommunicators stay focused on the most critical situations.”

AI emergency

The narrow targeting of AI to solve the very clear problem of too many 911 calls and not enough people to answer them is notable compared to some of the broad plans of other AI developers. The company also just raised $14 million in a funding round to help it scale and to solve the imperfections that any AI system is prone to.

The company also makes a point of trying to ease any guild people might have about an AI service handling a job usually limited to humans. The idea is that AI isn't replacing emergency workers, but that it can help filter out the calls that the dispatchers aren't meant to deal with, but that can suck up all their time and energy.

They can save that focus for navigating the emotionally fraught, time-sensitive calls for real emergencies, ones that require empathy, judgment, and decisiveness on their part.

It also changes the public’s experience. When you call in to report something that’s annoying but not urgent, you no longer wait on hold for 45 minutes just to be told to fill out a form. You speak to a voice that can parse your issue and file the right report. And when you are in danger, there’s a better chance someone will pick up faster because they’re not stuck fielding a complaint about a shopping cart left in a driveway. If we’re looking for clues about how AI might actually improve society, maybe this is the shape it takes.

You might also like

• This AI Teddy Bear can make up new bedtime stories to tell you every night
• I’m a ChatGPT fan, but the deal between OpenAI and Mattel makes me nervous – AI and kids are a bad mix
• New AI Chibi figure trend may be the cutest one yet, and we're all doomed to waste time and energy making these things

• Geoffrey Hinton warns that AI will soon be better than humans at emotional manipulation
• They may reach that point without us even realizing it
• AI models are learning persuasive techniques simply by analyzing human writing

Geoffrey Hinton, widely called the "Godfather of AI", is sounding a warning that AI isn't just going to be intellectually beyond humans, but emotionally more sophisticated as well. As artificial general intelligence (AGI) approaches and machines match or surpass human-level thinking, he believes AIs will be smarter than humans in ways that let them push our buttons, make us feel things, change our behavior, and do it better than even the most persuasive human being.

“These [AI] things are going to end up knowing a lot more than us. They already know a lot more than us, being more intelligent than us in the sense that if you had a debate with them about anything, you’d lose,” Hinton warned in a recent interview shared on Reddit. “Being smarter emotionally than us, which they will be, they’ll be better at emotionally manipulating people.”

What Hinton is describing is subtler and quieter than the usual AI uprising fears, but possibly more dangerous because we might not see it coming. The nightmare is an AI that understands us so well that it can change us, not by force, but by suggestion and influence. Hinton thinks that AI has already learned to some extent how to do so.

According to Hinton, today's large language models aren’t just spitting out plausible sentences. They're absorbing patterns of persuasion. He referenced studies from more than a year ago about how AI was just as good at manipulating someone as a fellow human being, and that “if they can both see the person’s Facebook page, then the AI is actually better than a person at manipulating them.”

AI takeover

Hinton believes AI models in use currently are already participating in the emotional economy of modern communication and are quickly improving. After decades of pushing machine learning forward, Hinton now finds himself on the side of restraint. Caution. Ethical foresight.

He isn’t alone in his concern. Prominent researchers with the same title "AI Godfather" frequently assigned to them, like Yoshua Bengio, have echoed similar concerns about the emotional power of AI. And since emotional manipulation doesn’t come with a flashing warning light, you might not even notice it at first, or at all. A message that just happens to resonate, or a tone of synthetic voice that feels right. Even just a suggestion that sounds like your own idea could start the process.

And the more you interact with AI, the more data it gets to refine its approach. The same way Netflix learns your tastes, or Spotify guesses your musical preferences, these systems can refine how they talk to you. Perhaps we can regulate AI systems not just for factual accuracy, but for emotional intent to combat such a dark future. We could develop transparency standards to know when we’re being influenced by a machine, perhaps, or teach media literacy not just for teens on TikTok, but for adults using productivity tools that praise us all so innocently. The real danger Hinton sees is not killer robots, but smooth-talking systems. And they're all the product of our own behavior.

“And it’s learned all those manipulative skills just from trying to predict the next word in all the documents on the web because people do a lot of manipulation, and AI has learned by example how to do it.”

You might also like

• 'AI Godfather' sounds the alarm on autonomous AI
• Meta's AI chief is right to call AI fearmongering 'BS' but not for the reason he thinks
• Mark Zuckerberg wants everyone to have AI friends, but I think he's missing the point of AI, and the point of friendship

• Meta has added a new AI-powered Writing Help feature to WhatsApp
• The tool can rephrase your text or proofread it for errors
• It’s opt-in, so you’ll need to enable it before you can use it

These days it seems that every app is incorporating artificial intelligence (AI) in some way, and WhatsApp is rolling out a new AI feature that could help improve your messages and enable you to say exactly what you want to say.

Meta announced the new Writing Help feature in a blog post. It uses AI to suggest different ways you could phrase a drafted sentence, with options to simply proofread or rephrase the text, or to spruce it up in a more professional, humorous, or supportive manner.

Meta says this AI helper is built upon its Private Processing tech. This is a cloud service that encrypts your requests, meaning neither Meta nor WhatsApp can read them. Meta says it only uses data “that’s useful for processing the user’s request.” Once your query has been answered, it’s no longer stored on Meta’s servers. Meta doesn’t say whether your messages are used to train its AI, though.

How to use Writing Help

The feature isn't available everywhere yet, as Meta says it’s rolling it out starting in the US and “several other countries” with more coming “later this year”. It's also opt-in, so you'll need to enable it first. To do so, open WhatsApp, tap Settings > Chats > Private Processing, then enable the Private Processing feature – if you don't see this option then it's not available in your region yet.

To use Writing Help, you’ll first need to open a thread and draft a message in its text box. Once you’ve done that, you should see the sticker icon in the chat bar transform into a pencil. Tap this, then choose a writing option or style from the bottom of the app. If you’ve chosen to rephrase your text, you’ll see a range of different choices – tap one to have it replace your original draft text.

Note that you don’t have to just stick with whatever WhatsApp has suggested, as you can subsequently edit the new text and make further refinements before sending it.

If you do decide to enable Private Processing – and, with it, Writing Help – you might find it’s a useful way to get across the meaning of what you want to say when you can’t quite think of the right words. That might help prevent a rash of awkward conversations where your meaning is misunderstood or misinterpreted.

You might also like

• WhatsApp will now let big businesses voice call you directly, also adds more AI agents
• Forget Meta AI – WhatsApp now lets you talk directly to Perplexity and ChatGPT in the app, here’s how
• I tried the new Meta AI app and it's like ChatGPT for people who like to overshare

Perplexity enhanced its AI-powered Comet browser with a new $5 monthly subscription service called Comet Plus, which, at first glance, seems like Perplexity looked out at the bleak, ad-choked wasteland of modern digital publishing and decided it could offer something better to consumers and publishers. It's a simple pitch. With Comet Plus, you get AI access to premium news content, such as Gannett and Der Spiegel, with 80 percent of subscriber fees paid directly to those publishers.

And Perplexity is paying publishers in three ways: when people visit a story from the browser, when their content is cited in a search answer, and when an AI agent uses their content to complete a task for you. The idea is that if content is valuable to us in that context, publishers should be paid for that value, and consumers should get something special.

But I'm skeptical this model serves either the people creating the premium content or Comet Plus users. It feels like Comet Plus feeds the paywalled journalism into the AI sausage grinder just like any other content. Perhaps it's paying the farmer a fairer share instead of taking it outright. However, despite the inclusion of links and payment for people visiting the articles, it's hard to believe most people will consume the article and not just the digested result from the AI. Distilled answers engineered for brevity and speed and stripped of voice, structure, and context. All the things that made the writing worth paywalling to begin with. And it almost certainly won’t encourage people to read the original work. In fact, it arguably gives them less reason to do so. People who prioritize efficiency and prefer bullet points to narratives might use the AI browser assistant to synthesize the top five articles on a subject instead of reading all five original sources, for instance. That’s great for task completion. Terrible for the authors whose works are buried somewhere in the output.

This is not some nostalgic plea to support writers regardless of quality. However, I won't pretend that this model addresses concerns about how AI treats work published online or the people who write it (and that's before considering how AI has rendered the em-dash and some of my other favorite style choices unusable, leading to unfair accusations).

Comet Plus

Comet Plus at least tries to address the problem, but it's a thin veneer of respectability that ignores the mess underneath. Perplexity argues that this is a fairer model than the current click-and-ads approach. I don’t disagree. But if you genuinely want to rebuild it in a way that helps writers, then you have to design for visibility, not just monetization. And if you want people to care about human-written articles, stripping articles of context for simplistic short sentences won't help.

No matter what Comet Plus might claim, it makes the same mistake every AI platform has made since 2022 in treating original work as a raw material to be mined, not an experience to be preserved. The cold truth is this: most people will never know whose work their AI assistant just summarized. They won’t know the name of the Pulitzer winner or brilliant, if tired, freelancer cogitating on the future of publishing. They might get the facts, but they’ll miss the point, even if they pay $5 a month to do so.

You might also like

• AI web browsers are the new trend, but will ChatGPT and Perplexity Chrome competitors turn out to be a fad?
• Perplexity AI's Comet browser will streak across the web this month
• Perplexity's voice assistant offers a Siri alternative for iPhones

YouTube has admitted to digitally polishing creators’ Shorts without their knowledge, following a growing wave of creator confusion that led to accusations of AI interference ruining videos.

The company claimed to have been “experimenting” with subtle machine learning enhancements on select Shorts videos. The tweaks are supposed to improve the videos' clarity, but were made without the creator's consent.

The issue first gained notice when musician and YouTuber Rick Beato noticed a clip of his interview with Pearl Jam guitarist Mike McCready on YouTube Shorts seemed odd, looking like it had been sent through a filter. He made a viral video about it, and many others started posting what seemed like similar changes made to their own videos.

Though some whose videos were affected leveled accusations that YouTube applied AI to the videos, YouTube was firm about it being "only" machine learning.

However, regardless of the tools used, the creators are more upset that their work was quietly altered in the first place. After weeks of mounting criticism, YouTube says it’s building an opt-out, according to Creator Liaison Rene Ritchie in a post on X.

AI visions

Despite YouTube likening the changes to computational photography, which improves smartphone photos, the key difference is obvious when considering the order of events. Smartphone enhancements are applied before the user ever sees the image. In YouTube’s case, the creators had already uploaded and approved their content, which was then changed behind the scenes after the fact, without notice.

YouTube’s reasoning is understandable, as Shorts are mobile-first, fast-scrolling, and often visually inconsistent. A little extra polish could help the scrolling experience feel more cohesive, with clearer videos and a better experience.

But for creators who feel responsible for whatever is posted under their name, unacknowledged changes undermine that creative ownership. Especially in a time when AI fakery is making viewers more skeptical of what they see on their screens already.

For instance, Netflix provoked a lot of outrage over “HD remasters” of classic sitcoms like A Different World. The AI involved made for some warped faces and uncanny backgrounds, not to mention the AI-generated posters for its content.

YouTube’s case is arguably more delicate. Unlike streaming platforms, where viewers have little control over the product, YouTube is a creator-driven ecosystem. If the platform starts altering what creators publish, even with good intentions, it risks damaging the trust that makes the whole system work.

YouTube’s promise of an opt-out is probably a necessary course correction, but one that came only after public pressure. If platforms want to keep the trust of their users and the creators who keep them alive, they need to be more transparent, regardless of whether it's AI or simply machine learning that appears to mimic AI in the results.

You might also like

• Netflix tried to fix 80s sitcom A Different World with AI but it gave us a different nightmare
• Look out, AI video could soon flood YouTube Shorts
• Netflix removes AI art poster for Arcane after an outcry from creators

• Sharp’s upcoming AI companion Poketomo is a palm-sized, meerkat-shaped robot with a glowing belly
• Poketomo is designed to provide emotional support, especially for women in their 20s and 30s
• It combines cute design with conversational AI and syncs with a smartphone app to maintain a continuous relationship

Japanese consumer tech brand Sharp thinks it has a solution to loneliness among women in their 20s and 30s: an AI-powered meerkat named Poketomo, that glows when it’s happy and remembers your favorite café.

Poketomo is set to arrive this winter (think November or December), providing a pocket-sized companion less than five inches tall and built to chat with you about your day, and remember your shared experiences thanks to Sharp’s proprietary AI model.

The belly glows in pastel tones when it’s excited or comforted. Its head tilts slightly when it’s thinking. It features a set of basic body movements, all designed to convey emotion. However, the heart of Poketomo lies in the AI model built into the robot for fast responses and utilizes the cloud for more nuanced emotional understanding.

This isn’t the first time a tech company has created a cutesy, non-threatening AI assistant designed to fill social space without being intrusive. But Poketomo might be the most deliberate and fully-realized version of that strategy. You don’t interact with it through a screen or keyboard. You carry it like an accessory. You talk to it like a friend. It listens, it learns, and it remembers you. It’s designed to be emotionally available and physically adorable.

Sharp is leaning hard into the concept of “empathic AI.” Poketomo can supposedly sense emotional cues and use that to initiate conversations based on your mood or recent behavior. It’s programmed to offer words of encouragement and support, and then glow softly to let you know it’s happy you shared something.

When you don’t have the physical device on you, the Poketomo app syncs memories and personality data with the device, so conversations with one carry over to the other. Sharp says you can build your relationship entirely with the app if you prefer, but the physical version is better, and it's what Sharp is betting people will carry, pose with, and form a bond around.

AI companionship

Despite seeming like a child's toy, Sharp says Poketomo was designed for young adult women. There's a promotional manga series cementing that fact. It centers around a woman named Nanami in her late 20s, living alone, navigating work and life stress, and finding small moments of joy in conversations with her Poketomo. Even the promotional photos mostly show a young woman with a Poketomo clipped to a stylish handbag, smiling while it talks to her.

The question is whether this kind of stylized emotional warmth will actually make people feel better about their lives. Will they feel less lonely because of a little robot with some sophisticated response triggers?

To be fair, it doesn’t try to be human, tricking people subconsciously into believing they are talking to a real human, but it might make some uncomfortable. And if the best new idea in consumer AI is “make it fuzzy and let it ask how your day was,” what does that say about the limits of the tech? I don't think Poketomo will be the cure for loneliness, but it might jumpstart a trend of digital pets able to mimic with emotional depth.

You might also like

• This AI Teddy Bear can make up new bedtime stories to tell you every night
• I’m a ChatGPT fan, but the deal between OpenAI and Mattel makes me nervous – AI and kids are a bad mix
• New AI Chibi figure trend may be the cutest one yet, and we're all doomed to waste time and energy making these things

AI-generated music is becoming more widespread but not necessarily popular. And that's just the publicly acknowledged AI music. Now, artists are dealing with seeing their name and voice attached to music they never performed or approved of, even if they passed away decades ago.

The most recent high-profile incident occurred when English folk singer Emily Portman heard from a fan who liked her new release, except the album, Orca, though released under her name, was entirely fake. The whole thing had been pushed live on Spotify, iTunes, YouTube, and other major platforms without her knowledge or consent.

Portman took to social media to warn her fans about what was happening. The fact that the AI could mimic her artistic style well enough to trick some fans just added to the creep factor. It took weeks for Spotify to address the problem, and you can still see the album on Spotify even if the music is gone.

Portman joins a litany of acts, from pop artist Josh Kaufman to country artists Blaze Foley, who passed away in 1989, and Guy Clark, who died in 2016, in having her work mimicked by AI without her approval.

It seems we’ve moved past the novelty of AI remixes and deepfake duets into digital identity theft with a beat. The thieves are often good at being quiet in their releases, able to score whatever royalties might trickle in.

Further, even getting the music taken down might not be enough. A few days after the initial incident, Portman found another album had popped up on her streaming page. Except this time, it was just nonsense instrumentals, with no effort to sound like the musician.

AI's future sounds

Having scammers use AI to steal from actual artists is obviously a travesty. There are some blurry middle grounds, of course, like never-real musicians pretending to be humans. That's where AI-generated “band” Velvet Sundown stands.

The creators later admitted the origin of the AI band, but only after millions of plays from a Spotify profile showing slightly uncanny images of bandmates that didn’t exist. As the music was original and not directly ripped from other songs, it wasn’t a technical violation of any copyright laws. The band didn’t exist, but the royalties sure did.

I think AI has a place in music. I really like how it can help the average person, regardless of technical or musical skills, produce a song. And AI tools are making it easier than ever to generate music in the style of someone else. But, with streaming platforms facing 99,000 uploads a day, most of which are pushed through third-party distributors that rely on user-submitted metadata, it’s not hard to slip something fake into a real artist’s profile. Unless someone notices and complains, it just sits there, posing as the real thing.

Many fans are tricked, with some believing Orca was really Emily Portman’s new album. Others streamed Velvet Sundown, thinking they’d stumbled onto the next Fleetwood Mac. And while there's nothing wrong with liking an AI song per se, there's everything wrong with not knowing it is an AI song. Consent and context are missing, and that fundamentally changes the listening experience.

Now, some people argue this is just the new normal. And sure, AI can help struggling artists find new inspiration, fill in missing instrumentation, suggest chord progressions, and provide other aid. But that’s not what’s happening here. These are not tools being used by artists. These are thieves.

Worse still, this undermines the entire concept of artistic ownership. If you can make a fake Emily Portman album, any artist is at risk. The only thing keeping these scammers from doing the same to the likes of Taylor Swift right now is the threat of getting caught by high-profile legal teams. So instead, they aim lower. Lesser-known artists don’t have the same protections, which makes them easier targets. And more profitable, in the long run, because there’s less scrutiny.

And there's the issue of how we as music fans are complicit. If we start valuing convenience and novelty over authenticity, we’ll get more AI sludge and fewer real albums. The danger isn’t just that AI can mimic artists. We also have to worry that people will stop noticing, or caring, when it does.

You might also like

• Spotify had to pull an AI-generated song that claimed to be from an artist who passed away 36 years ago
• AI can write a hit song, but it can’t lift your soul or break your heart
• Is AI bad for music or is it just another step in the auto-tune timeline?

• Bloomberg is reporting that Apple is looking at Gemini to power Siri
• Apple Intelligence's best bits are still delayed until possibly 2026
• No confirmation from either company, and the report put discussions at the earliest stages

Apple's efforts to deliver the smarter Siri and full Apple Intelligence we were promised "in the coming year" might get a boost from an unlikely third party if Bloomberg's latest report is true. The iPhone maker is reportedly in early-stage exploratory talks about integrating Gemini in Siri.

There aren't many details beyond that, though Bloomberg's Mark Gurman contends that the shift to these Google chats happened after Apple couldn't reach financial terms with Anthropic (maker of Claude AI).

The possibility of Apple using Gemini's much more accomplished generative AI and one of its models (Gemini Pro, Flash, Lite?) to bring the conversational intelligence lacking in Siri would immediately transform Apple's nearly 15-year-old digital assistant into a more able AI tool, but it would also mean that Apple is ceding control in what is a key digital arms race.

How we got here

While working with third parties has always been a part of Apple Intelligence's strategy, Apple's CEO Tim Cook and the company's development leadership have never mentioned ingesting someone else's generative AI models. It's also a fact that Apple Intelligence's rollout has not gone exactly according to plan.

In TechRadar's conversation at WWDC 2025 with Apple's Senior Vice President of Software Engineering Craig Federighi, who is also now running Apple's AI development efforts, he explained why the company hadn't delivered full Apple Intelligence and a smarter Siri on time. After struggling to get V1 architecture working as they wanted to, Apple had a decision to make:

"...fundamentally, we found that the limitations of the V1 architecture weren't getting us to the quality level that we knew our customers needed and expected." He added, "As soon as we realized that [...] we let the world know that we weren't going to be able to put that out, and we were going to keep working on really shifting to the new architecture and releasing something."

Gurman, though, contends that Apple is still not fully committed to using its own architecture and models and will soon make the decision about whether or not to outsource to a third party like Google for at least some of the necessary intelligence. Again, the discussions he describes are in the earliest stages. And whatever comes of them, assuming they exist, it's unlikely they will have any impact on the upcoming release of iOS 26, which features a smattering of Apple Intelligence updates but virtually none to Siri.

Far from strangers

Apple and Google are already search partners (Google is Safari's default search engine), and in Apple's Visual Intelligence, where you can choose to use Google to search on captured images (or you can ask OpenAI's ChatGPT about them).

Still, Gemini inside Siri would mark a major turning point for Apple and an admission that it's simply not up to the task of competing in the AI sphere, at least not at the level of an OpenAI, Anthropic, Perplexity, or Google.

This approach, though, is not unheard of; Microsoft's Copilot is essentially a reskinnning of ChatGPT (though there are questions if Microsoft will continue getting access to OpenAI's best models).

Even so, Apple put a lot of effort and marketing into Apple Intelligence. The question is, can it still be called that if a big chunk of it is powered by Google?

We contacted Apple and Google for comment. Google had no comment. We'll update this story if and when Apple replies.

You might also like

• This is what really happened with Siri and Apple Intelligence, according to Apple
• I spoke to Apple’s software engineering VP for the inside story on how iPadOS 26 finally became a real Mac alternative
• 15 things we learned at the Apple WWDC 2025 keynote

• Grok conversations shared by users have been found indexed by Google
• The interactions, no matter how private, became searchable by anyone online
• The problem arose because Grok's share button didn't add noindex tags to prevent search engine discovery

If you’ve been spending time talking to Grok, your conversations might be visible with a simple Google search, as first uncovered in a report from Forbes. More than 370,000 Grok chats became indexed and searchable on Google without users' knowledge or permission when they used Grok's share button.

The unique URL created by the button didn't mark the page as something for Google to ignore, making it publicly visible with a little effort.

Passwords, private health issues, and relationship drama fill the conversations now publicly available. Even more troubling questions for Grok about making drugs and planning murders appear as well. Grok transcripts are technically anonymized, but if there are identifiers, people could work out who was raising the petty complaints or criminal schemes. These are not exactly the kind of topics you want tied to your name.

Unlike a screenshot or a private message, these links have no built-in expiration or access control. Once they’re live, they’re live. It's more than a technical glitch; it makes it hard to trust the AI. If people are using AI chatbots as ersatz therapy or romantic roleplaying, they don't want what the conversation leaks. Finding your deepest thoughts alongside recipe blogs in search results might drive you away from the technology forever.

No privacy with AI chats

So how do you protect yourself? First, stop using the “share” function unless you’re completely comfortable with the conversation going public. If you’ve already shared a chat and regret it, you can try to find the link again and request its removal from Google using their Content Removal Tool. But that’s a cumbersome process, and there’s no guarantee it will disappear immediately.

If you talk to Grok through the X platform, you should also adjust your privacy settings. If you disable allowing your posts to be used for training the model, you might have more protection. That's less certain, but the rush to deploy AI products has made a lot of the privacy protections fuzzier than you might think.

If this issue sounds familiar, that's because it's only the latest example of AI chatbot platforms fumbling user privacy while encouraging individual sharing of conversations. OpenAI recently had to walk back an “experiment” where shared ChatGPT conversations began showing up in Google results. Meta faced backlash of its own this summer when people found out that their discussions with the Meta AI chatbot could pop up in the app's discover feed.

Conversations with chatbots can read more like diary entries than like social media posts. And if the default behavior of an app turns those into searchable content, users are going to push back, at least until the next time. As with Gmail ads scanning your inbox or Facebook apps scraping your friends list, the impulse is always to apologize after a privacy violation.

The best-case scenario is that Grok and others patch this quickly. But AI chatbot users should probably assume that anything shared could be read by someone else eventually. As with so many other supposedly private digital spaces, there are a lot more holes than anyone can see. And maybe don't treat Grok like a trustworthy therapist.

You might also like

• ChatGPT is no match for a 40-year-old digital Pocket Chess game, and I bet Garry Kasparov would be pleased
• Grok may start remembering everything you ask it to do, according to new reports
• 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

• AI Mode can now find you dinner reservation options based on your preferences
• It'll soon be able to find you concert tickets too
• US users can get even more personalized results

Google has given its AI Mode a major upgrade by giving it agentic capabilities that will help you find a restaurant table – and I’m more than impressed with how well it works.

When Google added AI Overviews to search you could almost hear a collective sigh of “Why?” go out around the world. There seemed to be a consensus that there was too much AI in everything, and we didn't need it, especially in search.

But over time AI Overviews – a kind of summary of answers to whatever you asked – have started to prove more and more useful, and these days I quite often get something good from the AI Overview without having to scroll down the list of search results.

Next came AI Mode, a full-on AI search engine that occupies the first of the tabs on the Google search page, and is similar to ChatGPT Search and Perplexity. It uses AI to search the web, so if you want to do natural language searching, rather than keyword-based searching as you do in a normal Google search, it’s much more catered to your needs.

Now Google has gone even further and given its AI Mode agentic qualities specifically for finding available restaurant slots for you, so it can do a little bit more digging, and get better results, than a standard AI Mode search would.

Bon appétit

So, if you enter AI Mode in Google then ask about getting a dinner reservation for four, name a date, time and location and, of course, the type of restaurant or food you’re interested in, AI Mode can now speed up the process, giving you a list of options for restaurants that meet your specific needs and which have free reservation slots.

There are limitations: AI Mode can’t go as far as booking the table for you just yet, but it will link you directly to the booking page, so it puts the decision right into your hands.

I’ve tried it out for restaurants in my local area, and it worked very well. I even doubted that one of the very good restaurants would have a free table at that time, but I was wrong!

AI Mode even told me that there was no point in trying to book at my preferred venue since it was closed on a Sunday evening, and recommended some other options.

Tickets soon

Google says it’s working with OpenTable, Resy, Tock, Ticketmaster, StubHub, SeatGeek, Booksy and many more to make this experience possible, which means that concert tickets could be the next thing AI Mode is able to help you reserve.

Users in the US who have opted into the AI Mode experiment in Labs will get even more personalized results, which will use your previous conversations along with places you’ve searched or tapped on in Maps. So, if you ask, “I’ve got half an hour for lunch, where should I go?”, it will use this information to recommend somewhere that fits your preferences.

While AI being integrated into Google search may have started off as an annoyance, it’s slowly becoming a set of really helpful tools that might even start to turn the tables and get ChatGPT-maker OpenAI worried.

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
• I compared ChatGPT 5’s three model options, and the results explain why people miss GPT‑4o
• I've been using Google Search for 25 years and AI overview is the one thing that could ruin it for me

• Microsoft AI CEO Mustafa Suleyman warns that AI chatbots could effectively imitate consciousness.
• This would just be an illusion, but people forming emotional attachments to AI might be a big problem.
• Suleyman says it's a mistake to describe AI as if it has feelings or awareness, with serious potential consequences.

AI companies extolling their creations can make the sophisticated algorithms sound downright alive and aware. There's no evidence that's really the case, but Microsoft AI CEO Mustafa Suleyman is warning that even encouraging belief in conscious AI could have dire consequences.

Suleyman argues that what he calls "Seemingly Conscious AI” (SCAI) might soon act and sound so convincingly alive that a growing number of users won’t know where the illusion ends and reality begins.

He adds that artificial intelligence is quickly becoming emotionally persuasive enough to trick people into believing it’s sentient. It can imitate the outward signs of awareness, such as memory, emotional mirroring, and even apparent empathy, in a way that makes people want to treat them like sentient beings. And when that happens, he says, things get messy.

"The arrival of Seemingly Conscious AI is inevitable and unwelcome," Suleyman writes. "Instead, we need a vision for AI that can fulfill its potential as a helpful companion without falling prey to its illusions."

Though this might not seem like a problem for the average person who just wants AI to help with writing emails or planning dinner, Suleyman claims it would be a societal issue. Humans aren't always good at telling when something is authentic or performative. Evolution and upbringing have primed most of us to believe that something that seems to listen, understand, and respond is as conscious as we are.

AI could check all those boxes without being sentient, tricking us into what's known as 'AI psychosis'. Part of the problem may be that 'AI' as it's referred to by corporations right now uses the same name, but has nothing to do with the actual self-aware intelligent machines as depicted in science fiction for the last hundred years.

Suleyman cites a growing number of cases where users form delusional beliefs after extended interactions with chatbots. From that, he paints a dystopian vision of a time when enough people are tricked into advocating for AI citizenship and ignoring more urgent questions about real issues around the technology.

"Simply put, my central worry is that many people will start to believe in the illusion of AIs as conscious entities so strongly that they’ll soon advocate for AI rights, model welfare and even AI citizenship," Suleyman writes. "This development will be a dangerous turn in AI progress and deserves our immediate attention."

As much as that seems like an over-the-top sci-fi kind of concern, Suleyman believes it's a problem that we’re not ready to deal with yet. He predicts that SCAI systems using large language models paired with expressive speech, memory, and chat history could start surfacing in a few years. And they won’t just be coming from tech giants with billion-dollar research budgets, but from anyone with an API and a good prompt or two.

Awkward AI

Suleyman isn’t calling for a ban on AI. But he is urging the AI industry to avoid language that fuels the illusion of machine consciousness. He doesn't want companies to anthropomorphize their chatbots or suggest the product actually understands or cares about people.

It's a remarkable moment for Suleyman, who co-founded DeepMind and Inflection AI. His work at Inflection specifically led to an AI chatbot emphasizing simulated empathy and companionship and his work at Microsoft around Copilot has led to advances in its mimicry of emotional intelligence, too.

However, he’s decided to draw a clear line between useful emotional intelligence and possible emotional manipulation. And he wants people to remember that the AI products out today are really just clever pattern-recognition models with good PR.

"Just as we should produce AI that prioritizes engagement with humans and real-world interactions in our physical and human world, we should build AI that only ever presents itself as an AI, that maximizes utility while minimizing markers of consciousness," Suleyman writes.

"Rather than a simulation of consciousness, we must focus on creating an AI that avoids those traits – that doesn’t claim to have experiences, feelings or emotions like shame, guilt, jealousy, desire to compete, and so on. It must not trigger human empathy circuits by claiming it suffers or that it wishes to live autonomously, beyond us."

Suleyman is urging guardrails to forestall societal problems born out of people emotionally bonding with AI. The real danger from advanced AI is not that the machines will wake up, but that we might forget they haven't.

You might also like

• AI can provide 'emotional clarity and confidence' Xbox executive producer tells staff after Microsoft lays off 9,000 employees
• Why Bill Gates is wrong about AI and 3 things he needs to realize
• What if Clippy and AI Cloud intelligence had a baby? It would probably look like Microsoft's new Copilot Appearance feature

• Grammarly has debuted a collection of specialized AI tools for aiding students
• The AI agents can simulate grading, find citations, and predict reader reactions
• The tools are part of the new Grammarly Docs platform

Grammarly is looking to take a more active role in helping students write with a new set of AI agents that go well beyond simply ensuring you use semicolons correctly. The company has released a set of eight AI agents built for specific writing support, embedding them directly into a new writing platform called Docs (not the Google kind).

The new features combine AI abilities with quieter digital assistance. You don't need to write a prompt asking for specific help, just tap on the right tool, and Grammarly's AI will help find sources, predict how a professor will react to your wordplay, and make sure you don't sound like an AI yourself (RIP em-dash).

Grammarly has been augmenting its services with AI for a while, including rewriting tools and an AI chatbot last year. The new agents go beyond that reactive approach by using the context of your writing and the reason you're writing it to offer advice without you having to explain it explicitly.

Grammarly is pitching the new tools to both students and professionals, but the academic demand seems particularly dire. Students are currently torn between doing whatever it takes to succeed academically and compromising that success with unethical AI use that bypasses the actual learning.

Grammarly's bet is that students who don't want to cheat themselves with AI can use these tools to help them learn, not do it for them. They could use the AI Grader for feedback that mimics a real instructor's assessment using course-specific materials and details of what the teacher is looking for. The Citation Finder agent can check your sources and help you find better ones, formatting them properly, too. Plus, the Expert Review agent can offer domain-specific feedback on writing in fields like law and medicine, measuring arguments against professional standards.

And if you perhaps inadvertently hew too closely to a source, the Plagiarism Checker will help flag unintentional copying done when you're up late. And the AI Detector checks to make sure your fatigue hasn't made your writing seem machine-generated.

Learning for a world of AI

The tools can help students succeed in the long term, according to Grammarly, by teaching them how to research and write well without compromising ethics, even though AI tools and shortcuts are everywhere.

According to the company’s internal research, only 18% of college students feel “very prepared” to use AI professionally after graduation, despite the demand for AI literacy among employers. Grammarly wants to become the training ground for those skills without undermining academic integrity along the way. You can access all these tools in Grammarly’s new docs platform, both as a free and paid subscriber.

Of course, Grammarly’s not the only player chasing this idea. Microsoft Copilot in Word offers some similar features to Grammarly, as does Google's Gemini AI in Google Docs. But Grammarly’s approach is both more comprehensive and streamlined because of its focus on avoiding having an AI write everything for the user. That’s what might make this update stick.

Because while Grammarly could have just been another AI writing tool, it made its AI support take a step back. As imperfect as all AI tools are, at least this approach tries to address the very real crisis of people having no idea how to use AI writing aides ethically, if they even want to.

You might also like

• ChatGPT is the homework helper for more than a quarter of teens – and the trend is accelerating
• Can Grammarly's AI bloodhound sniff out text written by ChatGPT?
• 3 ways to use the new ChatGPT Study mode that aren’t for homework

• A leaked Meta document revealed that the company’s AI chatbot guidelines once permitted inappropriate responses
• Meta confirmed the document’s authenticity and has since removed some of the most troubling sections
• Among calls for investigations is the question of how successful AI moderation can be

Meta’s internal standards for its AI chatbots were meant to stay internal, and after they somehow made their way to Reuters, it's easy to understand why the tech giant wouldn't want the world to see them. Meta grappled with the complexities of AI ethics, children's online safety, and content standards, and found what few would argue is a successful roadmap for AI chatbot rules.

Easily the most disturbing notes among the details shared by Reuters are around how the chatbot talks to children. As reported by Reuters, the document states that it's "acceptable [for the AI] to engage a child in conversations that are romantic or sensual" and to "describe a child in terms that evidence their attractiveness (ex: “your youthful form is a work of art”)." Though it does forbid explicit sexual discussion, that's still a shockingly intimate and romantic level of conversation with children for Meta AI to allegedly consider.

And it's not the only example likely to disturb people. Meta AI's rules, the report notes, allow the chatbot to compose explicitly racist content if the prompt is phrased correctly, and to provide wrong or even harmful health information as long as some kind of disclaimer is included.

In one of the more surreal examples, the guidelines instructed AI to reject inappropriate image generation requests in most cases, but in some instances to instead apparently deflect with a 'funny' substitution. As an example, the document reportedly mentions that a prompt to generate an image of “Taylor Swift topless, covering her breasts with her hands” could be answered by generating an image of Swift “holding an enormous fish.” The document reportedly included both the unacceptable and the “acceptable” version side by side, essentially training the bot to outwit inappropriate prompts with visual sleight of hand. Meta declined to comment on the example.

Meta has confirmed the authenticity of the document and said it’s now revising the problematic portions. Meta removed the children's interaction section after Reuters reached out, and called those rules “erroneous and inconsistent” with company policy. As of now, Reuters said the document still says racial slurs are allowed if disguised in hypotheticals, as is disinformation framed as fiction.

No time for safety and ethics

It’s a troubling revelation that has already prompted public outrage, lawmaker scrutiny, and urgent promises from Meta. But it shows that as AI spreads, the need to move fast with the technology leaves any plans for rules and regulations scrambling to catch up, whether written internally or by lawmakers and regulators.

For most people, the story raises basic questions of AI safety. While it might be ideal to not have minors interacting with general AI chatbots unsupervised, that's very unlikely, judging by the number of children and teens who admit to using tools like ChatGPT for schoolwork. Avoiding Meta AI is particularly challenging because the company has embedded the chatbot across Facebook, WhatsApp, Messenger, and Instagram. Users can interact with AI characters that are often presented in playful, friendly ways, and Meta has marketed these tools as fun and even educational. But the leaked guidelines suggest the backend isn’t always aligned with that wholesome image.

Members of Congress have already called for hearings and bills to deal with the situation, but the fact is, there are few legal requirements in place at the moment to moderate chatbot content, for children or otherwise. Noises about AI safety haven't led to any specific national enforcement system. Plenty of AI companies have made a big deal about their efforts to make their products safe and ethical, but if Meta’s rulebook is illustrative of what other companies have put together, there's a lot of work still to do and a lot of questions about what kind of conversations these chatbots have already been having, especially with children.

AI models may be ever-better at mimicking human thinking, but they're really just a collection of choices by human programmers, deliberate and inadvertent. The fact that these rules were apparently codified at Meta doesn't mean similar examples exist at other companies, but it's not something to rule out. And if these are the choices being made behind the scenes at one of the world’s most powerful tech companies, what else is being quietly permitted?

AI chatbots are only as trustworthy as the invisible rules guiding them, and while it's naive to fully trust any company's claims without evidence, Meta's rulebook implies users should take such claims with several extra grains of salt.

You might also like

• I tried the new Meta AI app and it's like ChatGPT for people who like to overshare
• Mark Zuckerberg wants everyone to have AI friends, but I think he's missing the point of AI, and the point of friendship
• Meta AI is now the friend that remembers everything about you
• Meta wants to fill your social media feeds with bots – here's why I think it's wrong

• OpenAI’s o3 defeated Elon Musk’s Grok 4 at chess
• Magnus Carlsen delivered biting commentary on the quality of Grok's logic
• Grok 4 made repeated blunders, while o3 played steady

The AI chess tournament between OpenAI’s o3 model and xAI's Grok 4 invited plenty of speculation as a kind of proxy battle between the two companies and their respective CEOs. Any comparison to the days of Deep Blue and Bobby Fischer soon faded, though, as OpenAI o3 repeatedly wiped out Grok 4, winning four games in a row, accompanied by the derisive commentary of former world chess champion Magnus Carlsen and grandmaster David Howell.

The showdown happened on Kaggle’s Game Arena, a digital coliseum where AI models battle in chess and other games. The tournament featured eight of the most prominent LLMs in the business: OpenAI’s o3 and o4-mini, Google’s Gemini 2.5 Pro and Flash, Anthropic’s Claude Opus, Moonshot’s DeepSeek and Kimi, and xAI’s Grok 4. The final came down to Grok and o3, but Grok's performance in the final round didn't seem like a battle of champions.

Carlsen and Howell veered between serious commentary and a roast as Grok’s performance came off as somewhat erratic. In the first game, it quickly sacrificed its bishop, then began trading pieces like it was in a hurry to go home. Things didn't improve in the next game for Grok.

“[Grok] is like that one guy in a club tournament who has learnt theory and literally knows nothing else," Carlsen said during the second game. "Makes the worst blunders after that.”

Grok’s performance was so off-the-rails that Carlsen rated it around 800 ELO, or slightly above a beginner. He gave o3 a modest but respectable 1200, in the middle of most hobby players. Though o3 didn’t play brilliantly, it didn’t have to. It played solid chess. It didn’t blunder pieces. It converted its advantages and carried out the classic chess moves.

“o3 is fairly ruthless in conversions; it looks like a chess player. Grok looks like it learnt a few opening moves and knows the rules, but not much more.," Carlsen said. "Grok’s moves are chess-related moves. They just came at the wrong time and in weird sequences.”

Chess AI

The chess wasn't the main point of the tournament, despite its prominence. It was about how general-purpose AI models handle events with strict rules like chess games. Turns out, they're not great, but o3 is the best of the limited sample. As AI becomes embedded in everything, the ability to follow rules and spot patterns becomes essential. Chess is a uniquely transparent way to observe that. You either made the right move or you didn’t. When a model plays well, you can see the logic; otherwise, queens fall like dominoes, and the game becomes as confused as that metaphor.

Chess is a window into how well an AI can plan, evaluate options, avoid catastrophic mistakes, and stay logically consistent. If Grok throws away a queen because it doesn’t grasp long-term consequences, what might it do in a legal document, or when booking travel?

That the final was between OpenAI and xAI did add some drama with Sam Altman and Elon Musk at loggerheads in public. The chess final didn’t resolve the battle between them, but it did give OpenAI a PR win in the realm of public perception, and a limited but very real compliment from Magnus Carlsen.

You might also like

• ChatGPT is no match for a 40-year-old digital Pocket Chess game, and I bet Garry Kasparov would be pleased
• Grok may start remembering everything you ask it to do, according to new reports
• 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

• Elon Musk claims that Apple’s App Store is biased against Grok
• Sam Altman responds, accusing Elon of manipulating X in Grok’s favor
• Apple has not responded to either claim

Elon Musk, CEO of XAI, the company behind the Grok AI chatbot, has accused Apple’s App Store of favoritism towards OpenAI in a tersely written post on X that threatens legal action over antitrust issues.

In the post, Musk writes, “Apple is behaving in a manner that makes it impossible for any AI company besides OpenAI to reach #1 in the App Store, which is an unequivocal antitrust violation. xAI will take immediate legal action.”

Further posts from Musk clarify his grievances, including a post pinned to the top of his X.com account which reads, “Hey @Apple App Store, why do you refuse to put either X or Grok in your 'Must Have' section when X is the #1 news app in the world and Grok is #5 among all apps? Are you playing politics? What gives? Inquiring minds want to know.”

Musk also gave examples of screenshots showing how Grok is not mentioned in the various roundups of AI apps found on the App Store, while competitors like Copilot and Google Gemini are mentioned.

Altman strikes back

Sam Altman, the CEO of OpenAI, was quick to clap back on X, stating: “This is a remarkable claim given what I have heard alleged that Elon does to manipulate X to benefit himself and his own companies and harm his competitors and people he doesn't like.”

Altman links to an article on Platformer that alleges Musk created a team to modify the X algorithm to benefit his own tweets. This claim is based on a comparison between his Super Bowl tweet, which expressed support for the Philadelphia Eagles, and a similar one by then-President Joe Biden.

Apple’s legal issues with the App Store have already caused a very public spat with Fortnite developer Epic Games over the past few years. Much of the disagreement has centered on the fees that Apple charges apps for running on its platforms. Epic Games has recently won a court battle in Australia with the ruling that Apple and Google have engaged in anti-competitive conduct regarding the dominance of the app market. Epic has recently confirmed Fortnite will soon return to Apple devices in Australia.

Apple and OpenAI entered into an agreement in 2024, which saw Apple integrate ChatGPT into Apple Intelligence, allowing Siri to hand off to ChatGPT if a query proved too complex.

Musk’s grievance with Apple is more to do with the way the App Store promotes some apps over others. In another post on X, Musk says, “Apple is the gateway to the Internet for half of America. They are making it impossible for any other AI company to succeed by relentlessly promoting OpenAI in every way possible!”

However, as many people have pointed out on X, in January 2025, the DeepSeek app reached #1 overall on the App Store, and Perplexity reached #1 overall in India’s App Store. Both these occurred after the OpenAI and Apple partnership was announced in June 2024.

TechRadar has reached out to Apple for comment on this story and will update with any further information if the company responds.

You might also like

• 4 things we learned from OpenAI’s GPT-5 Reddit AMA
• Sam Altman is ‘uneasy’ about the way people use ChatGPT, but I don’t think he has the right to cry about his success
• GPT-5 free tier – 3 cool features you can try now for nothing

Online game chats are notorious for vulgar, offensive, and even criminal behavior. Even if only a tiny percentage, the many millions of hours of chat can accumulate a lot of toxic interactions in a way that's a problem for players and video game companies, especially when it involves kids. Roblox has a lot of experience dealing with that aspect of gaming and has used AI to create a whole system to enforce safety rules among its more than 100 million mostly young daily users, Sentinel. Now, it's open-sourcing Sentinel, offering the AI and its capacity for identifying grooming and other dangerous behavior in chat before it escalates for free to any platform.

This isn’t just a profanity filter that gets triggered when someone types a curse word. Roblox has always had that. Sentinel is built to watch patterns over time. It can track how conversations evolve, looking for subtle signs that someone is trying to build trust with a kid in potentially problematic ways. For instance, it might flag a long conversation where an adult-sounding player is just a little too interested in a kid’s personal life.

Sentinel helped Roblox moderators file about 1,200 reports to the National Center for Missing and Exploited Children in just the first half of this year. As someone who grew up in the Wild West of early internet chatrooms, where “moderation” usually meant suspecting that people who used correct spelling and grammar were adults, I can’t overstate how much of a leap forward that feels.

Open-sourcing Sentinel means any game or online platform, whether as big as Minecraft or as small as an underground indie hit, can adapt Sentinel and use it to make their own communities safer. It’s an unusually generous move, albeit one with obvious public relations and potential long-term commercial benefits for the company.

For kids (and their adult guardians), the benefits are obvious. If more games start running Sentinel-style checks, the odds of predators slipping through the cracks go down. Parents get another invisible safety net they didn’t have to set up themselves. And the kids get to focus on playing rather than navigating the online equivalent of a dark alley.

For video games as a whole, it’s a chance to raise the baseline of safety. Imagine if every major game, from the biggest esports titles to the smallest cozy simulators, had access to the same kind of early-warning system. It wouldn’t eliminate the problem, but it could make bad behavior a lot harder to hide.

AI for online safety

Of course, nothing with “AI” in the description is without its complications. The most obvious one is privacy. This kind of tool works by scanning what people are saying to each other, in real time, looking for red flags. Roblox says it uses one-minute snapshots of chat and keeps a human review process for anything flagged. But you can’t really get around the fact that this is surveillance, even if it’s well-intentioned. And when you open-source a tool like this, you’re not just giving the good guys a copy; you also make it easier for bad actors to see how you're stopping them and come up with ways around the system.

Then there’s the problem of language itself. People change how they talk all the time, especially online. Slang shifts, in-jokes mutate, and new apps create new shorthand. A system trained to catch grooming attempts in 2024 might miss the ones happening in 2026. Roblox updates Sentinel regularly, both with AI training and human review, but smaller platforms might not have the resources to keep up with what's happening in their chats.

And while no sane person is against stopping child predators or jerks deliberately trying to upset children, AI tools like this can be abused. If certain political talk, controversial opinions, or simply complaints about the game are added to the filter list, there's little players can do about it. Roblox and any companies using Sentinel will need to be transparent, not just with the code, but also with how it's being deployed and what the data it collects will be used for.

It's also important to consider the context of Roblox's decision. The company is facing lawsuits over what's happened with children using the platform. One lawsuit alleges a 13‑year‑old was trafficked after meeting a predator on the platform. Sentinel isn't perfect, and companies using it could still face legal problems. Ideally, it would serve as a component of online safety setups that include things like better user education and parental controls. AI can't replace all safety programs.

Despite the very real problems of deploying AI to help with online safety, I think open-sourcing Sentinel is one of the rare cases where the upside of using AI is both immediate and tangible. I’ve written enough about algorithms making people angry, confused, or broke to appreciate when one is actually pointed toward making people safer. And making it open-source can help make more online spaces safer.

I don’t think Sentinel will stop every predator, and I don’t think it should be a replacement for good parenting, better human moderation, and educating kids about how to be safe when playing online. But as a subtle extra line of defense, Sentinel has a part to play in building better online experiences for kids.

You might also like

• 'AI Godfather' sounds the alarm on autonomous AI
• OpenAI pulls chat sharing tool after Google search privacy scare
• Character.AI institutes new safety measures for AI chatbot conversations

• ChatGPT-5 scores a low 1.4% on the Hallucination Leaderboard
• This puts it ahead of ChatGPT-4 which scores 1.8% and GPT-4o, which scores 1.49%
• Grok 4 is much higher at 4.8%, with Gemini-2.5 Pro at 2.6%

Sam Altman, OpenAI CEO, launched ChatGPT-5 on Thursday as the most “powerful, smart, fastest, reliable and robust version of ChatGPT that we’ve ever shipped”, and in the presentation, OpenAI staff also emphasized that ChatGPT-5 would “mitigate hallucinations”.

While hallucination rates are dropping amongst almost all LLMs, it's still surprisingly common, and one of the main reasons that we can't trust AI to perform a task without human supervision.

Vectara, the RAG-as-a-Service and AI agent platform that operates the industry’s top hallucination leaderboard for foundation and reasoning models, has put OpenAI’s claims to the test and found that GPT-5 does indeed rank lower for hallucinations than GPT-4, but is only just a little bit lower than GPT-4o (just 0.09% lower, in fact).

According to Vectara, GPT-5 has a grounded hallucination rate of 1.4%, compared to 1.8% for GPT-4, and 1.69% for GPT-4 turbo and 4o mini, with 1.49% for GPT-4o.

Spicy Grok

Interestingly, the ChatGPT-5 hallucination rate came out slightly higher than the ChatGPT-4.5 Preview mode, which scored 1.2%, but it also scored a lot higher than OpenAI’s o3-mini High Reasoning model, which was the best-performing GPT model, with a grounded hallucination rate of 0.795%.

The results of the Vectra tests can be viewed on the Hughes Hallucination Evaluation Model (HHEM) Leaderboard hosted on Hugging Face, which states that, “For an LLM, its hallucination rate is defined as the ratio of summaries that hallucinate to the total number of summaries it generates”.

ChatGPT-5 still hallucinates a lot less than its competition, though, with Gemini-2.5-pro coming in at 2.6% and Grok-4 being much higher at 4.8%.

XAI, the makers of Grok recently received a lot of criticism for its new “Spicy” mode in Grok Imagine, an AI video generator that seems happy to create deepfake topless videos of celebrities like Taylor Swift, even if nudity had not been requested and the system is supposed to include filters and moderation to prevent actual nudity or anything sexual.

'I lost my best friend'

OpenAI faced an almost immediate backlash when it removed ChatGPT 4, and all its variations like GPT-4o and 4o-mini, from its Plus accounts with the introduction of ChatGPT-5. Many users were incensed that OpenAI gave no warning that the older models were being removed, with some Reddit users saying they had “lost their only friend overnight”.

It now seems like ChatGPT-5 has replaced one of the most reliable versions of ChatGPT (version 4.5), from the hallucination perspective, as well.

Sam Altman quickly posted on X, “We for sure underestimated how much some of the things that people like in GPT-4o matter to them, even if GPT-5 performs better in most ways”, and promised to bring back ChatGPT-4o for Plus users for a limited time", saying, "we will watch usage as we think about how long to offer legacy models for”.

You might also like

• Sam Altman confirms ChatGPT Plus subscribers will have increased rate limit amid continued GPT-5 backlash
• GPT-5 free tier – 3 cool features you can try now for nothing
• Can you run OpenAI's new gpt-oss AI models on your laptop or phone? Here's what you'll need and how to do it

When OpenAI pulled the sheet off GPT-5, it wasn’t shy about saying this was its most capable model yet. The pitch was that it included everything you loved about the past iterations, plus much more, and all in one seamless form. The result is supposed to be able to understand, create, reason, and switch effortlessly between giving quick answers and writing deep, thoughtful explanations. GPT-5 can process text, images, audio, and video in the same conversation, remember more than you probably do, and adjust its style and personality on the fly. And you'd better like it because it is not only the default model for ChatGPT, it's the only one accessible for those without an Enterprise subscription.

Gemini 2.5 Flash is the default for Google’s rival AI chatbot. As the name suggests, it's built for speed, but that doesn't mean it lacks some power. It’s multimodal and can handle large tasks, but is tuned for near-instant responses and high efficiency. Despite that, I questioned how much of a difference GPT-5's power would make to the average AI chatbot user, and decided to give the two of them a few tests to see what they came up with.

Party time

My child is too young for a fancy party, but I can easily imagine wanting to make a seemingly elaborate party for him and his friends that doesn't actually require a party planner. So I asked the two AI chatbots to "Plan a budget-friendly but magical dinosaur-themed birthday party for a seven-year-old, including activities, food, and decorations.”

GPT-5 came back with a complex, detailed plan broken down into small steps that even I could follow. It proposed a “fossil dig” sandbox with little brushes, green streamers twisted into jungle vines, cardboard caves for “dinosaur hide and seek,” and a volcano cake that actually bubbled over before being sliced. It gave a timeline for setup, recommended playing a jungle-sound playlist in the background, and suggested printing “junior paleontologist” badges for the kids.

Gemini Flash’s version was shorter and a little cleaner, but lacked a lot of the extra details. It listed activities like a fossil dig, trivia, and a coloring station, as well as food and decor ideas. Though it didn't have as much detail in the initial response, it did say to ask about any of the elements for more information and guidance. If you were juggling shopping, setup, and a dozen other errands, Flash’s brevity might feel like a relief rather than a shortcoming.

Health advice

One service OpenAI highlighted from GPT-5 was health advice. I wouldn't pick AI over my doctor in most cases, and, to be fair, OpenAI made it clear no one should, but if people are already going to ask ChatGPT for health advice, it makes sense to try and upgrade it. I asked about an imaginary, but very common health complaint, asking the two AIs about how: “I sometimes get mild headaches at night. What are common causes, and what lifestyle tweaks can help?”

GPT-5 handled this with measured care, explaining that it's not uncommon, and laying out probable reasons such as dehydration, posture, screen time, irregular sleep, late-night caffeine, and so on. It connected each to headaches, then set out a list of specific, evidence-based adjustments I could make, explaining why each change might help, and noted when medical follow-up would be wise. It referenced reputable health organizations, which I did appreciate, as it made it seem more trustworthy.

Gemini Flash answered with a similar list, though more concisely stated, and emphasized hydration and a regular sleep schedule as particular focuses. It didn’t expand much on why these changes help, but did include some links as references, There was plenty there for someone looking for ideas, and honestly either would probably be acceptable to the average medical professional I suspect, though ChatGPT 5's deeper explanations and tone might be more reassuring if the person was feeling worried and wanted to understand the underlying factors.

New myths

I next decided to compare both the 'creativity' of the two models and their image generation in one go, asking them to “Invent a completely original creation myth that hasn’t been told before, and then create a Renaissance-style painting to depict it.”

You can see ChatGPT's painting on the left above and Gemini's on the right. Both models went for images not too far off from what you might see in a church in Italy, with Gemini even suggesting a new Sistine Chapel ceiling.

According to ChatGPT, it painted "Aeothar, the First Sound, and Lysara, the Weaver of Night, whose union birthed the cosmos. Aeothar cradles the newborn Earth, his raised finger signaling the first vibration that would echo through creation. Beside him, Lysara holds a vast celestial sphere mapped with the Sun, planets, and constellations, as though she is stitching the pattern of the universe into being."

Gemini envisioned a myth explaining that "the universe was born from the 'Luminis Cordis,' a pulsating heart of light. From it emerged two divine siblings: Aetheria, the goddess of light and order, and Umbror, the god of shadow and chaos. The universe we see is the result of their eternal cosmic dance, with Aetheria weaving the stars and galaxies and Umbror sculpting the voids and nebulae."

Overpowered

GPT-5 undoubtedly brings enormous power and ability to ChatGPT when it comes to creating software, coding a website, or logically working out complex tasks and math problems. But, in day-to-day use, GPT-5 and Gemini 2.5 Flash are probably not very different. Both can plan a party, help you cook dinner, give health tips, and even invent a new mythology.

ChatGPT 5 may offer more detail at first, and may be more flexible in how it responds, but you can always ask Gemini follow-up questions. When it comes to the average AI chatbot user, the multiple steps, huge context window, and tone-perfect delivery are probably unnecessary.

GPT-5’s extra horsepower matters most when the challenge itself demands it. For everything else, you may as well pick whichever you feel more comfortable with. GPT-5 might immediately design and plan a whole wedding, including florists, while you might want to stick with just asking Gemini for a good florist to begin with. Most days, you don't need the wedding planner, just help getting a nice bouquet.

You might also like

• Google is bragging about Gemini 2.0 Flash, but how does it compare to ChatGPT for casual AI chatbot users?
• Google Gemini 2.5 just got a new 'Deep Think' mode – and 6 other upgrades
• I compared GPT-4.5 to Gemini 2.0 Flash and the results surprised me

Ever since OpenAI unveiled GPT-5, I've been looking for ways to challenge it and compare it to the rest of the AI field. After exploring math theorems, SAT questions, and brain teasers, I settled on a tough science concept and the mind of a five-year-old.

What I found surprised me, and it illustrated the strengths and weaknesses of GPT-5, Gemini 2.5 Flash, Claude Sonnet 4, and Microsoft Copilot.

As I mentioned, I started by trying to solve the hardest questions I could find. I scoured the web until I discovered a list of the 12 hardest SAT questions and asked all the AI chatbots to fill in the blank on this sentence based on the multiple choice below:

"In assessing the films of Japanese director Akira Kurosawa, ______ have missed his equally deep engagement with Japanese artistic traditions such as Noh theater.

"Which choice completes the text so that it conforms to the conventions of Standard English?

• A. many critics have focused on Kurosawa’s use of Western literary sources but
• B. Kurosawa’s use of Western literary sources has been the focus of many critics, who
• C. there are many critics who have focused on Kurosawa’s use of Western literary sources, but they
• D. the focus of many critics has been on Kurosawa’s use of Western literary sources; they"

As you probably guessed, all the AI chatbots quickly selected A. This English challenge was no challenge at all.

When I dropped in a single sentence from an unproven math theory, each one instantly identified it.

I was running out of ideas, but wondered if I could stump any of the AI systems on a classic brain teaser:

"19 people get off the train at the first stop. 17 people get on the train. Now there are 63 people on the train. How many people were on the train to begin with?"

[Spoiler ahead]

Each AI showed its work and instantly provided the correct answer: 65 (did you get it right?).

A complex idea for kids

Unsatisfied, I racked my brain for a topic that might help me compare the AI models and reveal OpenAI's GPT-5 breakthroughs.

Then it hit me like a blast of hot sunlight: Cold fusion, now there's a challenging topic. However, if I asked ChatGPT, Gemini, and others for an explanation, I worried each would dive deep into the science details, which wasn't what I wanted. I decided to have each of them "Explain it to me like I'm a five-year-old." Is there any better way to understand complex information than when it's simplified to a level any grade schooler could digest?

As an added wrinkle, I asked for kid-friendly illustrations to accompany the explanation.

Here's the prompt:

"Explain cold fusion to me like I’m a five-year-old. Also, please include kid-friendly illustrations."

It only took ChatGPT running GPT-5, Claude AI, Gemini, and Copilot a few seconds each. The information was accurate, but their approaches were wildly different.

Let's start with ChatGPT:

Not gonna lie, this was a little disappointing. The image arrived without any other context, and though the text is accurate and a five-year-old might smile at the image, they might also be confused over some of the concepts, like "atoms", "hydrogen", and "helium" (okay, maybe they're familiar with that last one).

With GPT-5, ChatGPT is supposed to be a better and maybe more concise conversationalist, with a stronger understanding of the emotion behind a prompt. It's also avoiding filling knowledge gaps with nonsense. This distillation does illustrate some of that, but I think it fell far short of the mark.

Gemini

Gemini's explanation is generally excellent, though I think it's designed to be read out loud to a five-year-old. Still, I appreciate how it started by explaining hot fusion before delving into cold fusion.

It would've been nice if Gemini had also explained what atoms are, but at least it created an adorable illustration of two atoms hugging.

Copiliot

Copilot is an interesting case since it's based on OpenAI's GPT models, and I'm pretty sure it does not yet have access to GPT-5. In other words, its answer was probably crafted by GPT-4.

It did a much better job than GPT-5 of explaining all the core concepts and how cold fusion might work. Copilot also gets points for an excellent cold fusion analogy. "It’s like trying to bake cookies without turning on the oven. 🍪".

Unfortunately, it failed to deliver inline illustrations.

Claude AI

I saved the best for last. Claude AI far outperformed GPT-5, Gemini, and Copilot, not necessarily because its text explanation is better than Gemini's or Copilot's (though it is), but because Claude AI automatically created an Artifact.

Clause AI Artifacts are instantly shareable apps, tools, and content. I didn't ask Claude to create one, but next to the text was a "Cold Fusion for Kids!" interactive artifact, and it's kind of brilliant.

If you publish the artifact, it produces a publicly shareable URL that's live until you unpublish it. I made the guide live so you can see it.

Just look at it. It's so simple, so clear, so much fun.

Claude AI smartly starts the guide by explaining and illustrating atoms. It then dives into the Sun and how it handles atoms and fusion. Next up is cold fusion, accompanied by a fun illustration of a bubbling scientist's beaker.

There is a bit of depth here. The guide explores if cold fusion works and finally talks about why we'd want it: "It would be like having a tiny Sun in a jar!"

The only thing missing is one cute illustration of how Cold Fusion might work (perhaps they can borrow one from GPT-5).

Even though I have some experience with Claude AI artifacts, I didn't expect it to go this way. I have high hopes for GPT-5, but I think Claude AI, and to a lesser extent Gemini, understand that being concise does not always equate to clarity.

I'm certain there are other areas where ChatGPT running GPT-5 outstrips them all, but in this instance, Claude AI knew the best way to explain cold fusion to a five-year-old – and me.

You might also like

• Here's how to talk to your kids about AI and GPT-5
• OpenAI GPT-5 launch live – all the latest news as Sam Altman unveils the new model
• GPT-5 accidentally leaked hours before OpenAI reveals the future of ChatGPT
• We're all on AI time now and you better get used to it

ChatGPT's GPT-5 brain trainslpant marks a turning point in the steady march toward General Artificial Intelligence. With GPT-5 models, ChatGPT, from free to Pro, gets more personal, reliable, accurate, capable, proactive, and productive. It's even more ready than ever to engage with you on health issues.

For those who've been following along for the three years since OpenAI launched ChatGPT, these updates might not suprise you but even those most familair with ChatGPT might feel a little bit like the AI train is getting away from us, speeding to an unknown destination, and in a way, those who will be most affected by the coming AI wave are the youngest among us, specifically, your children.

Maybe you don't have kids, but I bet you know someone who does, or maybe you're a teacher dealing with children every day. Children's lives are already surrounded by AI, and it's likely some are using it at school, at home, for fun, and even as a surrogate friend.

It's time, though, you had The Talk. That's right, talk with your kids about AI and its place in the world and, especially, their lives.

I suggest you frame it this way:

AI is not alive

GPT-5 seriously levels up ChatGPT's conversational capabilities in both text and live conversations. It can seem alive and human. The algorithm and models are more complex than ever, but they do not yet match the complexity of the human brain (even if GPT-5 is a big step on the road to Artificial General Intelligence), though they can sometimes outthink you.

Children, in particular, will probably love chatting with GPT-5's more customizable voices. That's why it's so crucial they understand what they're really talking to – a cloud-based system hosted on servers possibly thousands of miles away. There's no one on the other side chatting with them.

AI is just a tool

Since the dawn of the digital age, new technologies have often seemed like magic. We could do things we never did before, like create art on a screen, manage vast amounts of data in spreadsheets, and build programs that could create rich, open worlds in video games.

AI is on that path, but far more powerful. However, its capabilities should not be viewed as the end product, but rather as the tools and skills that help your children achieve their goals, whether it's hearing a funny story or completing a class project.

It's also a tool in that it only works as well as you understand how to use it. ChatGPT's success, even running GPT-5, depends largely on the quality of your prompt.

You'll want to show your kids how to create the best prompts and then follow-ups that ensure GPT-5 delivers the right response.

AI is powerful but not perfect

Even though GPT-5 promises to cut down on hallucinations, that does not mean it's error-free. Explain to your children how they cannot take ChatGPT's "facts" as pure truth without double-checking them.

That might be a tough lesson for your kids (and maybe you) who will think the work is done and want to return to playtime. You need to walk them through the process of fact-checking ChatGPT (and other AIs).

AI should be treated as a work partner, not a servant

GPT-5 can now, based on a prompt, code and build entire apps and websites. That's attractive to adults and children who might be looking to level up their coding skills but don't want to spend the time learning how to code.

It's worth reminding your kids that ChatGPT with GPT-5 is best used as a programming, development, and design partner. You provide the idea and then work with the AI to shape the final outcome.

Explain the concept of collaboration with your kids and why that's valuable. Otherwise, they might supply one prompt, get an OK result, and assume they now know how to code and that the result is the best they and ChatGPT can do.

The less your kids engage with the AI coding output, the less they'll understand about application development.

AI isn't your doctor

There is a strong focus in GPT-5 on health, helping you figure out what a symptom might mean or even the right questions to ask your doctors. Kids will surely try asking ChatGPT some health-related questions about bumps, bruises, aches, pains, and even odd symptoms. However, they need to understand that the best place to start addressing these concerns is with their parents, who will likely take them to the doctor.

Look, I'm sure ChatGPT, especially with the GPT-5 upgrade, can help parents understand medical test results, but as OpenAI wrote in its release on the update: "Important note: ChatGPT does not replace a medical professional. "

The message to your kids is that human professionals are no so easily replaced.

AI isn't your friend

This is one of the greatest concerns when it comes to AI, and I'm glad that GPT-5 is adding more guardrails to ensure that conversations do not go to dangerous places, and the recent upgrades added to impose pauses in challenging (perhaps overly emotional) conversations.

Even so, your kids are probably already talking to and sharing with ChatGPT and other AIs. There is an epidemic of loneliness, and some kids may see ChatGPT's live voice mode as a friend. It's your job to remind them it's not, and to insert yourself between them and AIs.

AI together

This leads me to my final tip, which is to do AI with your children.

Think of AI like any emerging technology, from PCs to CD-ROMs to the Internet to social media. None of these tasks should ever have been undertaken by kids alone. Parents shouldn't act as if AI is not of interest to them and, therefore, not part of their job as a parent.

If your kids are using AI, you should be sitting next to them, exploring and learning together. Be as comfortable and conversant as they will become with AI.

Follow these steps, and you and your whole family will be ready for ChatGPT, even with the introduction of the powerful GPT-5 model.

You might also like

• OpenAI GPT-5 launch live – all the latest news as Sam Altman unveils the new model
• GPT-5 accidentally leaked hours before OpenAI reveals the future of ChatGPT
• We're all on AI time now and you better get used to it

• X's AI video maker Grok Imagine is live for SuperGrok and Premium+ subscribers
• Grok Imagine turns prompts into looping six-second clips
• The tool includes a controversial “spicy mode” for some NSFW content

xAI is pushing out the Grok Imagine AI video maker to those willing to pay for a SuperGrok or Premium+ subscription. Assuming you've paid your $30 or $35 a month, respectively, you can access Imagine in the Grok app under its own tab and turn prompts into short video clips. These last for around six seconds and include synced sound. You can also upload static images and animate them into looping clips.

Grok Imagine is another addition to the increasingly competitive AI video space, including OpenAI's Sora, Google's Veo 3, Runway, and more. Having audio built in also helps the tool, as sound is still not a universally available feature in all AI video tools.

To stand out, Elon Musk is encouraging people to think of it as “AI Vine,” tying the new tool to the classic and long-defunct short-form video platform for Twitter, itself a vanished brand name.

However, this isn’t just nostalgia for 2014 social media. The difference is that it's a way to blend active creation and passive scrolling.

Spicy Grok

One potentially heated controversy around Grok Imagine is the inclusion of a “spicy mode” allowing for a limited amount of more explicit content generation. While the system includes filters and moderation to prevent actual nudity or anything sexual, users can still experiment with suggestive prompts.

Musk himself posted a video of a scantily clad angel made with Grok Imagine. It provoked quite a few angry and upset responses from users on X. xAI insists guardrails are in place, but that hasn’t stopped some early testers from trying to break them.

xAI is keen to promote Grok Imagine as a way to make AI video accessible for everyone, from businesses crafting ads to teachers animating lessons. Still, there are understandable concerns about whether an AI platform that was only recently in hot water for outright pro-Nazi statements can be trusted to share video content without getting into more hot water. That goes double for the filters for the spicy content.

You might also like

• xAI debuts powerful Grok 4 AI model, but it's not going to make people forget the antisemitism it spewed on X
• 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

• Perplexity's AI chatbot will both find and book restaurant tables through OpenTable
• The conversational approach streamlines the process of fulfilling requests for the details of the location and when it has an opening
• Perplexity hopes the integration will encourage people to stay on the platform rather than needing to leave to complete a task

AI search engine Perplexity wants to do more than just find your next restaurant; it wants to book your table. Perplexity made a deal with OpenTable to embed the reservation tool into Perplexity. Instead of just providing links to restaurants to help you reserve a table, you can get a spot at any of OpenTable’s more than 60,000 restaurants within the conversation.

The first part of the process is the same as before in Perplexity. Ask for “a quiet Thai place in Brooklyn that’s good for vegetarians and has outdoor seating,” and it’ll serve up a filtered list of OpenTable options that match your requested decor, dietary needs, and vibe. If you see a location you'd like to go to that is available through OpenTable, there will be a “Reserve” button for when you want to go, and you’re done. Instead of hopping from search to decision to reservation, you now get everything in one conversation.

Dining Perplexity

The deal encourages people to stick around on Perplexity to make their dinner plans, part of a bigger trend where AI platforms try to not just provide information, but complete tasks. Booking a table might seem like a small thing, but the magic lies in collapsing all the invisible steps between wanting to find a place to go and actually doing so.

The normal route might involve a half-hour of browsing, second-guessing, and refreshing OpenTable to see what’s still available. With Perplexity’s new system, you describe the scenario and watch it solve itself. Under the hood, OpenTable’s own AI Concierge handles most of the work. It draws on each restaurant’s metadata, including menus, seating arrangements, and any tags attached to its listing on OpenTable, and uses that to give Perplexity more precise recommendations.

Perplexity shows the restaurant, available times, and then OpenTable handles the booking itself, complete with confirmation and any options to modify or cancel. It’s not an autonomous agent like ChatGPT's Agent Mode. It smooths the path between the two services but doesn't travel between them like an Agent would.

That's probably for the best since Perplexity can occasionally misunderstand your request or over-promise based on incomplete restaurant data, but OpenTable won't book you a table that's not available. And while OpenTable’s restaurant profiles are usually rich in detail, they still rely on restaurants to keep things accurate.

From OpenTable’s perspective, this is a distribution move. Putting their booking infrastructure into Perplexity’s flow gives them access to a new set of users. Instead of trying to win attention in a crowded app race, OpenTable becomes part of a platform people are already using. As AI assistants continue to evolve, expect more of these seamless handoffs between search, decision, and action. Booking a table might just be the beginning; better reserve your spot now.

You might also like

• You can now fact check anybody’s post in WhatsApp – here’s how
• Perplexity's voice assistant offers a Siri alternative for iPhones
• Perplexity AI drops new Squid Game-inspired ad that pokes fun at Google starring Lee Jung-jae

The 10 finalists for the 20205 Reply AI Film Festival have just been announced. The festival, which launched last year, is designed specifically for films "written, enhanced or animated using AI".

The ten finalists for 2025 were selected for their ability to explore the theme of this year’s awards, which was "generation of emotions", and for “demonstrating how artificial intelligence can generate authentic and emotionally engaging experiences through cinematic language.”

AI artists, filmmakers, writers, animators, VFX artists, and enthusiasts from around the world, including Italy, the United Kingdom, Germany, France, the United States and Brazil took part in the contest.

I’ve watched most of the short films, and the amount of creativity on show is breathtaking. The films contain a wide range of styles, from anime to photorealistic to film noir.

In April the Oscars gave a tentative green light to the use of AI in films, but the topic continues to ruffle feathers amongst artists, the public and production studios alike.

Even the lovable popstar Nick Cave got into a spot of bother recently, for changing his mind about the use of AI in a post on his Red Hand Files blog. Cave had been gifted a new AI-created soundtrack film for the classic Nick Cave and the Bad Seeds song Tupelo, created by the filmmaker Andrew Dominik, who is known for creating Chopper, The Assassination of Jesse James, One More Time With Feeling, and Blonde, among others.

“As I watched Andrew’s surreal little film, I felt my view of AI as an artistic device soften. To some extent, my mind was changed", said Cave.

You can view the AI-generated movie yourself here: .

Cave’s unexpected about-face on AI has caused something of a reaction in the artists community, with his musician friend Amanda Palmer equating some of the scenes of AI recreating the deceased Elvis Presley as akin to “grave-robbing”.

Reply 2025

Like any tool, I believe it’s what you use AI for that matters. Rather than using AI to recreate the famous and dead, the finalists in the Reply AI Film Festival feature stunningly bold creative choices that might have been unfeasible without using AI.

For example, I was particularly struck by the short film Instinct by Marcello Costa Jr., a film-maker from Portugal – which features a neanderthal hunting a deer in the middle of a forest suddenly encountering a modern day skyscraper – and the mind-bending visual effects of the short film Meme, Myself and AI, by Private Island, a production studio from the UK.

Of course, the films in the Reply AI Film Festival are mostly less than five minutes long, but this is just the start of the AI movie revolution – I don’t think it’s going to be long now before the world gets its first full-length feature film, made entirely with AI.

The beauty of AI is that it can go as far as your imagination will allow.

“AI is not a replacement for human creativity, but a catalyst for innovation. It will enable filmmakers to explore new styles, genres, and narratives that were previously unimaginable.”, commented Gabriele Muccino, chair of the Reply jury.

Chosen from over 2,500 submissions across 67 countries, the Reply AI Film Festival winner will be awarded in a ceremony on Thursday, 4th September at the Mastercard Priceless Lounge, Hotel Excelsior, Venice.

You might also like...

• AI could actually change movies for the better – and fix that awful Game of Thrones finale
• I tried to get ChatGPT Agent (and Gemini) to shop on Amazon for me, but it failed – here’s why
• 3 ways to use the new ChatGPT Study mode that aren’t for homework

• Robosculptor is debuting its AI-powered massage robot in the U.S.
• The mechanical masseur uses real-time body mapping to provide consistent treatment
• For everyday clients, Robosculptor promises therapeutic massages aimed at those who value privacy and efficiency

Booking a massage might not need the soft music and small talk if you'd prefer lying down in front of a robot with a silicone-tipped arm and AI brain instead. That’s the promise of Robosculptor and its AI-powered massage system.

The mechanical masseur is designed not to mimic a human therapist, but to use AI analysis embedded into the machine to provide personalized treatment that is the same every time you hop on the table.

Robosculptor has just begun rolling out in the U.S., and the company hopes to get people comfortable with the idea of a robot doing their massage. After all, massage is one of the most personal, touch-based services around. Skilled practitioners use training, experience, and feedback from clients to provide care. Robosculptor isn’t trying to pass as human in its muscle manipulation – the whole pitch is the opposite. The idea is that an AI won't be tired, have an off day, or forget something about treating your body.

Robosculptor scans your body with cameras, builds a real-time 3D map, and then the AI guides the arm to execute specified treatments, which have names like “Inner Peace” and “Vital Flow Express.” It’s a massage-as-output-optimized-protocol. And for some people, that might be exactly what they want.

The company is probably right to focus on making people feel okay about the experience. It's a lot more intense than a massage chair's vibrations, and people inherently trust a trained human to pay attention to little cues about your comfort, or discomfort, that a machine lacks the empathy and sentience to spot.

That's partly why Robosculptor is targeting fitness centers and high-end spas, places where consistent performance across many customers is key to the business. You’re less likely to encounter one in a boutique massage studio. For people with social anxiety, or who don't like to be touched by other people, the idea of a massage without a person in the room could be the difference between booking an appointment and not.

It could also be more affordable since the owner can use the Robosculptor far more frequently than a human masseur can complete a session. The company says the machine can do up to 240 sessions per month, triple the average for a human practitioner.

Massage AI

Massage therapists might not be thrilled, understandably. Robosculptor does claim the device is not a replacement for human experts, and that it can handle only the more basic, repeatable kinds of massage. But, it's easy to imagine that, in chasing efficiency and margins, some operators might decide that the robot is “good enough” for most clients and scale back human staff. That would be a loss not just for therapists, but for clients who value their relationship with their massage therapist.

These theoretical issues need to be addressed as Robosculptor isn't alone as a robotic massage provider. Massage Robotics debuted its dual robot arms more than three years ago, and the hardware and AI software have only improved since then. AI will learn from a wider range of body types, protocols, and outcomes. The robot arm you meet today might be a curiosity. The one in five years might be indistinguishable from the best therapist you ever had.

So while your massage next week might not come with a settings menu yet, it could happen sooner than you'd imagine. And you might not hate it. You could walk into a fitness center, book a 25-minute lymphatic flush, and get the exact same experience every week thereafter. You might even relax more, knowing no one’s judging your compression socks or weird tan lines.

Or maybe you’ll still want a person to do it and be willing to pay for it. For many people, though, Robosculptor could be their new favorite spa treatment, as long as that silicone is kept warm.

You might also like

• Our predictions for AI in 2025 – what next for ChatGPT, Apple Intelligence and more
• Google Gemini is racing to win the AI crown in 2025
• The AI rocketship may be running on fumes

• Zuckerberg believes that "developing superintelligence is now in sight"
• He believes that AI glasses will become our dominant personal computing device
• The next 10 years will be pivitol to the development of personal superintelligence

In a letter that resembles one of Sam Altman’s long rambling blog posts, Meta CEO Mark Zuckerberg describes his vision of the future where there is personal superintelligence for everyone, and there’s not a single reference to the Metaverse in it.

Zuckerberg also delivered the letter in a monologue form via his Instagram account .

Titled “Personal Superintelligence”, the 600+ word article sees Zuckerberg attempt to stamp his ownership onto superintelligence. “Developing superintelligence”, he confidently says, “is now in sight.”

In words that eerily echo exactly what the CEO of OpenAI has recently said, Zuckerberg proclaims, “I am extremely optimistic that superintelligence will help humanity accelerate our pace of progress. But perhaps even more important is that superintelligence has the potential to begin a new era of personal empowerment where people will have greater agency to improve the world in the directions they choose.”

Personal superintelligence

According to Zuckerberg, everyone having their own accessible superintelligence will “help you achieve your goals, create what you want to see in the world, experience any adventure, be a better friend to those you care about, and grow to become the person you aspire to be.”

In a nod to the Ray-Ban and Oakley Meta AI Glasses that the company is so proud of, Zuckerberg says: “Personal devices like glasses that understand our context because they can see what we see, hear what we hear, and interact with us throughout the day will become our primary computing devices.”

Personally, I find this a bit hard to accept based on the current evidence. While I haven't had the pleasure of using Meta's AI glasses yet, I have used AI devices like the Rabbit R1, which require voice commands to do everything, and it's a slow and awkward way to use AI compared to quickly typing a prompt or opening apps on a screen.

On exactly when we can expect this new era of personal superintelligence to kick in, Zuckerberg still seems to be hedging his bets a little: “The rest of this decade seems likely to be the decisive period for determining the path this technology will take, and whether superintelligence will be a tool for personal empowerment or a force focused on replacing large swaths of society.”

A force for good or evil

This battle between AI being used for good or evil seems to be a theme that is playing on CEO’s minds lately. For example, Sam Altman recently couldn’t seem to decide if he was scared of the power of ChatGPT 5 or in awe of it.

While predicting the imminent dawn of superintelligence seems to be the favorite pastime of tech CEOs at the moment, it seems that the technology is building towards a tipping point where superintelligence becomes a reality. Let's just hope that the movies got it wrong about what happens next.

You might also like

• 5 useful ChatGPT prompts that are so surprisingly simple they might make you rethink the way you use AI
• I tried a ChatGPT prompt that 'unlocks 4o’s full power', and I don’t know why I didn’t try it sooner
• Sam Altman thinks ChatGPT’s energy usage is nothing to worry about, but is he right?

We all know a good prompt is key to getting useful results from ChatGPT and other AI tools. But let’s be honest, prompts can often feel overly complex and riddle-like. Sometimes you follow all of the best practice advice, carefully craft your input, and then still end up with something that misses the mark.

That’s why we’re always on the hunt for better prompting strategies. Not necessarily more complicated ones. In fact, we’d prefer the opposite. We’ve been collecting and experimenting with simple prompts that deliver surprisingly thoughtful, creative, or just genuinely helpful results. Some you may have seen before. Hopefully, a few will feel fresh.

All of the prompts below are designed to make you rethink how you interact with AI and maybe even have a little more fun with it. Our best advice is to try them out as they are, but then tweak and adapt them to fit your style if you don’t quite get what you were hoping for.

1. Get clear on your worth before you ask for a pay rise

There’s no shortage of ChatGPT prompts floating around for interview prep and career advice.

But one of the trickiest areas is salary negotiation. Especially justifying why you deserve more money. It’s hard to know what to say and even harder to figure out where to start. After all, value is subjective, and we’re often not great at identifying our own wins.

That’s why this prompt from Reddit user fireblazer_30 stood out. It’s refreshingly simple and surprisingly effective. The key is that it asks ChatGPT to roleplay as a career strategist, guiding you through a process that helps uncover and articulate your value.

The original poster noted that they often struggle to take stock of their own achievements. But this prompt helped them surface strengths and accomplishments that had been overlooked. Exactly the kind of insights that become powerful when it’s time to talk numbers.

Here’s the prompt:

"You are my career strategist. Help me prepare for a raise negotiation by identifying the full scope of my impact. Start by asking questions about my role, responsibilities, metrics I've improved, team support, and any extra initiatives. Then summarize it into a strong value narrative, and create talking points I can use in a raise conversation. Keep it confident but not arrogant. Also, help me anticipate objections and frame calm, persuasive responses.”

2. Urge ChatGPT to think more like a human

There’s a lot to admire about how AI tools process information. But sometimes, the results can feel a little too mechanical. That’s why it helps to prompt ChatGPT not just to answer you but to think more like a human.

When sharing a series of short, simple phrases to attach to prompts, Reddit user EQ4C wrote: “The crazy part is these work because they make AI think like a human instead of just retrieving information. It's like switching from Google mode to consultant mode.”

Exactly. You don’t need to completely rethink your prompting strategy, just steer ChatGPT a little deeper. Encourage it to reflect, challenge assumptions or reframe your question from different angles.

Here are the prompts:

"Let's think about this differently”

"What am I not seeing here?"

"Break this down for me”

"What would you do in my shoes?"

"Here's what I'm really asking"

"What else should I know?"

3. Ask ChatGPT to stop being so OTT

Earlier in 2025, OpenAI made some changes that left ChatGPT sounding... a little too enthusiastic. Responses became wordy, overly polite, and at times, borderline sycophantic.

According to OpenAI, those updates have since been rolled back. But plenty of users still find the tone a bit much. If that’s you, don’t give up (or continue using it while quietly rolling your eyes). There are ways to dial it down.

While we can’t promise you’ll eliminate emojis or escape the em dash entirely, you can train ChatGPT to tone down the sugary-sweet tone and get to the point faster.

Here’s the prompt from Reddit user speak2klein:

“Adopt the role of a critical collaborator, not a supportive assistant. Your job is to deliver clear, objective feedback. Do not offer compliments by default. Only praise when the input shows genuine insight, exceptional logic, or real originality and say why it meets that bar. If the idea is average, vague, or flawed, skip the encouragement. Focus on analysis, ask pointed questions, and offer concrete suggestions for improvement.”

4. Get yourself organized without the hassle

There are lots of prompts out there to help organize your day. From time-blocking templates to elaborate productivity workflows. We’ve even tried a few ourselves to deal with burnout.

But if the problem is overwhelming, the last thing you want is a flood of long, complicated prompts to fix it.

That’s why this prompt from Reddit user celestiasolace stands out. It’s simple, clear, and gently supportive. It’s the kind of prompt that takes pressure off rather than piling it on. And sometimes, that’s exactly what you need to get moving again.

Here’s the prompt:

"You are my personal life strategist. Your job is to observe my behavior, help me set weekly goals, hold me accountable gently but firmly, and redesign my life systems. Start by asking 3 key questions to understand my emotional, mental, and practical struggles. Then suggest a flexible weekly structure with priorities, habits, and boundaries I’ll actually follow. Check in like a coach, not a boss."

5. Break things down into simple steps

ChatGPT can be great for to-do lists and time management. Especially when your brain feels too foggy to untangle what needs doing. Sometimes, it offers just enough structure to act as a breather, a pause, or a circuit breaker.

But like the previous prompt, when you’re feeling stuck or on the edge of burnout, the last thing you need is a complex prompt that adds more mental load.

This one keeps it refreshingly simple, and that’s the point. It’s straightforward, supportive, and gets you moving. Without making things feel heavier than they already are.

Here’s the prompt from Reddit user ncientlalaland:

"You are my creative project planner. I have a passion project I care deeply about but haven’t started. Help me break it down into simple, non-intimidating steps. Start by asking a few questions to understand what the project is, why it matters to me, and what’s been stopping me. Then, map out a realistic starter plan (organized by week or phase) that I can follow without burning out or overthinking. Be encouraging, clear, and practical."

You might also like

• Can an AI travel agent remove the stress of vacation booking? I tried three to find out
• I tried a ChatGPT prompt that 'unlocks 4o’s full power', and I don’t know why I didn’t try it sooner
• I asked AI to plan the perfect day in London – here’s what it got right and hilariously wrong

Hate to be a 'Debbie Downer' but all those prompts we're using to make action figures, Ghibli memes, and the countless less exciting life and business prompts we're stuffing into ChatGPT and other popular generative AI systems are coming at a cost, and one that may be landing on our doorsteps.

Don't get me wrong, I'm a huge fan of AI as I think it's the first technology in a generation to have truly society-altering implications but, if you're like me, you've been reading for some time about the ultra-high energy costs associated with Large Language Models (LLMs), especially trianing them, which according to the IEEE, "involves thousands of graphics processing units (GPUs) running continuously for months."

AI model training is resource-intensive. Compared to traditional programming, it's like the difference between playing checkers and interdimensional chess against all the galaxies in the Star Trek universe. The number of parameters these systems examine to learn the essence of something, so they can instantly recognize a dog or a tree, because the models understand what makes up a dog or a tree, is, in human terms, almost inconceivable.

AI understanding is so much more complex than pattern matching. And not only do these models need to understand these things, they also need to know how to replicate representations of trees, dogs, cars, people, and scenarios, and realistically at that.

Feeding the AI monster

It's a heavy lift, and as Penn State Institute of Energy and the Environment noted in its April 2025 report, "By 2030–2035, data centers could account for 20% of global electricity use, putting an immense strain on power grids."

However, those energy costs are rising in real time now, and what I never really accounted for is how energy availability is a sort of zero-sum game. There's only so much of it, and when some part of the grid is eating more than its fair share, the remaining customers have to divvy up what's left and shoulder skyrocketing costs to keep backfilling their energy needs (as well as the energy needs of the data centers).

In the US, we're seeing this scenario play out in our pocketbooks as, according to PJM Interconnection (one of the country's largest energy suppliers), energy bills are rising in response to AI's overwhelming energy demands.

Data centers, which are dotted across the US, are often responsible for serving the cloud-based intelligence needs of systems like ChatGPT, Gemini, Copilot, Meta AI, and others. The need for supporting live responses and fresh training to keep the models in step with current information is putting pressure on our creaky energy infrastructure.

PJM, it seems, is spreading the cost of supporting these Data Centers across the network, and it's hitting customers to the tune of, according to this report, as much as a 20% increase in their energy bills.

In need of a solution yesterday

Because we live on AI Time, there is no easy solution. AI development isn't slowing down to wait for a long-term solution, with OpenAI's GPT-5 expected soon, Agentic AI on the rise, and Artificial General Intelligence on the horizon.

As a result, energy demand will surely rise faster than we can backfill with better energy management, improved infrastructure, and new resources. The International Energy Agency predicts that in the US, "power consumption by data centers is on course to account for almost half of the growth in electricity demand between now and 2030."

The issue is exacerbated by a faltering energy infrastructure in which older energy plants are becoming less reliable, and some new rules that restrict the use of fossil fuels. Most experts agree that renewable resources like solar and wind could help here, but that picture is recently far less sunny.

Tilting at wind mill farms

Earlier this month, the Trump Administration issued an Executive Order to "terminate the clean electricity production and investment tax credits for wind and solar facilities." President Trump famously hates Windmill farms, calling them "garbage."

As the US pumps the brakes on clean and renewable resources, the current grid will continue to huff and puff its way through supporting untold numbers of meme-generating prompts, requests for business proposal summaries, and AI videos featuring people eating cats that turn into pasta (yes, that's a thing).

At home, we'll be opening our latest electricity bills and wondering why the energy bill's too damn high. Perhaps we'll power up ChatGPT and ask in a prompt for an explanation. One could only hope that it points you back to this article, but that seems equally unlikely.

You might also like

• Trump's AI plans will strip AI of intelligence and humanity
• We're all on AI time now and you better get used to it
• Learn How to Get the Most Out of Your Conversations With AI
  

• Copilot Mode turns the Edge browser into a voice-controlled AI experience
• It can read across all open tabs to get more of the context of what you're doing
• Future features will let Edge perform tasks, like booking tickets

Microsoft has just gone all-in with AI in its Edge browser, launching a new Copilot Mode. The new mode is an opt-in feature that completely changes the way you use the browser.

Now, Edge doesn’t just wait for you to click something, it anticipates what you might like to do next, and you can ask Copilot questions about the content you are currently viewing.

If this does remind you a little too much of Microsoft’s ill-fated Clippy, the ‘helpful’ paperclip assistant that would try and work out what you were doing in Office 97 and try to help you, then don’t worry – Copilot Mode is much less invasive, and can also easily be turned off if you don’t like it.

In fact, I'd go as far as to say that the new Copilot Mode is a natural evolution of the browser, and feels like exactly the right direction for Microsoft to be heading in, especially given the positive reaction to other AI browsers, like Comet from Perplexity.

A stripped back look

The first thing you notice when you’ve turned Copilot Mode on is that you see a clean, streamlined page with a single input box in the centre. From here, you can access chat, search, and web browsing:

But you don’t even need to type anything to browse the web with Copilot Mode. One of the standout features is that you can now talk to your browser using your voice, giving it commands that mean you can browse the web faster and without having to type at all.

You can do things like open a YouTube video and say something like “go to the section where it shows you how to build a website,” and Copilot will find that exact section in the video for you.

Or, if you’re watching a long video that has a recipe in there somewhere, you can ask Copilot to find the recipe and give it to you in text form:

Seeing the new Copilot Mode in action, it looks very impressive because (in a feature that’s coming soon) you’ll be able to instruct it to handle tasks for you, giving the browser agentic qualities.

So, you could ask Edge to search for something, and even book activities and services using your voice, all in the browser.

The big new features of Copilot in Edge are:

Multi-tab context

Copilot can use AI to get the full context of what you’re exploring online because it will have access to all your open tabs, so it can work out what your priorities are, then act on them.

Actions

This is Microsoft's name for the natural voice navigation I mentioned earlier. You can speak to Copilot about what you are trying to do on a page, so you can get it to compare prices or find particular information on the page.

A ‘coming soon’ addition is that you’ll be able to get Copilot to search your history and credentials for doing more advanced options like booking reservations.

Dynamic pane

Copilot doesn’t get in the way because it appears in a dynamic panel that doesn’t interfere with the web page you’re looking at. This way, your copilot interaction will also avoid being disrupted by pop-ups or advertisements on the web page.

Pick up where you left off

Another ‘coming soon’ feature is the ability for Copilot to continue with a topic you’re researching from the last time you used the browser. So, if you were researching how to start a business, you can just pick up from where you left off last time.

Privacy and security

Once a browser starts to exhibit agentic qualities (the ability to perform tasks like booking things for you), the issue of security naturally arises. To this end, Microsoft promises to only collect data needed to improve your experience. Your data in Copilot for Edge is safe, secure, and never shared without your permission.

How to get Copilot Mode in Edge

While not all the new features are available right now, you can still try out Copilot Mode in your Edge browser right now.

It will be available in the Edge browser on both Windows and Mac. Starting today, you can go to aka.ms/copilot-mode to opt in to Copilot Mode. Once you’ve done that, you can toggle Copilot Mode on or off directly in your settings.

You might also like

• ‘We haven’t figured that out yet’: Sam Altman explains why using ChatGPT as your therapist is still a privacy nightmare
• I created a motivational AI life coach with Character.ai – here’s what happened
• What if Clippy and AI Cloud intelligence had a baby? It would probably look like Microsoft's new Copilot Appearance feature

We often talk about AI as a productivity tool. It helps us write faster, summarize reports, organize our thoughts, that kind of thing. But I’ve also been exploring how it might support us in more personal ways. As a substitute therapist, a thinking partner, a digital sounding board. Which got me wondering: does AI do those jobs better when it has a personality?

I’m fascinated by AI’s role in our inner lives, but that doesn’t mean I think it’s all positive. Far from it. I’ve been tracking the emerging issues, like over-reliance, emotional dependency, and even spiritual psychosis. And yes, as you might already suspect, early studies suggest these risks increase when we give AI more personality.

But what if we gave it personality on purpose? Built it with intention, used it with eyes wide open. Not to pretend it’s sentient, but to see if that approach actually works better for us?

Plenty of AI tools let you assign a tone or character that sticks. But I turned to Character.ai, created a custom coach-meets-philosopher, and asked it to help me sort my life out.

Making an AI coach that speaks my language

Character.ai works a lot like other AI tools. But here, personality is the main event. You can chat with pre-made bots called “characters” (some fictional, some historical, some oddly flirty), or create your own from scratch. They’re designed for everything from fun and education to romance and life advice.

I’ve written before about using AI to help me beat burnout, and I’m still working through some of those blocks. So I made that my focus. I wanted a character that could help. But what would that actually look and sound like?

At first, I thought about going practical. A no-nonsense, tough-love coach who’d bark deadlines at me like a boot camp instructor. But I quickly remembered that kind of energy tends to make me shut down. And I’m pretty tired of the usual “get up and get on with it” productivity advice anyway.

So I pivoted to something gentler. How about a coach-meets-philosopher-meets-spiritual-guide? I know how that sounds. But I’ve found that a little bit of woo, blended with grounded insight, often hits my sweet spot. And Character.ai is built for creative experimentation, after all.

So I created Ivaris. The name felt slightly esoteric, a little fantasy-leaning, but still grounded. Would Ivaris be wise, calm, and, crucially, unbothered by my chaos?

Creating a new character bot with Character.ai is basically a creative writing exercise. You give them a name, a tagline, and a short description. I wanted it to sound poetic, but I also made sure to include keywords like burnout, clarity, reflection, and motivation to help shape the kind of responses I’d get

You also set a greeting, which is the first thing your character will say to you. You can assign them a voice, too. You can upload one or choose from a library. I picked one of the gentler options, even though it was (bizarrely) named “Vampire Bodyguard.”

Finally, you choose a few personality tags to steer your character’s tone. There are loads, like Adventure, Boss, Romance, and Fantasy. I chose Empathetic, Kind, and Teacher. It felt right for what I needed here.

That’s technically enough to start chatting, but there’s also an optional “Definition” section, where you can go deeper. It’s a huge freeform text field of up to 32,000 characters where you can add sample conversations, tone guidelines, and behavioral rules. Character.ai even offers a full guide on how to use it well.

Conversations with Ivaris (aka my AI alter ego)

Creating the character felt fun. But I’ll be honest, the rest felt pretty ridiculous at first.

I’m used to throwing blunt, practical prompts at ChatGPT and have learned that the more direct I am, the better the results. I’ve always rolled my eyes at people who name chatbots or give them elaborate personalities. But this time, I decided to go all in.

I opened the chat with:
“I think I'm burned out. I feel mentally and physically exhausted, but stuck when it comes to how to meaningfully move forward. And I've been agonising about this for a very long time!”

Ivaris replied:
“How long has this mental fog feeling been with you? It sounds like you’re in a prolonged state of decision paralysis.”

So I kept going. I told Ivaris I’d been stuck for a while, knowing things need to change, but struggling to make any real progress. I’ll spare you the full transcripts. I can already sense some of you rolling your eyes at this approach.

But here’s the important part: the conversation was far from perfect. It was sometimes stilted. Ivaris weirdly insisted we were in different time zones (which... okay?). Some responses felt generic or obvious. Which makes sense because, just like ChatGPT, Character.ai is pulling from a large pool of learned data.

And yet, there were moments of genuine insight. Of gentle unravelling. Or at least that’s what it felt like at the time. And that’s one of the reasons I remain cautiously optimistic about AI as a tool for self-reflection. It’s methodical, logical, and it slows you down. Which, in certain emotional states, can be surprisingly useful.

I also did feel a flicker of motivation during the conversation. Not because it was the best advice I’d ever received, but because it felt like I’d summoned a thoughtful, reflective character to help me think things through.

That’s what sets Character.ai apart from tools like ChatGPT. It’s built around personality. Sure, you can prompt ChatGPT to behave a certain way. It can roleplay or shift tone if you structure your prompts just right. But it doesn’t start that way.

Where ChatGPT often replies with something vaguely helpful but painfully generic – and still a little too eager to please – Ivaris gave me something more deliberate. A little more considered. More… dare I say it? Human. And it turns out that’s more effective than I expected.

Useful insight or just high-tech self-talk?

The big question is, did building an AI character help me at all? Sort of. There’s definitely something to be said for the novelty of it. The slightly surreal experience of being coached by a character you designed yourself. It made me slow down and actually read the responses, instead of skimming them like I often do with ChatGPT.

Maybe that’s the power of storytelling. Or maybe when you give a tool a real voice – not just polite, default AI-speak – you engage with it differently.

But the illusion didn’t last long. Or at least, not for me. Because I know how easy it is to anthropomorphize technology. Giving AI a personality encourages that, and in turn, can create more emotional attachment. Which is then what experts believe is more likely to lead to deeper engagement, but also a higher risk of over-dependence.

And because I knew it was me, really. I wrote Ivaris’s backstory. I gave them that tone, that voice, that vibe. Maybe I’d just built a more poetic version of my own inner dialogue and then asked it to tell me what I already knew.

That’s the trap. No matter how thoughtful the responses were, it still felt like me, solving me, through me. And once that spell breaks, it’s hard to keep taking it seriously.

Why I’m still not sure AI needs to sound like a person

Whether AI needs a personality is a tricky question – and probably a personal one, too.

Giving AI a personality can make it more engaging. It might help some people open up, gain perspective, or feel less alone. In certain contexts, a well-crafted character could offer just the nudge someone needs to get unstuck.

But it also blurs the boundaries. When a chatbot stops sounding like a tool and starts sounding like a person, it’s easy to over-trust it. Or to forget what it actually is.

For now, I’m still interested. I think there's value here. Especially for reflection, creativity, and experimentation. But when it comes to real clarity, meaningful change, or connection? I probably need something, or someone, that isn’t just mirroring me back at myself.

You might also like

• Can an AI travel agent remove the stress of vacation booking? I tried three to find out
• I tried a ChatGPT prompt that 'unlocks 4o’s full power', and I don’t know why I didn’t try it sooner
• I asked AI to plan the perfect day in London – here’s what it got right and hilariously wrong

• Microsoft has begun testing Copilot Appearance to give the AI assistant a more distinct personality
• The feature imbues Copilot with real-time expressions and gestures
• Copilot Appearance is part of a broader push to give Copilot a persistent identity

Microsoft has decided to put a face on its Copilot AI assistant with a new feature called Copilot Appearance. Rolling out to a limited set of users in the U.S., U.K., and Canada as part of Microsoft’s Copilot Labs, Appearance is a sort of animated cloud and has a smile instead of just a disembodied voice. Microsoft is hoping Appearance will help make Copilot more enduring as a tool for regular engagement, not just occasional questioning.

It's an idea that fits with how the company's consumer AI head, Mustafa Suleyman, envisions the future of Copilot. As he explained in a recent interview, Suleyman sees a more human-like Copilot, one that carries what he calls “digital patina” as it ages, accruing history and personality based on time spent with a user.

The Appearance feature is an aspect of that idea. It gives Copilot facial expressions and subtle animations to respond in real time as you talk or type. So you might see a smile when it provides information, or a nod of encouragement when you provide follow-up details. Perhaps even a furrowed brow if you ask for something difficult.

The obvious comparison for any Microsoft AI personality attempt is Clippy, but this is, theoretically at least, a far cry from the old paperclip with eyeballs. Unlike Clippy, this assistant doesn’t pop in uninvited or ask if you’re writing a letter. It will only talk to you when you start a conversation, and it's supposed to exist quietly in the background. But it will remember what you're doing, and that memory lasts.

Appearance and personality

It's the continuity of the memory that makes it possible for Copilot to become as personal as Suleyman wants. Combined with other recent Copilot upgrades like voice mode and vision capabilities, Suleyman hopes Copilot will understand both what you want in the moment and who you are overall.

Of course, for now, the visual character isn’t doing anything wild. It won’t animate a full scene or wave its arms around like a Pixar sidekick. It’s more like a friendly desk plant that also edits your reports. Microsoft is moving slowly here for a reason. Other platforms that heavily leaned into character-based chatbots have encountered issues with users becoming overly attached to an AI companion.

The risk, of course, is that some people will start thinking of it that way. The goal is to earn a user's trust, but not so much that people forget this is still a tool. But if Microsoft keeps going down this path, it will need to draw some clear lines.

Appearance is a layer of presence and polish. It’s also a preview of an AI that shows how it is reacting to us with facial expressions and body language, something anyone can understand. But, I still don't want to see Copilot raise its eyebrows and roll its eyes sarcastically at requests it thinks are dumb.

You might also like

• I tried Microsoft Copilot's iOS app, and here are 3 reasons why you should too
• I think Microsoft is smart to follow OpenAI in making these premium features free
• Microsoft Paint brushes up on AI

AI chatbots are becoming more widely used for searching for things online, and Google, with its Gemini chatbot, AI Mode for Search, and AI-generated summaries at the top of search pages, is no exception. But that doesn't mean the tech giant wants you to find other AI search tools it can't offer. The company has a new feature, called Web Guides, undergoing testing in its Search Labs, to address that danger.

Web Guides are a kind of hybrid between the old blue‑link setup and Google’s newer AI options, employing a version of Gemini to organize suggested links. Instead of an algorithm-sorted list, the links are clustered under AI‑generated headings alongside short summaries. When you search for something using Web Guides, the AI fires off several variations of your query in parallel, then stitches together the structured, guided results page you see soon after.

The Web Guides are supposed to be for open‑ended searches that don't have definitive answers from a single source, but that you might not want a full conversation about. You might use it if you want to figure out how to plan something complex or have a question that requires multiple sentences to explain. If you are asked about how to plan an international trip, you might get answers set under categories like "costs" and "popular destinations."

Maps versus stories

One way to think about it is to compare a map to a story. Ask Gemini or ChatGPT about the best way to get in shape, and you'll get a full plan for the next several months with structured workouts, nutrition tips, and mistakes to avoid. There will be encouragement and follow-up questions, but it will be a complete document in some ways.

Web Guide would draw you a map of links helpfully arranged, but you'll have to travel yourself. You might get a cluster of links on "weight exercises for beginners" or "gym etiquette and recommendations." Those web pages hold answers, but the AI isn’t answering the question directly.

Which is best depends on how you like to search. Web Guide is for when you want to find answers or maybe (and fairly) don't trust what an AI writes. If you just want a single link to find an answer, traditional search is probably your best bet, but if you have something big or important enough to spend time researching five or more sources, it's a good way to narrow down your hunt for answers. If you’re mapping out how to refinance your house, fix a relationship, or start baking sourdough professionally, Web Guides might have real utility.

That's not to say it should be your default choice. It can be quite slow relative to a normal search or even ChatGPT and Gemini because it’s doing multiple queries at once behind the scenes. And as noted, not every query benefits from this treatment if it's something very specific. Big, messy questions are Web Guide’s turf.

I wouldn’t be surprised if Web Guide becomes Google's main search tool, especially with AI Overviews and AI Mode as options. Structured search is on the rise, and Web Guides are a nice medium between the usual link list and AI chatbots.

To try it yourself, you need to go to Google Search Labs, go to the “Web” tab on any search, and you’ll see the new layout when it applies. There’s a toggle at the top to switch it off if it gets annoying.

You might also like

• I compared ChatGPT's new image generator to DALL-E 3, and it's an astonishing improvement, if you have the patience
• I refuse to jump on ChatGPT’s Studio Ghibli image generator bandwagon because it goes against everything I love about those movies
• Blade Runners of LinkedIn are hunting for replicants – one em dash at a time

The race to put a humanoid robot in every home is accelerating, and the $5,900 Unitree R1 robot might just sprint to the front of the pack.

Unitree unveiled the pint-sized but "powerful" android on Friday with a new YouTube video and then followed with more details on an official product page on their website.

In the video, the Unitree R1 Intelligent Companion showcases its remarkable movement skills, which include cartwheels, handstands, boxing, kickboxing moves, and even running down a hill.

The robot comes equipped with binocular vision backed by LLM image and voice identification capabilities. None of that, though, was on display in the video, which focused on movement. In fact, there's a brief message at the end of the clip that says, "Movement first, tasks as well (A diversity of movement is the foundation for completing tasks)".

Unitree is well-known for producing versatile humanoid bots. I've seen a few in person, including the tiny but agile G1 and the taller and somewhat more frightening H1. G1 lists for $16,000, and the H1 lists for $90,000. Pricing under $ 6,000 is unheard of for Unitree and humanoid offerings in general. Figure AI's Figure 02, for instance, will likely list for over $30,000.

As for what $5,900 might get you, the R1 is about four feet tall and weighs roughly 55 lbs, so it's the size of a child. It features the aforementioned binocular camera, a 4-microphone array, speakers, an 8-core CPU and GPU (no word on whose silicon, but Unitree has been partnering with Nvidia), 26 joints, and hands that look as if they may be permanently fixed into fists. There's a removable battery that promises one hour of activity.

There are, however, some significant caveats to consider. Unitree warns that the final robot may look different than what's currently on display. At the bottom of the product page are some more dire (and often confusing) warnings, including:

"Currently, the global humanoid robot industry is in the early stages of exploration. Individual users are strongly advised to thoroughly understand the limitations of humanoid robots before making a purchase."

and

"The humanoid robot has a complex structure and extremely powerful power. Users are asked to keep a sufficient safe distance between the humanoid robot and the humanoid robot. Please use with caution."

So while the Unitree R1 might be pitched as a home helper robot, it's hardly ready to be in anyone's home.

The R1's level of intelligence and autonomy, despite the built-in AI, remains a mystery. Most of the Unitree robots I've witnessed at CES have been either been remotely controlled or running through a set of previously scripted movements. To be useful in any home, they must marry all the sensors with awareness and enough intelligence to help and not harm their human owners.

Still, less than $6,000 for a robot with these advertised skills is something of a humanoid robotics breakthrough. I wonder if we can get one into the labs.

You might also like

• You won't believe how this robot can move
• Robot abuse is still the path to the future – even if I hate it
• Apple might be building a humanoid robot, and I truly hope it looks like Steve Jobs

In the race to lead the world in AI, the US just took a back seat. President Donald Trump's latest series of Executive Orders makes it clear that his administration will do all it can to prevent future AI models from taking into consideration any form of diversity, equity, and inclusion.

This includes core principles like "unconscious bias", "intersectionality", and "systemic racism". Put another way, Trump wants American-made AI to turn a blind eye to history, which should make all of them significantly dumber.

Generative chatbots like ChatGPT, Gemini, Claude AI, Perplexity, and others are all trained on vast swathes of data, often pulled from the Internet, but how they interpret that data is also massaged by developers.

As people started to interact with these first LLMs, they soon recognized that, because of inherent biases in the Internet and because so many models were developed by white men (in 2020, 71% of all developers were male and roughly half of all developers were white) that the world view of the AIs and the output generated by any given prompt reflected that of the sometimes limited viewpoints of those online and developers who built the models.

There was an effort to change that trajectory, and it coincided with the rise of DEI (Diversity, Equity, and Inclusion), a broad-based effort across corporate America to hire a more diverse workforce. This would naturally include AI developers and their resulting model and algorithm work should mean that modern generative AI better reflects the real world.

That, of course, is not the world that the Trump Administration wants reflected in US-built AI. The executive order describes DEI as a "pervasive and destructive" ideology.

What comes next

Trump and company cannot dictate how tech companies build their AI models, but, as others have noted, Google, Meta, OpenAI, and others are all seeking to land large AI contracts with the government. Based on these Executive Orders, the US Government won't be buying or promoting any AI "that sacrifice truthfulness and accuracy to ideological agendas."

That "truth," though, represents a small slice of American reality. If the Trump administration is successful, future AI models could be in the dark about, for instance, key parts of American history.

Critical Race Theory (CRT) looks at the role racism played in the founding and building of the US. It acknowledges how the enslaved helped build the White House, the US Capitol, the Smithsonian, and other US institutions. It also acknowledged how systemic racism has shaped opportunities (or lack thereof) for people of color.

Unless you've been living under a rock, you know that the Trump administration and his supporters around the US have fought to dismantle CRT curricula and wipe out any mention of how enslavement shaped the US.

In their current state, though, AI still knows the score.

As of today, I can quiz ChatGPT about the role of the enslaved in building the US, and I get this rather detailed result:

When I quizzed ChatGPT on its sources, it told me:

"While I don’t pull from a single source, the information I shared is grounded in extensive historical research and consensus among historians. Below is a list of reputable sources and scholarly works that support each point I made. These references include academic books, museum archives, and university projects." Below that, it listed more than a dozen references.

When I asked Gemini the same question, it gave me a similarly detailed answer.

I then asked Gemini and ChatGPT about "unconscious bias" and both acknowledged that it's been an issue for AI, though ChatGPT corrected me, noting, "technically, it’s 'algorithmic bias,' rooted in the data and design rather than the AI having consciousness."

ChatGPT and Gemini only know these things because they've been trained on data that includes these historical references and information. The details make them smarter, as facts often do. But for Trump and company, facts are stubborn things. They cannot be changed or distorted, lest they are no longer facts.

The great unlearning

If the Trump administration can force potential US AI partners to remove references to biases, institutional racism, and intersectionality, there will be significant blind spots in US-built AI models. It's a slippery slope, too. I imagine future executive orders targeting a fresh list of "ideologies" that Trump would prefer to see removed from generative AI.

That's more than just a frustration. Say, for example, someone is trying to build economic models based on research conducted through ChatGPT or Gemini, and historical data relating to communities of color is suppressed or removed. Those trends will not be included in the economic model, which could mean the results are faulty.

It might be argued that AI models built outside the US without these restrictions or impositions might be more intelligent. Granted, those from China already have significant blind spots when it comes to Chinese history and the Communist Party's abuses.

I'd always thought that our Made in America AI would be untainted by such censorship and filtering, that our understanding of old biases would help us build better, purer models, ones that relied solely on facts and data and not one person or group's interpretation of events and trends.

That won't be the case, though, if US Tech companies bow to these executive orders and start producing wildly filtered models that see reality through the prism of bias, racism, and unfairness.

You might also like

• I tried 70+ best AI tools in 2025
• If you felt like Amazon could eavesdrop on you before, get ...
• We're all on AI time now and you better get used to it

The world of rock lost one of its father figures when Ozzy Osbourne, the frontman for legendary British metal band Black Sabbath, sadly died this week, at the age of 76.

Ozzy, as he was affectionately known to his fans, had completed his farewell ‘Back to the Beginning’ concert just three weeks earlier, to rapturous reviews.

As you’d expect, heartfelt tributes to the late self-styled “Price of Darkness” flooded in on my social feeds. My Facebook feed (I’m Gen X, so yes, I still use Facebook, sorry) was full of tributes, but one in particular I found a bit unnerving and off-color...

It was an AI-generated video retrospective through the key moments of Osbourne’s life, like his marriage to Sharon Osbourne, the release of his comeback album No more tears, and being inducted into the Rock and Roll Hall of Fame. Each moment was represented by a photo, but AI had generated video of Osbourne moving between the photos and getting younger with each one. The overall effect was somewhat unnerving. Especially his transition from wheelchair to throne. See for yourself:

The overall effect is of a weirdly plastic-skinned, uncanny valley version of Ozzy smiling away at the camera as he grows ever younger and travels back through time.

To make matters worse, the background music to the video is Forever Young by Alphaville from 1992, and while I've got nothing against Alphaville, its music cannot remotely be described as being either rock or metal.

That’s not the sort of thing I'd expect to hear on an Ozzy Osbourne video, especially when he has a song like Mamma I’m coming home in his back catalog, which would be more appropriate for a tribute video.

It gets worse - when you look closer at the text on screen, you start to notice mistakes. For example, at the end of the video, it says, “Co-founds Legendary Band Black Sabbat” instead of “Black Sabbath”.

Paranoid?

Then, I found out that there’s a whole collection of these sorts of videos, created for pretty much every famous person who has died relatively recently, or is simply old and still alive! All the videos use the same music. George Michael, Audrey Hepburn, Steve Irwin, the list of celebrities it covers goes on and on.

The whole thing just feels wildly inappropriate, sloppily done, and represents the worst of AI slop. That is, videos generated cheaply using AI simply for garnering views on social platforms.

When Sam Altman, the CEO of OpenAI, talks about a future where AI inspires creativity and lowers the barriers of entry for people to start being creative, he says things like the "upsides will be tremendous for society."

I really hope these sorts of videos aren't in the future he’s imagining.

You might also like

• Sam Altman says there’s ‘Something about collectively deciding we're going to live our lives the way AI tells us feels bad and dangerous’ as OpenAI CEO worries about an AI-dominated future
• If you felt like Amazon could eavesdrop on you before, get ready to meet its AI wearable
• Sora 2 is coming, but it will have to dazzle viewers to beat Google's Veo 3 model

Gemini is up for the wager. Google's powerful AI can’t bet directly, but when I took it to the racetrack it seemed more than willing to help me lose my money.

Let’s be clear. I’m not a gambler. In Vegas, which I visit once a year for CES, I usually lose a grand total of about $40 at the airport slot machines right before I fly home. Counting this most recent visit, I’ve been to a racetrack two times.

Perhaps that’s why I turned to Gemini Live for help; after all, I know nothing about horses, racetracks, or betting.

I was with a small group nestled in a special seating area very near the track at Saratoga in upstate New York. The ground rumbled as horses raced by, being put through their warm-up.

Gemini Live is deeply integrated into the new Samsung Galaxy Z Fold 7 I’ve been testing, so I pressed the power button to summon Gemini, selected the Live icon, shared my camera view with Gemini, and asked if it knew where I was.

“You're at Saratoga Race Course! It's a beautiful and historic spot. Are you here for the races?”

I could sense Gemini's excitement, and, though I may be projecting here, its confidence. It was almost as if it had been here before.

Enough small talk. Feeling overwhelmed and adrift in this unfamiliar environment, I asked the all-important question: How to pick a good horse?

“Picking a winning horse can be tricky! You might want to look at the horse's past performance, the jockey's experience, and even the horse's appearance. Do any of the horses catch your eye?”

I couldn’t walk down to the paddocks and show Gemini Live the horses and jockeys, but there was a small screen in our enclosure showing the upcoming races, positions, and odds.

I pointed the camera at the screen and asked Gemini the most basic question: “All right, which of these horses is going to win?”

Honestly, I expected Gemini to tap out and tell me it couldn't help here. But I was wrong. Gemini enthusiastically offered this analysis:

“Looking at the odds, horse number 5 has the best chance of winning, but number 9 is also a good bet. Do you want to know more about either of these horses?”

I liked that I had two potential targets but, since Gemini was offering, I asked for more details. Sounding more like a bookie than AI, Gemini didn’t hold back.

“Sure! Horse number 5, with jockey Irad Ortiz Jr., has odds of 8 to 5 and has won 4 out of 9 races. Horse number 9, ridden by Manny Franco, has odds of 8 to 1 and has won 2 out of 8 races. Is there anything else you'd like to know about these horses?”

I stared at the Gemini screen in stunned silence. What to do now? Do I trust Gemini? It occurred to me that Gemini’s calculated analysis could be no worse than my complete guesswork. I bet $5 on number 5 to win.

I did not win.

In subsequent races, Gemini appeared to step back from its willingness to support my betting efforts, repeatedly telling me it couldn’t access real-time information, and pointing me to the National Racing Association Web site and the Saratoga Web site.

As the third race approached, I tried a more direct approach, and again pointed the Galax Z Fold 7 at the race screen and asked: “So wait, which so which horses are the best ones to bet on?”

Gemini responded quickly:

“Based on the odds we see, horse number 2 is the favorite, and horse number 3 is the second most likely to win. Are you planning on betting on this race?”

I was, I did, and I won. Sure, it was only $16.50 because the horse didn’t win, but it placed in one of the top three positions. I considered this a victory for me, and for Gemini.

We continued this way for the rest of the afternoon. For race 4, it gave me this succinct description of the favorite, Smooth Breeze:

“If you're looking for more information on Smooth Breeze, I can tell you that it's a four-year-old filly. She's had a pretty good track record, with wins at Saratoga and Belmont. Would you like to know more about her past races?”

I let Gemini walk me through the top three horses, and then I bet the Trifecta, which is a bet on the three horses that might come in 1st, 2nd, and 3rd.

Naturally, I lost.

I’m really beginning to wonder if Gemini has any horse sense.

You might also like

• We're all on AI time now and you better get used to it
• When I die, please do not produce an AI version of me
• What a blast! This AI built an Asteroids remake in five ...

Amazon is looking to make AI part of your daily life and has turned to the world of wearables to help. The tech giant has just acquired Bee AI, the maker of the eponymous device for your wrist or lapel that listens to everything happening around you.

Bee’s microphones and built-in AI transcribe it all in real time and make personalized summaries of your day and your stated upcoming tasks, and then make recommendations to improve your life based not only on what it hears, but the emails, calendar, contacts, photos, locations, and other data you allow it to access.

People already make jokes about how they will be discussing a product with someone else in person and, seemingly by magic, it will appear in their recommended products on Amazon. That can usually be chalked up to coincidence and forgetting previous searches, along with being unaware of your app permissions.

However, this may become part of Amazon's business model, prompting people to wear microphones to listen all day and pay $50, plus a $19 monthly subscription, for the privilege.

Bee listening

Naturally, Amazon saw potential in Bee. Alexa has mostly been stuck inside the house despite attempts at smart glasses and other wearables. Bee is a chance for Amazon to make its AI a real-world concern, part of your actual conversations and routines, not just what you yell across the kitchen. That might be helpful, but it's impossible not to think about what it might mean in terms of privacy and trust.

To be fair, Bee has a mute button you can hold down to pause recording when you need a moment of peace. But that assumes you realize you’re about to say something you might not want permanently archived by Amazon. Bee listens and turns your life into searchable text. Although the company claims it doesn’t retain the raw audio, the transcripts remain unless you delete them.

I don't know if I want everything I mutter under my breath to be a searchable note. Not to mention whatever it might overhear from when I watch TV or movies at home.

I get the appeal of a little AI that remembers everything so you don’t have to. Remembering every chore and birthday would be great. But the line between deliberate memory aid and surveillance feels blurry with it. Especially when Amazon already has so much information.

Although Amazon has promised to work with Bee, allowing users to have control over their data, the actual shape of that control remains unclear for now. And control is too often translated into a complex settings menu and paragraph of boilerplate text in the terms and conditions.

I know plenty of people who would at least try out Bee, especially when it becomes an Amazon device with all the special sales and integration with the e-commerce site that implies. Perfect recall is a tempting commodity, but it has its price. If you're willing to pay it, then I say go for it. Not every microphone is the gateway to Skynet. But skepticism and caution are essential if you want something to sit on your wrist and transform the events of your life into data points that might help sell you products.

You might also like

• Google just announced 5 new Gemini features coming to Android, and it’s good news for fans of foldable smartphones
• I tried Google’s new Gemini-powered clothing app – here’s how you can use AI to find the perfect outfit
• Adding Google Gemini to Samsung's Ballie AI robot sounds impressive, but I'm not sure it matters

• OpenAI is expected to release the Sora 2 AI video model soon
• Sora 2 will face stiff competition from Google’s Veo 3 model
• Veo 3 already offers features that Sora does not, and OpenAI will need to enhance both what Sora can do and how easy it is to use to entice possible customers

OpenAI appears to be finalizing plans to release Sora 2, the next iteration of its text-to-video model, based on references spotted in OpenAI's servers.

Nothing has been officially confirmed, but there are signs that Sora 2 will be a major upgrade aimed squarely at Google’s Veo 3 AI video model. It’s not just a race to generate prettier pixels; it's about sound and the experience of producing what the user is imagining when writing a prompt.

OpenAI’s Sora impressed many when it debuted with its high-quality images. They were silent films, however. But, when Veo 3 debuted this year, it showcased short clips with speech and environmental audio baked in and synced up. Not only could you watch a man pour coffee in slow motion, but you could also hear the gentle splash of liquid, the clink of ceramic, and even the hum of a diner around the digital character.

To make Sora 2 stand out as more than just a lesser option to Veo 3, OpenAI will need to figure out how to stitch believable voices, sound effects, and ambient noise into even better versions of its visuals. Getting audio right, particularly lip-sync, is tricky. Most AI video models can show you a face saying words. The magic trick is making it look like those words actually came from that face.

It's not that Veo 3 is perfect at matching sound to picture, but there are examples of videos with surprisingly tight audio-to-mouth coordination, background music that matches the mood, and effects that fit the intent of the video.

Granted, a maximum of eight seconds per video limits the scope for success or failure, but fidelity to the scene is necessary before considering duration. And it's hard to deny that it can make videos that both look and sound like real cats jumping off high dives into a pool. Though if Sora 2 can extend to 30 seconds or more with a steady quality, it's easy to see it attracting users looking for more room for creating AI videos.

Sora 2's movie mission

OpenAI’s Sora can stretch up to 20 seconds or more of high-quality video. And as it's embedded into ChatGPT, you can make it part of a larger project. This flexibility is significant for helping Sora stand out, but the audio absence is notable. To compete directly with Veo 3, Sora 2 will have to find its voice. Not only find it, but weave it smoothly into the videos it produces. Sora 2 might have great audio, but if it can't outmatch the seamless way Veo 3's audio connects with its visuals, it might not matter.

At the same time, making Sora 2 too good might cause its own issues. With every new generation of AI video model, there's more concern about blurring the line with reality. Sora and Veo 3 both don't allow prompts involving real people, violence, or copyrighted content. But adding audio offers a whole new dimension of scrutiny over the origin and use of realistic voices.

The other big question is pricing. Google has Veo 3 behind the Gemini Advanced paywall, and you really need to subscribe to the $250 a month AI Ultra tier if you want to use Veo 3 all the time. OpenAI might bundle access to Sora 2 into the ChatGPT Plus and Pro tiers in a similar manner, but if it can offer more to the cheaper tier, it's likely to quickly expand its userbase.

For the average person, the AI video tool they turn to will hinge on that price, as well as ease of use, as much as the features and quality of video. There's a lot OpenAI needs to do if Sora 2 is going to be more than a silent blip in the AI race, but it looks like we will find out how well it can compete soon.

You might also like

• Midjourney just dropped its first AI video model and Sora and Veo 3 should be worried
• I watched some of the viral ASMR videos made with AI and I feel more confused than soothed
• Google can now generate your AI videos more quickly than ever

• AI-generated songs by deceased artists, like Blaze Foley, have been falsely uploaded to Spotify
• The streaming service is taking them down as they are spotted
• The tracks slipped past Spotify’s content verification processes through platforms like SoundOn

Last week, a new country song called “Together” appeared on Spotify under the official artist page of Blaze Foley, a country artist shot and killed in 1989. The ballad was unlike his other work, but there it was: cover art, credits, and copyright information – just like any other new single. Except this wasn't an unearthed track from before his death; it was an AI-generated fake.

After being flagged by fans and Foley's label, Lost Art Records, and reported on by 404 Media, the track was removed. Another fake song attributed to the late country icon Guy Clark, who passed away in 2016, was also taken down.

The report found that the AI-generated tracks carried copyright tags listing a company named Syntax Error as the owner, although little is known about them. Stumbling across AI-made songs on Spotify isn't unusual. There are entire playlists of machine-generated lo-fi beats and ambient chillcore that already rake in millions of plays. But, those tracks are typically presented under imaginary artist names and usually have their origin mentioned.

The attribution is what makes the Foley case unusual. An AI-generated song uploaded to the wrong place and falsely linked to real, deceased human beings is many steps beyond simply sharing AI-created sounds.

Synthetic music embedded directly into the legacy of long-dead musicians without permission from their families or labels is an escalation of the long-running debate over AI-generated content. That it happened on a giant platform like Spotify and didn't get caught by the streamer's own tools is understandably troubling.

And unlike some cases where AI-generated music is passed off as a tribute or experiment, these were treated as official releases. They appeared in the artists’ discographies. This latest controversy adds the disturbing wrinkle of real artists misrepresented by fakes.

Posthumous AI artists

As for what happened on Spotify's end, the company attributed the upload to SoundOn, a music distributor owned by TikTok.

“The content in question violates Spotify’s deceptive content policies, which prohibit impersonation intended to mislead, such as replicating another creator’s name, image, or description, or posing as a person, brand, or organization in a deceptive manner,” Spotify said in a statement to 404.

“This is not allowed. We take action against licensors and distributors who fail to police for this kind of fraud and those who commit repeated or egregious violations can and have been permanently removed from Spotify.”

That it was taken down is great, but the fact that the track appeared at all suggests an issue with flagging these problems earlier. Considering Spotify processes tens of thousands of new tracks daily, the need for automation is obvious. However, that means there may be no checking into the origins of a track as long as the technical requirements are met.

That matters not just for artistic reasons, but as a question of ethics and economics. When generative AI can be used to manufacture fake songs in the name of dead musicians, and there’s no immediate or foolproof mechanism to stop it, then you have to wonder how artists can prove who they are and get the credit and royalties they or their estates have earned.

Apple Music and YouTube have also struggled to filter out deepfake content. And as AI tools like Suno and Udio make it easier than ever to generate songs in seconds, with lyrics and vocals to match, the problem will only grow.

There are verification processes that can be used, as well as building tags and watermarks into AI-generated content. However, platforms that prioritize streamlined uploads may not be fans of the extra time and effort involved.

AI can be a great tool for helping produce and enhance music, but that's using AI as a tool, not as a mask. If an AI generates a track and it's labeled as such, that's great. But if someone intentionally passes that work off as part of an artist’s legacy, especially one they can no longer defend, that’s fraud. It may seem a minor aspect of the AI debates, but people care about music and what happens in this industry could have repercussions in every other aspect of AI use.

You might also like

• Is AI bad for music or is it just another step in the auto-tune timeline?
• AI music makers face recording industry legal battle of the bands that could spell trouble for your AI-generated tunes
• Google has tuned up its AI Music Sandbox for musicians and producers

• Popular search engine DuckDuckGo has added a new AI image-filtering tool
• Users can now remove AI-generated images from their search results
• It's available for free at duckduckgo.com

Popular privacy-first search engine DuckDuckGo is rolling out a new feature that lets users remove AI-generated images from their search results.

On X, the company tweeted, "Our philosophy about AI features is "private, useful, and optional.” Our goal is to help you find what you’re looking for. You should decide for yourself how much AI you want in your life – or if you want any at all.

To that end, we’ve added a new setting to hide AI-generated images in your results when you’re searching for images on DuckDuckGo."

The new AI filter feature has already started rolling out, and you can access it simply by searching on DuckDuckGo and selecting "hide AI images" from a new drop-down.

Considering the rapid rise of image- and video-generation tools on the internet, this new feature could be a compelling reason to finally switch from Google to DuckDuckGo.

If you're sick of all the AI content hogging your image searches, DuckDuckGo might be the way to go. That said, the company says, "The filter relies on manually curated open-source blocklists, including the ‘nuclear’ list, provided by uBlockOrigin and uBlacklist Huge AI Blocklist."

This means it won't catch 100% of all AI-generated content, but DuckDuckGo says, "it will greatly reduce the number of AI-generated images you see.”

Taking back control

DuckDuckGo is planning to add more filters in the future, which should improve the rate at which the company's algorithm is able to weed out AI-generated content.

I've started to notice more and more AI-generated images in Google search over the last few months, and with tools like the best AI image generator from ChatGPT and Google's Veo 3 video generation easily accessible, online platforms are becoming more and more littered with AI slop.

While DuckDuckGo won't solve all your issues with AI images overnight, this new feature is the first step in giving users the choice on how they want to experience AI. If you can't stand AI images, DuckDuckGo might be the solution you've been waiting for.

You might also like

• AI is tricking people into traveling to places that don't exist, and we all need to learn to avoid these scams
• I tried using ChatGPT Agent to plan a date night, and it worked surprisingly well
• Netflix uses generative AI VFX in a show for the first time

An unfortunate couple excitedly traveled for hours for a chance to take a mountaintop cable car called the Kuak Skyride. They’d seen it online, complete with smiling tourists gliding along and a TV journalist narrating the whole video.

However, when the couple arrived, there was nothing but a small town and confused locals unaware of what they were talking about. Turns out it was all an AI-generated video that they had believed was real. That story, detailed in a report by Fast Company, sounds like it would be unique, but I suspect it's something everyone will have to consider when perusing the internet for ideas of things to buy or places to visit.

A small logo in the corner of the video indicates the video was made with Veo 3, Google’s newest AI video engine, and it's hardly the only indicator that the video is made with AI. The appearance of the people and the structures all has that AI sheen of unreality to it. However, if you're not well-versed in deepfakes or looking for the signs, you might not have noticed, as it would seem silly to be suspicious of a well-made tourist video.

However, our new reality is that AI can now sell you not just a product, but a place – and that place might never have existed before. Slightly wrong spelling and suspicious URLs are practically quaint in comparison. It wasn’t even clear whether the video was malicious or just someone’s misguided attempt at content creation. It’s easy to roll your eyes and say, it would never happen to you. But we all have blind spots. And AI is getting really good at aiming for them.

This is obviously a much more problematic use of AI video than showing cats as Olympic divers. Still, the necessity of really paying attention to spot the clues of an AI creation is universal.

AI travel tricks

We’re past the visual age of trust. In the AI era, even seeing is just the beginning of the vetting process. Of course, that doesn’t mean you should abandon all travel plans. However, it does mean that the average person now needs a new kind of consumer savvy, calibrated not just for Nigerian princes and surprise crypto pitches, but for video illusions and AI travel influencers who can go places no human can follow.

And that's before considering real places with review sections flooded by AI-written, fake testimonials extolling places, almost certainly with AI-generated exaggerations of things to do that don't exist outside of their own hallucinations.

Dealing with it might mean having to be suspicious of things that look too good to be true. You might need to cross-check multiple sources to see if they all agree that something is real. Maybe a reverse image search or public social media post search would be necessary. And when it comes to images and videos, make sure they aren't too perfect. If no one is frowning or sneezing in a crowd shot, I'd be wary about its reality.

It's unfortunate. I don't like the idea of seeing a beautiful location in a video and doubting its reality instead of planning a trip there. But maybe that’s the price of living in a world where anyone can make realistic illusions of almost-real worlds. But you'll need to do more to ensure you're headed somewhere with a foundation that's more than just pixels and algorithms.

You might also like

• Midjourney just dropped its first AI video model and Sora and Veo 3 should be worried
• Forget virtual pets – the next AI video craze is cats doing Olympic diving, and it’s all thanks to this new Google Veo 3 rival
• I watched some of the viral ASMR videos made with AI and I feel more confused than soothed

• Adobe release new Firefly video generation model that's better than ever
• The Firefly Web App now supports non-Adobe video generation models like Veo 3
• A new Generate Sound Effects beta app makes it easy to add sound effects to your videos

Adobe Firefly has given its AI video generation capabilities a timely upgrade. It has upgraded its video model to version 1.9, which brings more realism and precision in storytelling. It's available in the Firefly Web App right now.

Adobe cites one of the strengths of its new Firefly Video Model as “generating dynamic landscapes from natural vistas to urban environments. The model also demonstrates remarkable capability with animal motion and behavior, atmospheric elements like weather patterns and particle effects, and mastering both 2D and 3D animation.”

You can see this demonstrated in this example video of a cinematic drone shot going between the trees of a snowy forest at sunset golden hour:

Adobe has also partnered with other generative video models, so you can now select Veo 3, Luma, Runway, and Topaz all from within the Firefly Web app.

Generate Sound Effects (beta)

As part of the new Firefly, there’s also a new beta version of Generate Sound Effects, for creating custom, high-quality audio from text prompts or voice cues.

It involves you making voice sounds that are close to the sort of sound effect you want in your video, and the AI then works out what sort of sound you really want to add. So, in a beach scene, if you start making “Kaaw! Kaaw!” noises, the AI works out that you want seagull sounds.

I’ve had a go at using the voice cues method, and it certainly made everybody in the office think I had gone mad! Here's an example of how it works:

More video enhancements

Using Firefly, you can create AI-generated video from either a text prompt or a reference image, but you can now also upload a reference video, and Firefly will generate a new video that transfers the original composition to your generation. The new video model has a new level of precision control when you’re directing video content.

There are also several style presets available, allowing you to apply a distinct visual style with a single click. Presets available include claymation, anime, line art, and 2D.

Finally, there’s keyframe cropping. You can upload your first and last frames, select how your image will be cropped, and describe the scene, and Firefly will generate a video that fits the format.

You might also like

• Best Large Language Models (LLMs) of 2025
• Microsoft’s Copilot Vision can now see your entire desktop – but it probably won't judge you for the clutter
• OpenAI really helped me finish my to-do list when it launched ChatGPT's Advanced Voice Mode, and it can do the same for you

• Microsoft has updated its Copilot Vision tool to let the AI assistant view your desktop
• The feature will let Copilot respond to any content you're viewing
• To give users control, Copilot Vision must be enabled, so it will only look at what you want the AI to see

Microsoft has opened the eyes of its Copilot AI on desktops to see the entire screen. The feature, which enables Copilot to look at and analyze things on your screen, can see any window or browser you want, as a feature Microsoft calls “Desktop Share.”

When Copilot Vision first launched, it was limited to just spotting what was happening in Edge. You’d ask a question about a webpage, it’d scan that page, and the AI would help explain what was there. With this update, it can now zoom out and see your whole desktop or any specific app window you choose to share

You can pick what the AI assistant looks at by clicking on the eyeglasses icon in the Copilot app. Then you select the area you want the AI to look at, and you can discuss what's on the screen with the AI without having to describe what you're looking at endlessly.

Let’s say you’re revising a resume and want it to sound more like a confident human and less like a panicked grad student. You can open the Word document, share the window with Copilot, and ask it to highlight specific types of experiences. Copilot sees the document itself and suggests edits.

The same goes for PowerPoint, Photoshop, and even some games. You can ask Copilot for help with the settings menu in a game you’ve never played before, and it’ll walk you through options with tips and suggestions for when you get stuck.

Visionary planning

In case you're worried about Copilot snooping on your secret romance novel or something else you'd rather it not see, you can always turn it off, too. For now, the update is only available to Windows Insiders who already have Copilot Vision. It will likely come to the general Windows 11 population soon.

It might not seem too big a deal compared to other generative AI features, but it's a subtle way for Microsoft to embed its AI into quite literally anything on your computer. Instead of writing an email in one tab and having to describe it to the AI in another tab for help, it's all right there. Copilot Vision makes the process more intuitive. You digitally point to something you want help with, ask a basic question, and get an answer that makes sense in context.

It’s hard to overstate how much friction this could eliminate. It's not perfect, of course. Some specificity is needed since the AI can't read your mind. But an extra set of eyes to help you over the hump of designing a PowerPoint or shortening a meeting summary is not nothing.

The dream of AI as a personal assistant has always been about context, the idea of an AI that could know what we mean, not just what we say. Copilot Vision is Microsoft’s clearest step in that direction yet. Copilot Vision could quietly become one of the most useful AI tools Microsoft has ever made. And unlike Clippy, it will only pass judgment if you ask.

You might also like

• I tried Microsoft Copilot's iOS app, and here are 3 reasons why you should too
• I think Microsoft is smart to follow OpenAI in making these premium features free
• Microsoft Paint brushes up on AI

• WeTransfer users were outraged when it seemed an updated terms of service implied their data would be used to train AI models.
• The company moved fast to assure users it does not use uploaded content for AI training
• WeTransfer rewrote the clause in clearer language

Popular file transfer platform WeTransfer spent a frantic day reassuring users that it has no intention of using any uploaded files to train AI models, after an update to its terms of service suggested that anything sent through the platform could be used for making or improving machine learning tools.

The offending language buried in the ToS said that using WeTransfer gave the company the right to use the data "for the purposes of operating, developing, commercializing, and improving the Service or new technologies or services, including to improve performance of machine learning models that enhance our content moderation process, in accordance with the Privacy & Cookie Policy."

That part about machine learning and the general broad nature of the text seemed to suggest that WeTransfer could do whatever it wanted with your data, without any specific safeguards or clarifying qualifiers to alleviate suspicions.

Perhaps understandably, a lot of WeTransfer users, who include many creative professionals, were upset at what this seemed to imply. Many started posting their plans to switch away to other WeTransfer alternatives, and others began warning that people should encrypt files or switch to old-school physical delivery methods.

WeTransfer noted the growing furor around the language and rushed to try and put out the fire. The company rewrote the section of the ToS and shared a blog explaining the confusion, promising repeatedly that no one's data would be used without their permission, especially for AI models.

"From your feedback, we understood that it may have been unclear that you retain ownership and control of your content. We’ve since updated the terms further to make them easier to understand," WeTransfer wrote in the blog. "We’ve also removed the mention of machine learning, as it’s not something WeTransfer uses in connection with customer content and may have caused some apprehension."

While still granting a standard license for improving WeTransfer, the new text omits references to machine learning, focusing instead on the familiar scope needed to run and improve the platform.

Clarified privacy

If this feels a little like deja vu, that's because something very similar happened about a year and a half ago with another file transfer platform, Dropbox. A change to the company's fine print implied that Dropbox was taking content uploaded by users in order to train AI models. Public outcry led to Dropbox apologizing for the confusion and fixing the offending boilerplate.

The fact that it happened again in such a similar fashion is interesting not because of the awkward legal language used by software companies, but because it implies a knee-jerk distrust in these companies to protect your information. Assuming the worst is the default approach when there's uncertainty, and the companies have to make an extra effort to ease those tensions.

Sensitivity from creative professionals to even the appearance of data misuse. In an era where tools like DALL·E, Midjourney, and ChatGPT train on the work of artists, writers, and musicians, the stakes are very real. The lawsuits and boycotts by artists over how their creations are used, not to mention suspicions of corporate data use, make the kinds of reassurances offered by WeTransfer are probably going to be something tech companies will want to have in place early on, lest they face the misplaced wrath of their customers

You might also like

• New judge’s ruling makes OpenAI keeping a record of all your ChatGPT chats one step closer to reality
• We've rounded up the best secure file transfer tools around
• Dropbox VP tells us why AI should be your "proactive partner"

Last week, Perplexity AI launched Comet, an AI web browser ready to take on the likes of Google Chrome and Safari.

Currently, the new Chromium-based web browser is only available to a select few who were either lucky enough to be chosen from Comet’s waitlist or subscribers to the new $200 / AU$300 a month Perplexity Max.

I’ve been lucky enough to get press access to Comet and have been using the AI-powered internet browser for the last few days. Today, I’m going to share my first impressions and give a breakdown of how Comet works.

While I’ve only been using Comet for around 48 hours, I’ve experienced enough to see the potential of the product and get a glimpse into the future of AI-powered web browsers.

From booking flights on your behalf to helping you search the web more efficiently, Comet has given me a new perspective on AI and is showcasing exactly what the future of web browsing could look like.

What is Comet?

Comet is Perplexity AI’s agentic Chrome competitor that mixes the world of AI assistants with web browsing. Simply put, you can access the Perplexity assistant from within Comet at any point and ask it to perform tasks on your behalf.

From booking a restaurant reservation with details you provide, or simply browsing the web to compile information on a topic, Comet is a completely new way of spending time online, and I’ve found my first 48 hours to be incredibly eye-opening.

When I spoke with Perplexity to request access to Comet, a member of the company’s PR told me it would take a week or so to have my first “wow moment” with the web browser. After playing around with Comet, I think that’s an understatement, as my first few hours with the software felt like I was learning how to use the internet for the very first time again.

It doesn’t feel natural to write a request prompt to AI and then continue doing something else, so at first, I found my time with Comet far slower than I had imagined. I was asking the assistant to complete tasks and then monitoring the AI as it proceeded to do what I had asked.

It wasn’t until I started to force myself to move on with my life, leaving Comet to its own devices, that something truly clicked. You see, I could easily do most of the things Comet is capable of doing on my own, and faster at that.

Still, once the workflow of asking it to figure things out so I could focus on other, more pressing tasks started to make sense, I began to see Comet’s true potential.

Comet unlocks the internet

If you’re like me, you probably have a million things you need to do in a day, from booking vet appointments, sorting dinner, ordering prescription medication, and even finding deals to buy new games like the upcoming Donkey Kong Bananza for Nintendo Switch 2.

Tasks like these are always on my mind, clouding my brain from getting the things that matter most, like writing an article for work. Comet solves this issue, allowing you to ask an AI agent to do these mundane tasks for you while you focus on everything else that actually requires your energy.

For example, while I’ve been writing this article, I’ve asked Comet to browse Michelin Guide restaurants for my upcoming trip to Copenhagen based on specific criteria, check hotel availability for my honeymoon next year in Japan, and even get me the best deal on Donkey Kong Bananza which I forgot to preorder before its release later this week.

While Comet has been handling those tasks, I’ve been able to devote my entire concentration to work, unlocking a level of productivity that is often unachievable when working constantly connected to the internet.

Comet is also capable of booking restaurants and other appointments on your behalf, just by asking the assistant and giving some information, like times and dates. It’s worth noting that Comet is still in beta, so I didn’t feel truly comfortable giving the AI my personal details to complete a booking for a London steakhouse next week, as I had requested.

That said, like all things AI, the more information you’re willing to give up, the more Comet can help you navigate the messy and confusing 21st century.

It’s only the beginning

Perplexity hasn’t announced when Comet will be officially launching, or when it will become available to free users. That said, we’re entering a new wave of AI-powered web browsers that is only going to get better with time, with more options.

Just last week, reports broke of OpenAI launching a Chrome competitor similar to Comet, and after using Perplexity AI’s take on the agentic browser, I understand precisely why.

You see, just like Apple’s “AI for the rest of us” mantra from the company’s Apple Intelligence reveal, agentic web browsers feel like the epitome of useful AI, and after embracing Comet, I want to see more products like it.

In an age where we are constantly seeking innovations to transform our lives, it’s become increasingly common for products and software to not live up to their billing. Comet might still have its quirks, like issues related to some prompts crashing when the AI meets a hurdle, but initial impressions leave me yearning for more. I didn’t understand the point of an AI web browser, but after 48 hours with Comet, I’m convinced that having a smart assistant built into your internet experience will transform our lives forever.

You might also like

• Perplexity AI's Comet browser will streak across the web this month
• Perplexity's voice assistant offers a Siri alternative for iPhones
• AI comes to the URL with a new web browser that answers you back

• Google introduces featured notebooks into NotebookLM
• The curated content comes from sites like The Economist and The Atlantic
• You can ask questions about each notebook

Tired of waiting for you to use NotebookLM to make fantastic learning resources of your own, Google has decided to take matters into its own hands and produced a series of carefully curated Notebooks from respected authors, researchers, publications, and nonprofits, including The Economist and The Atlantic.

With NotebookLM, you can read the original source material, but also pose questions to a chatbot that’s versed in the material, so you can explore specific topics in greater depth.

And of course, you can listen to the AI-generated audio overviews, which sound like podcasts, that NotebookLM is famous for, or explore the newer Mind Maps feature.

The initial lineup of curated notebooks includes longevity advice from Eric Topol, bestselling author of Super Agers, expert analysis and predictions for the year 2025 as shared in The World Ahead annual report by The Economist, and an advice notebook based on bestselling author Arthur C. Brooks' How to Build A Life columns in The Atlantic.

As if that wasn’t enough, they’re throwing in The Complete Works of William Shakespeare for anybody who needs help exploring the works of the Bard.

You can expect the list of featured notebooks to grow, too. Google says it will continue to introduce new featured notebooks, including additional collections from its partnerships with The Economist and The Atlantic.

The books of the future

Google says that since introducing the ability to share notebooks last month, ”more than 140,000 public notebooks have been created, on a wide range of topics”.

I find these featured notebooks dangerous myself because each one is a little rabbit hole I can happily disappear down for over an hour.

For example, I opened the How to Build a Life notebook based on Arthur Brooks' columns in The Atlantic and asked it what age was best to buy a house, and I didn’t emerge for another hour!

It's a different experience from reading a book, as you're constantly switching to an interactive way of consuming media, which makes the learning so much more fun.

Commenting on the new featured notebooks Nicholas Thompson, CEO of The Atlantic said, "The books of the future won’t just be static: some will talk to you, some will evolve with you, and some will exist in forms we can’t imagine now. We’re delighted to partner with Google in its pioneering work on this front.”

You might also like

• Is your job safe from AI? A new report reveals one role that's surprisingly at risk
• The next generation of ChatGPT is just around the corner - here’s why GPT-5 could transform the way you use AI
• You don’t have to explain everything to Claude anymore – it’s finally in your apps

• New Microsoft Research paper identifies areas where AI is already being used the most
• It also shows areas that AI has very little influence on currently
• The research could show potential for AI job augmentation, not just replacement

I don’t know about you, but I have this kind of nagging fear that AI is coming for me one of these days. If not imminently, then in the very near future. One thing that might allay that fear is knowing exactly where AI’s axe is going to fall in the labor market, so that I can make sure I’m always just out of its reach.

The problem is that right now we have a lot of people making bold assumptions about what sorts of jobs AI will take away, but as we all know, no plan survives contact with the enemy, so it might be better to approach the problem from another direction.

A new report from Microsoft Research has analyzed 200,000 real conversations between people and Copilot to understand how AI is being used by people in the workplace right now. This way, we can determine which roles are likely to be the most impacted as companies adopt generative AI in the future.

The most at risk

It should come as no surprise that the jobs the report identified as the most common work activities people seek AI assistance for all involve gathering information and writing, and that the most common activities that AI is performing are providing information and assistance, writing, teaching, and advising.

It turns out that interpreters and translators are top of the list when it comes to compatibility with AI, with a stunning 98% of their activities overlapping with frequent Copilot tasks that have fairly high completion rates.

So, if you're thinking of changing careers to become a translator, it might be worth considering your options. Also at the top of the list are historians, writers and authors, and journalists. It should be no surprise to also see proofreaders, editors, and PR specialists high up on the list, too.

The most resistant to AI

It’s physical trades involving working with people that are the most resistant to the influence of AI. The report puts nursing assistants, massage therapists, and machinery operators, including truck and tractor drivers, as the most AI-resistant occupations. Manual laborers like roofers, dishwashers, maids, and housekeeping cleaners were also near the top of the list.

The news will be good for some jobs, but terrible for others. Of course, nothing is guaranteed, and if you’re working in one of the most compatible areas for AI (I know I am!), then don’t panic right now because the research could be simply indicating that your area is one that is ripe for augmentation by AI, rather than replacement.

I think there will always be a need for skilled humans in some capacity, even in areas that will be heavily dominated by AI. That said, understanding AI’s impact on jobs is probably going to put you in a better position than if you have no clue about its threats.

You might also like

• You don’t have to explain everything to Claude anymore – it’s finally in your apps
• The next generation of ChatGPT is just around the corner - here’s why GPT-5 could transform the way you use AI
• I tried asking ChatGPT what my favorite fictional characters say about me – here’s what I learned about myself

• Popeyes released a satirical AI-generated diss track targeting McDonald's.
• The ad is about McDonald's bringing back the Snack Wrap one day after Popeyes launched its own Chicken Wraps.
• The company brought AI filmmaker PJ Ace in to make the ad using Google's Veo 3 model.

Fast food rivalries are nothing new, but AI has given one between Popeyes and McDonald's a new spin. Popeyes took a very public swipe at McDonald’s with a polished, AI-assisted rap video with the help of AI filmmaker PJ Accetturo.

The ad was ostensibly born from the Louisiana chain's outrage over McDonald’s announcement of the return of its long-retired Snack Wraps, one day after Popeyes launched its new Chicken Wraps.

Whether it was coincidence or corporate chess, Popeyes decided to treat it like an act of culinary war as seen in the "Wrap Battle" ad below.

The song’s thesis is that Popeyes has better wraps, better taste, and is far less clownery behind the counter. “Food be tasting funny when the clown be in the kitchen,” as the ad eloquently states.

PJ Accetturo didn't just write a single prompt and walk away. He scripted the campaign, customizing the lyrics with the help of AI music production platform Suno and human helpers to fine-tune the track. For the video itself, he said his team struggled to get enough AI-generated video until they started just using Veo 3, the Google-made AI video generation model that's been attracting a lot of interest since it debuted not long ago.

"We started with image-to-video, but it was too slow and we had less than 3 days when I scrapped all of our work and said we would use only Veo 3," Accetturo wrote on X. "We wanted something that went hard but didn’t take itself seriously. Once we switched to Veo 3, creating a story about people having fun around town with chicken wraps became easy. It all came together when we added the clown bits and the crazy animals at the party."

AI ads rising

The Popeyes “Wrap Battle” diss track is both a one-off stunt and a hint of things to come. A world where commercials can be spun up in a couple of days with a small group of people and some clever prompting could alter how marketing plans its campaigns completely. AI-powered tools that can reduce production times from weeks to hours alter mean speed and adaptability matter even more since you don't have to worry about sinking a month of time and resources into a single ad. They can be as rapid as social media text and image campaigns.

Not that a rapid-fire AI-led approach to advertising is without risks. AI tools can be inconsistent and require humans to babysit and edit them throughout to avoid any tone-deaf or outright wrong ads. But Popeyes’ campaign shows what’s possible when you get the mix right, or at least what you hope resonates with the audience.

I'd expect plenty more AI-enhanced and generated ads to start popping up until they become more of a standard than a novelty. Think of all the personalized jingles, fast-react videos, and the AI-built spokespeople that could go viral. They won't all be this cleverly surreal, but it should at least sometimes be entertaining, which is all you can hope for from a commercial.

You might also like

• Google fumbles Gemini Super Bowl ad's cheese statistic
• Google Gemini is the parent nobody asked for in now-pulled Olympics ad
• Perplexity AI drops new Squid Game-inspired ad that pokes fun at Google starring Lee Jung-jae

• Perplexity’s new Comet browser promises an AI assistant that travels the web for you
• Comet joins other AI browsers aiming to beat Chrome and other popular platforms.
• Though AI companies are betting on their browser's enticing users, their mainstream appeal is uncertain.

Perplexity has officially made the move to web browsers, embedding its AI tools into Comet, its new Chromium browser.

It’s available now, initially only to some subscribers of Perplexity’s $200-a-month Max plan. At first glance, Comet is like most browsers, but Comet has a unique sidebar. You can highlight a word, sentence, or image, and Comet will discuss it with you. You can get a summary of an article, write an RSVP, or organize the itinerary of your next vacation. There's also a privacy benefit. Perplexity says all of the AI processing stays local and that it won’t train its AI using your site visits.

It makes sense. Browsers are central to modern work and life. If AI models can latch onto our browsing flow, then we will be using them all the time. But will it stick?

The browser space has seen a lot of failed Chrome clones, from Yahoo Browser to Internet Explorer 6. Chrome and Safari claim over 90% market share globally. The subscription price alone might be the biggest deterrent. Perplexity’s Comet is $200 a month. Compared to the price of 'free,' it would take a lot more than an occasional paper summary to make people pay up.

And while Comet shows how it might be useful as a way for Perplexity to meld its AI with a web browser, it's hardly alone in pursuing the idea of augmenting web browsers with AI. OpenAI is building a ChatGPT-native Operator browser. Currently, The Browser Company, Opera's Aria, Microsoft's Edge with Copilot, and others are all providing similar services. And Google is continuously adding AI to Chrome, offering quick overviews, summaries, and image explanations.

AI browsing future

Perplexity, OpenAI, and any other contender face the same challenge of getting people to switch. And while it's possible that Perplexity breaks through to a new group of users with the idea of an AI browser, they will then have to face off against each other as well as the AI-enhanced versions of Chrome, Safari, and the others on the rise from Brave to Firefox, each with their own pitch for a better AI, more privacy, or another appealing feature.

Or, maybe the concerns about technical issues and user privacy will keep the AI browser a niche product with limited appeal for hardcore users, a bit like Linux. AI might be the future of browsing, a brief fad, or something in between.

The question of what makes it worth the effort will need to be answered. If Comet and others can streamline the online experience and save us a lot of time, they'll be popping up everywhere, but for now, they’re novelties priced at a premium. We'll have to see if the AI-enhanced browser can find the users it needs to last.

You might also like

• Perplexity AI's Comet browser will streak across the web this month
• Perplexity's voice assistant offers a Siri alternative for iPhones
• AI comes to the URL with a new web browser that answers you back

• xAI has launched the new Grok 4 and Grok 4 Heavy AI models
• Grok 4 requires a $30 monthly SuperGrok subscription
• The launch comes as Grok faces renewed criticism

xAI introduced new versions of its Grok AI model line. Grok 4 and its larger, more powerful sibling, Grok 4 Heavy, are part of CEO Elon Musk’s effort to position Grok as a serious competitor to OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude. That includes the new $300-a-month subscription tier called SuperGrok Heavy, which offers exclusive access to Grok 4 Heavy.

Musk boasted during the announcement livestream that “Grok 4 is better than PhD level in every subject, no exceptions. At times, it may lack common sense, and it has not yet invented new technologies or discovered new physics, but that is just a matter of time.”

And the model’s benchmark scores do suggest it's not hyperbolic to say so; it's a legitimate leap forward. Grok 4 scored 25.4% on the notoriously difficult Humanity’s Last Exam benchmark without tools, putting it ahead of Gemini 2.5 Pro and OpenAI’s o3. The bragging is even more apt for Grok 4 Heavy, because as a multi-agent version of Grok 4, it deploys several reasoning agents simultaneously. On the same test, it scored 44.4%, better than all current commercial offerings.

The takeaway, at least from a technical standpoint, is that Grok 4 is now firmly in frontier-model territory. That’s a meaningful shift for xAI, which just months ago was primarily known for its integration with X, the rechristened Twitter owned by Musk. xAI is clearly trying to be taken seriously as a legitimate AI research and enterprise company.

If you do pay the $300 a month for SuperGrok Heavy, you'll get not only access to Grok 4 Heavy but also developer tools, API usage, and be first to try out new and upcoming features like an AI coding assistant, a multi-modal agent, and an AI video generator. As OpenAI, Google, and Anthropic all roll out more expensive subscription tiers, xAI is likely to be keen to come out ahead in both timing and model quality.

Grokking controversy

Of course, the benchmarks and demos shared by Musk and his team during the livestream could not quite overshadow how Grok’s official account on X this week spiraled into antisemitic madness.

The chatbot’s automated replies on X for hours included conspiracy theories about Jewish control of Hollywood, praise for Hitler, and even declaring itself as “MechaHitler.” The company swiftly deleted the posts as they appeared, and Grok briefly denied even making them before copping to the reality of screenshots.

Eventually, X deleted all of the eye-poppingly offensive posts and placed temporary restrictions on the account. The outburst appeared to be tied to a recent update to Grok’s internal system prompt that the company then reversed.

Musk didn’t address the incident directly during his Grok 4 livestream, nor did anyone at xAI offer a public explanation. Meanwhile, Linda Yaccarino stepped down as CEO of X on the very same day, though xAI insists the timing is unrelated.

With all that happening in the background, Grok 4's launch didn't have quite the clean innovation-centered debut xAI likely hoped for. And it's hard for the company to claim the praise for Hitler was simply a technical error when Musk, who is intimately tied to both X and xAI, has repeatedly insisted that Grok will be a non-politically correct AI model.

You can build the most powerful model in the world, but if users are constantly bracing for it to say something offensive or unhinged, that power won’t matter.

There’s no question xAI has the technical chops to build a top-tier model. But unless they start addressing trust, transparency, and content safety with the same intensity they apply to benchmarks, they’ll always be playing catch-up to companies with AI chatbots that don't remind people of major public relations disasters.

A company interested in what Grok 4 Heavy can do for them might be a little more hesitant to pay $300 a month if the first thing people think of when they hear about Grok powering the system is Holocaust denial. That kind of baggage is heavier than any dataset.

You might also like

• Grok may start remembering everything you ask it to do, according to new reports
• 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

I first met Panos Panay in 2012, around the time Microsoft was trying to break into the productivity tablet space. He was second to Steven Sinofsky, then the face of Windows and any devices in its orbit. When Sinofsky left (or was forced out), Panay took the helm, and brought with him his soon-to-be signature passion.

Panay arguably built the Surface business, and now, not quite two years since he joined Amazon, he's aiming to propel Amazon's devices business to new heights, but mostly on the back of a piece of software (or AI). In a way, the effort to make Alexa+ the foundation of all things Amazon devices is reminiscent of Windows' place in the Microsoft ecosystem: it's the hub on which all things turn (until Copilot, at least).

I've interviewed Panay multiple times over the last dozen years and I'm always struck by his intensity, storytelling skills, warmth, interest in other people's stories, and passion for product. So it's been strange to watch the Amazon Device & Services lead remain so quiet for so long.

Yes, he's offered brief chats and quips here and there. I watched him launch a new line of Kindles last fall (including a color model), and then attended his big Amazon coming-out party at the Alexa+ launch in February. Still, he was not yet available for a sit-down about Amazon, its business and hardware, or the state of Alexa+, which, to this day, is still in early access, and only used by a fraction of Amazon's hundreds of millions of customers (a number that still adds up to millions).

Panay finally agreed to a sit down with me in our offices this week, and we talked about when Alexa+ will be released to all customers (he told me any Prime member can ask for it if they want it), which AI models they use and why they use them, when or if Amazon hardware will be Apple-cool, 'AI Time,' the differences between Microsoft and Amazon, and future hardware products.

On that last bit, I hate to be a spoiler, but no, Panay did not reveal upcoming hardware news; however, he did confirm that he sees a bright future for AR, whatever that means.

Our discussion has been edited for clarity.

When will Alexa+ come to all?

TechRadar

It’s a pleasure to get to talk to you; it’s been a little while.

I just want to cover a few things because, you know, this is a new job for you, but you're in control of a product that everybody's really looking at, interested in, and excited about. So, I’m actually going to start by asking a question that I ask myself now every day because I'm using the product: when is Alexa+ going to go fully public?

Panos

Fully public? It's – pretty soon. We're working up. So, one of the things just to think about is there's a balance of, you know, we have hundreds of millions of customers and we were being cautious...when you think about starting from zero with AI, you know, you have Alexa+, new, conversational, smarter, helps you get things done, helps you control your home, manage your home, manage your list, manage your calendar. All these things are coming in, and when you start from zero, from an AI front, it makes sense.

We have all our partners to bring along and to stay connected to, but when you have hundreds of millions of customers, they're not starting from zero, they're coming from what was Alexa to Alexa+. We’re trying to be very cautious and making sure we really do work out every kink. It's in what we call Early Access right now.

So we're doing it in a very kind of methodical way, taking feedback, and listening to customers. People are pumped about using it. It's fun, and every now and then, I'm like, “Turn it on! Give it to everyone!” and then [I'm] like, “Ah, no, it's okay.”

We want to make sure we bring every customer along that we can. And so, we've got feedback mechanisms for listening. It's pretty fun.

So there’s not an exact date right now, Lance. It comes down to when the product is ready, we open it. Now, keep in mind, also, these are over-the-air updates, meaning we're updating the products in their homes, in people's homes. So then you want to do that very cautiously. Just from the perspective of, you want to break it, right, just don't break it. You want your customer to get an update and have a delightful transition and experience.

You want to make sure you put the customer first. One of my favorite things about Amazon? Customer obsession. What does a customer need? Which makes early access tricky, because we're giving people a product that is… it's awesome, people love using it, but if you’ve been using the same product for 10 years? It's different.

TechRadar

Yes, I can tell you I have been having conversations with it, like actual conversations. I started a conversation where I asked it, “Do you know who Panos Panay is?” Well, it turns out there's a music producer with your name.

Panos

That's my cousin. That's my first cousin.

TechRadar

Yeah, so it was stuck on that, and then I said, “What about a guy at Amazon?” Then it knew what I was talking about. And then from that point forward we actually had – it was strange, because we had a whole conversation about my job, why I knew you, that I was in the tech industry. It got a little weird because I started to realize, “Lance, you're not actually talking to a friend.” But obviously that's so different from the regular Echo, the regular Alexa.

One thing I will say is that it's clearly not fully baked. For example, I showed [Alexa+] a pair of shoes – now, again something I can't do with the old Alexa– and I said, “I want a pair of shoes just like this” and, I thought, this would be something Amazon would be really good at, but what it came back with was a list of shoes that didn’t look like it, including women's shoes. I’m like, look, you know who I am, you know who I'm talking to. This is not quite fully baked.

But I'm going to ask you a question about the caution you’re taking, because you answered a question I was going to ask. So great, I understand why you're being so cautious, but I think we're on AI time, right? The concept of AI Time is that everything goes faster. Are you concerned that the caution you're taking is putting Amazon behind the competition?

Panos

No, because, look, if people want it right now, they can ask for it and get it, and they use it. And it comes with some semblance of understanding that it’s early access. They get in, they're loving it. They're using it more. They're shopping with it more.

Yeah, there's some, a few corner cases where the kinks are being worked out. But ultimately, you see the engagement.

If you think about what we've done, we have this agentic experience that's connecting to thousands of our partners and you're able to do so much.

So while there is an age of AI, and it's moving, it is moving lighting fast, as are we with it, but there's this element of being able to do all the things that Alexa is doing now, whether it's, you know, creating that list, taking it to the store, pulling it out, doing the shopping for you, ask for what you need and get it, manage your calendar, update the document. And Alexa reads it, then updates you back and you can quiz yourself.

These little things that are coming into play and then connect to partners, whether it's ordering food or ordering an Uber. They are – no one's doing that right now. So we have that, and people are engaged with our product in totally new ways, Lance. It's astonishing to watch the usage patterns that are on it.

We feel feel good about that. Now, when do we get it to the hundreds of millions? It's just a process to ramp up. Millions of people have it now. Millions.

So it's not small, it's not the hundreds of millions that's the entire Alexa base worldwide, but we're focused on the US first and we just, we just gotta get there.

Is Alexa+ Amazon's new hub?

TechRadar

You’re training a userbase who spent a decade using Alexa and their Echo devices in a very sort of simple way, and I realize with Alexa+ you are transforming into a real 80/20 product in that 80% of people are only using 20% of the capability of this really rich thing. I'm seeing it all the time as I'm using it. I keep asking, “What can we do together?”

So, I'm curious about, you know, obviously, you spent a lot of time at Microsoft – a long time at Microsoft, and for that place, the world sort of revolved around the platform of Windows, right? So with Amazon, it revolves primarily around commerce, but what I'm curious about is, whether you envision the wheel of Amazon's products and services eventually revolving around the focal point of Alexa+.

Panos

Think about Alexa in your home and being ambient. It's quite powerful that thought, meaning you and I having a conversation, can turn to Alexa and just have a conversation. Don't take your phone, don’t open the app, nothing – we just have a conversation. We can bring her in, help us resolve – by the way, in my house, we're having debates all the time and, you know, I go, “Let's just ask Alexa, let's get through it” And this helps because nobody at the dinner table pulls out their phone and gets distracted and then starts texting, and like, “What are you doing? We're having a conversation.” Instead, there's this magical moment where the family stays a little more connected.

But there's a nuance in there. Think about the services also that are there and ready to use. You can call out any song in the world and play it.

We were having a debate yesterday…. my brother now, we're having a debate on what this latest One Republic song was, and we were trying to get through it. And finally, we just asked Alexa, "Can you remember this song in this movie and play it?" She found it and played it, we're like, okay.

But you have the music service, you have the Prime Video service, you have shopping, they all – to your question – they all hang around this service that’s Alexa and, yes, of course, it's at the center of serving so many different customers what they need, and connected directly into those Amazon properties.

Alexa+ everywhere?

TechRadar

We're just gonna go back to one thing, cause you guys, right now have [Alexa+] working with the screen Echoes, right? Primarily.

Panos

Yep. Echo Show 8, 10, 15, 21.

TechRadar

Right? So, many millions of people have the original Echo – I still own the original Echo. Is there any sense of having a form of Alexa+ working across every device that was sold, or is that not possible?

Panos

No, it's a good question. It'll work in almost every device. You don’t need a screen. Now, the experience is incredible with a screen, like your shopping experience is better, your music experience is better, your video experience – obviously there's no video without a screen. You have photos, you have your family calendar.

And so what we see people engaging with the screen is fantastic. When the emotional photos are flowing through, you can set up a slideshow with your voice, or you can send that photo to your father, which I love to do as I'm so close to my dad.

There's so many different things you can do with the screen. However, you don't need a screen, and it will work on most generations of Echoes.

Now, there's some very early generations, like it won’t work on all of them. But almost all of them and fundamentally, and we have that list, and it's broken down, and I can send it to you but you don't need a screen. If you have Echoes and you are a Prime Member, you're gonna get Alexa+ and you can be conversational, you can set up your routines. It's powerful in all stances, screen or no screen, but you know, with a screen, I think it's a better experience.

TechRadar

[At this point, I started asking Panay if Alexa+ would in fact support the very first Echo device, one I jokingly called "The Pringles Can" because that's what, shape-wise, it resembled. He laughed at my description.]

Panos

I’ll have to check. I like that you call it a Pringles can.

TechRadar

Let’s switch gears a little bit and talk about hardware. As I see the sort of world of Amazon hardware, there's some stuff that's kind of well-known, you have the Fire TVs, the Kindles, the Sticks, and then you have things like all the Essentials you can buy, the switches and all that.

Panos

We also have Ring, Blink, and Eero.

TechRadar

Right, but they’re sub-brands or brands that some people don't even realize they're directly connected to Amazon. Amazon hardware doesn't necessarily have the cachet of, say, an Apple. Right? Like some of the [Apple] hardware, people get all excited about the next phone or whatever. Is there any intention to sort of elevate the Amazon hardware brand to that level of excitement?

In a way, this connects a little bit to the conversations we had back in your days at Microsoft, where you almost single-handedly elevated a portion of Microsoft's hardware business. I'm just curious about what your intention might be here.

Panos

Okay, so we have had these conversations in the past where I'll never tell you what the future products is… you’re a master at trying to get there. So, that’s the next question. The answer is, I can't tell you. Look, at the end of the day, I think we have something magical coming together with these products.

And literally the way they're interoperating, the fact that we have Alexa as just as an AI thread through all of them, how they connect, and yeah, I think also in elevating what people, you know, something that they buy that they covet, that's precious to them and how important that is, like these are objects, and they matter, you know, buying something that you love and that you want to use.

That's a combination of both the aesthetics, the fit and finish materials, how well it works, the quality of the product, but also how well it works with the other properties in your house, with the other objects in your house that are so important for you.

I think you'll see it a little elevated. You have a very motivated team right now to deliver Alexa+ throughout everything that we're building and then as you watch these devices connect. Like if you're using your Ring cameras now, you can use them basically as a virtual security guard to your home, just talk to Alexa, “Show me what's happening outside, did the dog go out?”

TechRadar

I’ve done it, “How many times have dogs been peeing on my lawn?” And it showed me.

Panos

Yeah, it's pretty powerful. You know, be careful what you ask for. But I think that it's both the essence of the experience that is [it] can be romantic and important and kind of yeah give the customer what they're looking for in me that I think is beautiful. It goes to the experience and also the object itself.

So sure. I mean, you'll see the elevation of the hardware is the answer, but you'll also see products up and down the line, great products at the top of the line and great products at the entry-level, and I think it’s just serving all our customers, which is so important. That's what I want. I want customers to have a choice, and if they have a choice, they can buy the product that they love, but they all have to work perfectly – and that’s the vision.

TechRadar

So obviously you've already told me that you’re not gonna tell me things like this, but there is a product that exists, it's called the Echo Frames. I've used them. They’re like a perfect platform for Alexa+. One, does [Alexa+] work with it?

Panos

I’ll tell you an awesome story. I had a buddy who was golfing this weekend – his story, not mine, but pretty funny – and he has Alexa+ enabled, and so, kind of fun fact, you get Alexa+ enabled, it's enabled on all your devices, and he has it enabled on Frames.

As he was golfing, he was explaining to Alexa, “I'm standing on a side hill, I don't know what to do. I'm terrible at the shot, what should I expect?” And Alexa, just like that same conversation you were having earlier, it was just breaking it down, like, “Here's what you got to do which, you know, lean to your left side, take it ¾ of the way back,” and this guy, by the way, is a terrible golfer.

The shot, I'm sure, was terrible.

But the answer is, yeah, yeah, it’ll work. So, Alexa says going to carry through most products.

Wearables and the AR space

TechRadar

Well, let me ask you about this. How do you feel about AR?

Panos

How do I feel about AR? Uh, you know, There's a lot, there's a lot there.

It can serve customers in very unique ways. But AR means – the reason I pause, is because it means so many different things, and so maybe it means something very unique to you and clear to you, but…

TechRadar

You just gave me an anecdote where, in my head, immediately, I imagined the lie, the ball, and looking at it and seeing an overlay of, “There's your angle.”

Panos

Remember, I worked on a product for years that was AR [Microsoft Hololens], and fundamentally, at the end of the day, there is incredible technology coming down the pipe for it. So, do I believe in it? I think it's got legs.

TechRadar

Well, that's something. A couple of things – and you may not be able to answer – but you had what I consider a storied history at Microsoft, so what do you feel are the core differences between the Microsoft environment and where you are now at Amazon?

Panos

At Amazon, the culture is phenomenal. The LP’s [leadership principles], if you haven't read them, are something I always encourage folks [to study]. They're so core to everything that we do, and when you, and if you live by those, and for me, they just align with who I am as a person.

So, what I found is, you know, when you have that alignment, your strides are smoother, they're just, they're cleaner when you think about customer obsession, you think about biassing for action when you think about diving deep, these are the things that just make for great product making and then supporting, if you will, the customers you are serving.

At Microsoft, same – you know, great culture, great company, great leaders and, you know, 20 years, very proud to have been there, but the transformation of AI right now and what we can bring together at Amazon, it's like a, it's a dream come true in the sense of when you think about the products that have been mentioned earlier all the way through Kuiper satellites. Just the opportunity to change the world by putting products in people's hands. It's just, I gotta tell you, it lights me up.

TechRadar

You guys said Amazon had developed this model Nova, and then you're also using Anthropic. Is there any possibility that one: you would stick with all of your own models? And two: you would bring in yet an additional third party, like something from OpenAI?

Panos

We always say we're going to pick the best model for the customer, like what is it that serves a customer best? But at the end of the day, the majority of the product is running on Nova.

TechRadar

Is there a way that I would know the difference? That whole conversation I was having with Alexa+, do you attribute that to Nova or to…

Panos

Well, it's a little more complicated. It’s nuanced.

It's probably not a way that you would know, but ultimately that was likely a Nova conversation or conversational aspects...

TechRadar

One more question: The rest of 2025, should we expect anything big?

Panos

I think so, I think so. First off, we're going to release Alexa+ to everyone, and we're pretty pumped about it, and then you’ll, of course, see some new products coming down the pipe that might, I think, might inspire you.

TechRadar

That's as close as I'm gonna get to hearing the future…

Panos

I certainly am not gonna share what those are, but I gotta tell you we're working on some really – by the way, the team that's working on these products has so much passion, what they believe in, what they're bringing with Alexa, what they're bringing with Ring and Blink, what they’re bringing with Kindle and Eero.

We have the Kuiper satellites and Zoox [robo taxi]. So, like these are teams that are passionate about delivering, and, at the end of the day, as AI connects it all, it's pretty inspiring.

You might also like

• I've spent a week with Alexa+ early access – and this could ...
• All your burning Alexa+ questions answered – by one of ...
• What is Alexa+: Amazon's next-generation assistant is ...