Sizzle. Sizzle. That's the sound of your neurons frying over the heat of a thousand GPUs as your generative AI tool of choice cheerfully churns through your workload. As it turns out, offloading all of that cognitive effort to a robot as you look on in luxury is turning your brain into a couch potato.

That's what a recently published (and yet to be peer-reviewed) paper from some of MIT's brightest minds suggests, anyway.

The study examines the "neural and behavioral consequences" of using LLMs (Large Language Models) like ChatGPT for, in this instance, essay writing. The findings raise serious questions about how long-term use of AI might affect learning, thinking, and memory. More worryingly, we recently witnessed it play out in real life.

Google DeepMind, you EmptyMind

The study, titled, "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task," involved 54 participants split into three groups:

• LLM group: Instructed to complete assignments using only ChatGPT, and no other websites or tools.
• Search engine group: Allowed to use any website except LLMs, even AI-enhanced answers were forbidden.
• Brain-only group: Relying only on their own knowledge.

Across three sessions, these groups were tasked with writing an essay about one of three changing topics. An example of the essay question for the topic of "Art" was: "Do works of art have the power to change people's lives?"

Participants then had 20 minutes to answer the question related to their chosen topic in essay form, all while wearing an Enobio headset to collect EEG signals from their brain.

In a fourth session, LLM and Brain-only groups were swapped to measure any potential lasting impact of prior sessions.

The results? Across the first three tests, Brain-only writers had the most active, widespread brain engagement during the task, while LLM-assisted writers showed the lowest levels of brain activity across the board (although routinely completed the task fastest). Search engine-assisted users generally fell somewhere in between the two.

In short, Brain-only writers were actively engaging with the assignment, producing more creative and unique writing while actually learning. They were able to quote their essays afterwards and felt strong ownership of their work.

Alternatively, LLM users engaged less over each session, began to uncritically rely on ChatGPT more as the study went on, and felt less ownership of the results. Their work was judged to be less unique, and participants often failed to accurately quote from their own work, suggesting reduced long-term memory formation.

Researchers referred to this phenomenon as "metacognitive laziness" — not just a great name for a Prog-Rock band, but also a perfect label for the hazy distance between autopilot and Copilot, where participants disengage and let the AI do the thinking for them.

But it was the fourth session that yielded the most worrying results. According to the study, when the LLM and Brain-only group traded places, the group that previously relied on AI failed to bounce back to pre-LLM levels tested before the study.

TL;DR: AI makes us stupid, but we didn't need a study to prove it

To put it simply, sustained use of AI tools like ChatGPT to "help" with tasks that require critical thinking, creativity, and cognitive engagement may erode our natural ability to access those processes in the future.

But we didn't need a 206-page study to tell us that.

On June 10, an outage lasting over 10 hours saw ChatGPT users cut off from their AI assistant, and it provoked a disturbing trend of people openly admitting, sans any hint of awareness, that without access to OpenAI's chatbot, they'd suddenly forgotten how to work, write, or function.

This study may have used EEG caps and grading algorithms to prove it, but most of us may already be living its findings.

When faced with an easy or hard path, many of us would assume that only a particularly smooth-brained individual would willingly take the more difficult, obtuse route.

However, as this study claims, the so-called easy path may be quietly sanding down our frontal lobes in a lasting manner — at least when it comes to our use of AI.

That's especially frightening when you think of students, who are adopting these tools en masse, with OpenAI itself pushing for wider embrace of ChatGPT in education as part of its mission to build "an AI-Ready Workforce."

A 2023 study conducted by Intelligent.com revealed that a third of U.S. college students surveyed used ChatGPT for schoolwork during the 2022/23 academic year.

In 2024, a survey from the Digital Education Council claimed that 86% of students across 16 countries use artificial intelligence in their studies to some degree.

AI's big sell is productivity, the promise that we can get more done, faster. And yes, MIT researchers have previously concluded that AI tools can boost worker productivity by up to 15%, but the long-term impact suggests codependency over competency. And that sounds a lot like regression.

At least for the one in front of the computer.

Sizzle. Sizzle.

More from Laptop Mag

• Is Microsoft misleading users about Copilot? New claims point the finger at AI productivity
• Why OpenAI engineers are turning down $100 million from Meta, according to Sam Altman
• Google's latest Gemini 2.5 models are its biggest response to ChatGPT yet — and they're already live

On July 7, Google’s Gemini assistant will be able to send a WhatsApp message, place a call, adjust your volume, or set a timer, even if you’ve told it not to track what you’re doing. And no, it’s not something you need to opt into, nor can you opt out of it; it’s the new norm.

That’s because Google is quietly rewriting the rules of how assistant AI works on your phone. If you thought turning off Gemini Apps Activity meant turning off Gemini’s reach into your personal life, think again. Gemini is going live in your apps whether you like it or not.

This is according to a notification email seen by Android Police, where Google says that it will roll out an update that allows Gemini to “use Phone, Messages, WhatsApp, and Utilities on your phone, whether your Gemini Apps Activity is on or off.”

What’s actually changing?

Here’s what matters: Before now, if you disabled Gemini Apps Activity, the setting that governs what Gemini remembers and uses to train future AI models, you were cut off from deeper integrations. Gemini couldn’t place calls or send texts because it wasn’t allowed to access that data.

That barrier is about to vanish. Starting July 7, Gemini will gain access to four key phone services: Phone, Messages, WhatsApp, and Utilities (volume, flashlight, alarm, and media controls), regardless of whether you’ve turned activity tracking on.

This means Gemini will now handle everyday tasks that used to be blocked for privacy-conscious users. Want to say “text Alex on WhatsApp that I’m running late” with Apps Activity off? You’ll be able to. But it comes with a new kind of data compromise.

Google’s workaround: Access without memory

Google won’t save your assistant requests to your Gemini Activity page or use them to train its models if Apps Activity is disabled. But it will keep the data for up to 72 hours. Why? Google says this temporary window allows for “service reliability and safety checks,” which is corporate-speak for making sure things don’t break or misfire.

That data — the messages, call metadata, and voice prompts — lives in a kind of gray zone. You won’t see it in your history. You won’t be notified that it exists. And you can’t manually delete it before the 72-hour expiration.

If that feels a bit like having your privacy settings overruled, you’re not alone. What Gemini’s doing here is technically compliant with your preferences, but only because Google has quietly redefined what “off” means.

Can you stop it?

Yes and no. Gemini now has a second set of toggles buried in its settings menu. You can go to Profile → Apps inside the Gemini app and manually disable access to Phone, Messages, WhatsApp, or Utilities. But these are turned on by default, even if you’ve disabled tracking in the past.

There’s no upfront prompt. No opt-in. And unless you knew to look for them, or happened to read a very specific footnote in Google’s support pages, you’d never know those integrations were live. So, unless you go digging, Gemini now has deeper control of your device than the Assistant it’s replacing, and it’s not something you authorized.

More than an assistant

Sure, it’s easy to see this as a minor change, but it’s part of something much larger. Gemini isn’t just a voice assistant anymore. It’s being positioned as an “agent,” capable of understanding complex requests and acting across multiple apps. That’s a huge leap from the Google Assistant days of “Hey Google, set a timer.”

With cross-app prompts and generative suggestions rolling out across Android, Gemini is designed to handle multi-step tasks like “Find my next meeting and message the team I’ll be late.” The catch is that it needs deep integration into your personal apps to do that, and now it has it… even if you said no.

It’s not difficult to understand why this matters. Privacy isn’t just about what’s stored; it’s about what’s accessible. And Gemini is operating on a new model: Access-first, memory-later.

That might sound like a good compromise. After all, Google isn’t storing your requests forever or using them to train future models unless you explicitly allow it. But it also blurs the line between private and impermanent. Gemini doesn’t need to remember to function like it does. It just needs a few hours of access and your silence.

And that’s the problem. Because if you didn’t know about this, or didn’t check your app settings, you’re in a system that’s quietly opted you in, with no log, no prompt, and no accountability. Some may call that dystopian.

More from Laptop Mag

• Sony is fighting like cats and dogs against a small game developer from Hong Kong
• A second Windows 11 bug takes suspicious aim at Google Chrome — and the same Microsoft app is to blame
• The Asus ROG Strix G18 is proof that paying more for an RTX 5090 is a waste of money

In multiple interviews following the WWDC 2025 on June 9, Apple executives made it clear that the company has no intention of creating an Apple Intelligence-powered chatbot at this time to compete with ChatGPT, Claude, Google AI, and other similar platforms.

Craig Federighi, Apple's senior vice president of Software Engineering, and Greg Joswiak, senior vice president of Worldwide Marketing, reiterated this message in interviews with Tom's Guide, The Wall Street Journal, and other outlets.

While they readily admitted to the popularity and utility of these chatbots and left open the possibility of it happening someday, it's not the AI experience that Apple is currently looking to deliver to its customers.

That's fine for Apple, but an upgrade to Shortcuts in the developer beta for iOS 26 lets you ignore how Apple feels about making its own chatbot, and in seconds, you can have an Apple Intelligence chatbot up and running on your devices.

Here's a quick look at how to do it if you're running the iOS 26 developer beta. If you aren't eligible to install the developer beta, bookmark this page for when the first public beta arrives in July.

How to use Apple Intelligence as a chatbot

1. Open Shortcuts.
2. Tap the plus button in the upper-right corner.
3. Search for "Use Model" in the "Search Actions" bar
4. Select "On This iPhone" or "Private Cloud Compute" from the listed Models.
5. Add a Text element where you can ask your question or request something from the chatbot.
6. Toggle the Follow Up option if you want to be able to ask additional questions.
7. Select whether you want the Output from the chatbot to come as a notification or text saved to Notes, or to the clipboard.

That's all there is to it. Now, obviously, this isn't as smooth an experience as the standalone chatbot apps available, but it's interesting to experiment with how Apple Intelligence performs at this task.

Late but great?

Apple’s decision to skip the chatbot hype cycle may feel like a miss in the short term, especially as competitors like OpenAI, Google, and Anthropic flood the market with conversational AI personalities. But that’s not unfamiliar territory for Apple.

Historically, the company has let others go first — and often falter — while it refines its own vision for how a new technology should work.

Remember when Apple launched the iPad without a stylus, dismissing it altogether? Steve Jobs famously said, "If you see a stylus, they blew it." Years later, the Apple Pencil arrived, not as a tool critical to basic use of the iPad, but as an accessory that could augment specific use cases.

The same could be said of tablets. While Microsoft and others had early tablet PCs on the market for years, it wasn’t until Apple introduced the iPad that the category truly took off.

Even the Apple Watch came long after smartwatches from Microsoft, Pebble, Sony, Motorola, Samsung, and more. However, within a couple of years, it became the world's best-selling wearable device.

A similar pattern could be unfolding with generative AI. While competitors race to win the chatbot popularity contest, Apple is taking a quieter approach, threading AI into the fabric of iOS, iPadOS, and macOS.

From semantic photo search and inline writing tools to real-time translation, Apple Intelligence isn’t a destination — it’s a layer. There’s no standalone Apple chatbot because the company doesn’t want users to stop what they’re doing to engage with AI. Apple wants AI to support what its users are already doing.

Of course, the danger in this approach is that Apple risks letting others define the conversation about AI. That can shape public perception in a way that’s hard to unwind. Apple doesn’t want to be seen as behind, even if history shows that being second (or third, or fifth) can work out just fine.

This concern isn't helped by Apple's recent demonstration of how this can go wrong with the Apple Vision Pro. While there's plenty of time for Apple to crack AR and spatial computing, it's undeniable that the current Apple Vision Pro will not be that product.

So will Apple eventually release its own systemwide chatbot? Probably. But when it does, it will likely be less about keeping up with ChatGPT and more about delivering something that makes sense within Apple’s ecosystem of privacy, design, and usability.

For now, Apple is playing its usual long game. It will deflect complaints about being slow or late and focus on being the best. Eventually.

More from Laptop Mag

• Meta Quest 4: Everything we've heard about Meta’s upcoming VR/AR headset
• I used the Switch 2 and Switch OLED side by side for a week — here’s what surprised me
• 7 best early Amazon Prime Day iPad deals you don't have to wait to get

Gemini, Google's family of AI models, are often seen as playing second fiddle to OpenAI's GPT family, which powers ChatGPT — but they've come a long way since a bumpy launch, with Google's AI proving itself to be a great fit for those looking for anything from a speedy research aide to a solid smartphone assistant.

Unlike ChatGPT, Gemini is designed to run across a multitude of services, be it drafting messages in Gmail, transcribing calls in Google Meet, or acting as a second set of eyes and ears as the AI assistant for future Android XR smart glasses.

Now, Google has announced it is rolling out its latest upgrades to the models that make Gemini so powerful, pushing preview models like Gemini 2.5 Pro and Flash out of preview and into general availability — and making a new Flash-Lite model open for testing.

Here's what's new, and, more importantly, what it means for you.

Gemini 2.5: What's new, minus the jargon

If you're new to Gemini, or you're just a casual AI user, then talk of models, tokens, and benchmarks is likely to cause more confusion than clarity. So let's keep it simple: Google now offers three updated Gemini 2.5 models designed to tackle different kinds of tasks; Flash, Pro, and a newly introduced Flash-Lite.

• Flash is your everyday option. Think of it like OpenAI's GPT-4o, the model that powers most ChatGPT interactions. It's good for chat, but it can also help with brainstorming, writing, and even translating. It's the all-around model that strikes a balance between brain power and speed.
  
• Pro is, as the name might imply, a much more in-depth model, suited to taking on more demanding, professional tasks like coding, problem solving, data analysis, or more complex, multi-step prompts. It's slower in output, but it is Google's most intelligent model to date.
  
• Flash-Lite is the newest model to get an upgrade, now available for testing in Google AI Studio and Vertex AI, and it's best suited for developers and enterprises looking for a fast way to clear high-volume workloads that don't require too much power to solve.

For those already familiar with Google's models, the new Gemini 2.5 family is better at memory, reasoning, and responses, with improved performance and accuracy across the board.

All three models support multimodal inputs (text, images, and audio), tool use (like code execution), and have a massive 1 million token context window — ideal for handling large documents or specific and detailed prompts.

Both Gemini 2.5 Flash and 2.5 Pro also feature improved native audio output for more natural-sounding conversations, and enhanced reasoning with Deep Think.

For developers, Gemini 2.5 Flash now features rebalanced pricing, with slightly more expensive input tokens being met with cheaper output tokens.

Gemini 2.5 Flash and 2.5 Pro can now be accessed freely on the Gemini web portal and the Gemini mobile app by selecting your model of choice from the drop-down menu at the top of the conversation window.

More from Laptop Mag

• My favorite AI tool just hit Google Search, and it's actually useful — try it yourself
• A simple mistake in the Meta AI app could expose your deepest secrets
• ChatGPT may be the smartest software ever, but this Pong-era game console can do something it can't

In the earlier days of the internet, the World Wide Web was awash with ways to wander its wares, but now we're all mostly content to rely on Google as our go-to guide for getting things done online.

It's easy, it's practical, and, most importantly, it works. But trawling through Search results to find the answer you want (presented in a way that's easy to digest) can sometimes be a nightmare — and we don't all have the spare time to risk falling down a rabbit hole of distractions that leads you from "What's a good idea for a casserole?" to "Is the moon hollow, and how worried should that make me?"

So what if there was a way to quickly do all of that, sans the monotony of manually dragging your eyeballs through countless web pages? Well, there is.

It's called Audio Overviews, and it's a new podcast-like feature found in Google Searches that can help explain complex topics, provide an engaging way to learn new things, and let you do other things while Google Gemini translates thousands of Search results into a bite-sized, custom podcast tailored to your questions.

If you've ever found yourself staring blankly at a science paper titled Spectral Dispersion and Atmospheric Optics after Googling your 3 a.m. thoughts of "Why isn't pink found in rainbows?" and desperately wishing there was an episode of HowStuffWorks you could fall back on for an answer, Audio Overviews is your answer, and you can start using it right now in Google Labs.

Audio Overviews: The best NotebookLM feature heads to Google Search

Since its launch in 1998, there have been several major changes to improve the Google Search experience, but it's largely stuck to the same process: you type in your search request, hit search, and browse through the results.

Audio overviews bring something fresh to that experience, saving you the effort of trawling through results by using generative AI to compile a ton of useful information into a podcast-like audio snippet presented by two virtual hosts.

It's a feature ripped straight from the playbook of my favorite AI tool, NotebookLM — another Google service that makes for an excellent research aide, able to compile multiple sources of information into one easy-to-absorb document, or a similar AI-generated podcast snippet.

It's thanks to NotebookLM that I was ever able to write about complex topics like Microsoft's Majora 1 quantum processor, with its inner workings being so alien and confusing to me that it stands as the only time in life that I've gained knowledge and felt ultimately more stupid afterwards.

NotebookLM's Audio Overview feature gave me a crash course in quantum computing, helped me decipher the qubit, and ultimately served as an example of how to better deliver a complex topic to a curious mind.

Better still, it did it through the casual conversation of two impressively engaging virtual podcast hosts, effortlessly dropping knowledge bombs and explainers in the background while I went about my workday as normal.

Since then, I've been using NotebookLM to help me do the heavy lifting on topics I just don't have time in the day to fully dive into, and it's become an invaluable tool in my quest to convince strangers that I'm smart enough to have finished high school.

Now, with Audio Overviews arriving on Google Search, even more of us can reap the benefits of Google's AI-generated learning aides, and all become a little better off for it, if not in smarts, at least in time spent searching for answers.

How to use Audio Overviews in Google Search

You can test Audio Overviews in Google Search today by opting in to the experiment through Google Labs. Note, this feature is still in testing, and access is limited to Google Account owners in the United States.

More from Laptop Mag

• “It’s not good”: Arm boss warns U.S. export rules on AI chips could backfire
• AI slop ended up on Nvidia and CDC pages. All it took was some neglected URLs and a little SEO magic
• Apple’s AI-powered Siri reportedly has a new target date. Will it stick this time?

There's a strange phenomenon happening on what appear to be legitimate websites from major companies like Nvidia and federal agencies such as the U.S. Centers for Disease Control (CDC) that are hosting AI slop.

It appears that a spam marketing operation hijacked abandoned subdomains from companies and organizations, including Nvidia, the CDC, American Council on Education, Stanford, and NPR, according to a report from 404 Media. The AI slop found on these sites ranged from a hub for vaccines explaining what jets were flown in the movie Top Gun to Brazilian fart games on an Nvidia event page and just a lot of porn.

The subdomains, such as events.nsv.nvidia.com, which were archived by 404 Media, have long since been abandoned by the respective organizations. However, as these domains were connected to legitimate entities, having content about “adult video games” on the site could make it a top result on Google, leading to traffic. It's likely this traffic was somehow monetized by the organization behind this hijacking.

These sites have since been taken down by the respective entities or reverted back to their original form. As of the writing of this article, the website these domains were redirecting to, stocks.wowlazy.com, is no longer up. A summary at URLScan shows the site is seemingly still active and provides a screenshot, which appears to be AI-generated content.

What is AI slop?

AI slop is low-quality content generated by artificial intelligence. Not that long ago, this was mainly written articles filled with certain keywords that would ideally rank high in certain Google searches. Thanks to more advanced AI models such as Midjourney and Sora, now this slop can include AI images and video.

Just this past week, editors at Wikipedia revolted over the site's plan to test AI content. The idea was to test AI-generated summaries, called "simple summaries," that would be machine-generated to take existing Wikipedia articles and simplify them for readers. The human editors for the site claimed these AI summaries would ruin the site's reputation. While the Wikimedia Foundation says the rollout of this test is not happening, it does appear that it's not completely canceled and could be tried again with more input from the human editors.

Social media is a breeding ground for AI slop. Last year, Facebook was seemingly taken over by this content, with the most prominent example being the Jesus Christ shrimp image.

AI slop has made its way to all other platforms, and it will continue to become harder to distinguish the real content from AI-generated content. At its I/O event last month, Google released its AI-media tool, Veo 3, which was able to produce completely realistic video from text prompts. What followed soon after was a viral clip showcasing the completely realistic content generated by Veo 3, which was hard to identify as fake, except for the fake people in the video saying it was.

In one of the clearest examples of how everyday users of technology sometimes don’t understand it – and the developers of that technology haven't fixed it – the Meta AI app is allowing deeply personal, private, and altogether embarrassing content to be shared to its “Discover” tab. And those personal questions can spread to Facebook and Instagram.

This is precisely the worst place for questions like “Tips for asking an Asian girl about dating older men” and “Can my wife see [the questions I ask]?"

Let’s briefly back up: The Meta AI app launched in April. It uses Meta’s own large language model, Llama, which acts as a chatbot for users. In that way, it’s like ChatGPT or the Google Gemini app. You ask it a question, and it provides answers.

But the truly cursed part of this app is that those questions can somehow become public to other users, and clearly, not enough users realize it.

Why the Meta AI app is confusing users so much that they put their private questions into a public feed

The Meta AI app looks a lot like the Instagram app. There’s a Home tab, a Notifications tab, a History tab, and a Discover tab. The Discover tab is a lot like the Explore tab on Instagram – it’s a stream of content produced by other users of the Meta AI app. A lot of it is "AI slop" -- low-quality, AI-generated images of like, Cookie Monster in handcuffs or Pennywise the evil clown as a baby. And between the slop are questions the users very likely didn't intend for anyone else to see.

Why does this happen? There’s a confusing button in the top right of your chats with the Meta AI app called “Share.” Do not press if you are asking a private question.

That “Share” button essentially turns your question into a piece of content -- a "public prompt" -- posted to the Discover feed. The instructions are vague, and you might think hitting “post” wouldn’t result in your content appearing in the Discover feed. There's a one-time warning, but that's it.

You might also have just made a mistake. After being notified that hitting the “post” button would put your content in the Discover feed one time, you are no longer met with the warning on subsequent shares.

Even worse, these public prompts could show up on Facebook and Instagram as "suggested" prompts for other users.

How to stop the Meta AI app from posting your questions to the Discover feed or on Instagram or Facebook

Thankfully, you can delete public prompts in the Meta AI app. You click on your user icon in the upper right corner, click “view profile,” and you’ll see those prompts. You can long-press on each and be met with the option to delete it.

You can also adjust the settings to avoid this sort of thing, which we can't stress enough, shouldn't have happened in the first place, given the history of digital design innovation and research in the last few decades.

We digress: Go to "Data and privacy," click "Manage your Information," and then select "Make all public prompts visible to only you." Click "Apply to all."

Next, go back to the "Data and privacy" menu, click on "Suggesting your prompts on other apps," and uncheck Facebook and Instagram.

Facebook has more than 3 billion users, and the Meta AI app is available in 22 countries. When designing content for oceans of people, especially when your app clearly states that any questions you ask are private, why add a “social” element?

The answer is obvious if you’re Meta, a publicly traded company that relies on engagement and growth numbers to continually move up.

The Meta AI app: Bad design or bad thinking?

A thriving Discover section of the app – and which pieces of content a user clicks on in that app – is another data source about user interests, which Facebook can bundle and sell to advertisers. It’s just like Instagram’s Explore tab.

Is Meta AI's obtuse UI to blame, or is the vagueness intentional to get more content in the Discover feed? It could be both. It feels like a situation where Hanlon's razor could be applied: "Never attribute to malice that which is adequately explained by stupidity.”

More from Laptop Mag

• "Buckle up": Meta's Mark Zuckerberg warns of "intense" year filled with A
• In OpenAI, Google, and Meta's AI arms race, the real loser in 2024 was privacy
• Meta looks to bring a much-desired ChatGPT feature to Meta AI

Technology. The ever-advancing pinnacle of the processor. It's a modern-day blessing to most, but dreaded by those convinced it's five minutes away from flipping society onto its head like an egg served sunny-side down.

The word alone can strike fear into the hearts of man. Especially if you shout it loud enough into their ears as you pass them on the street.

And, if you were to take a small break from harassing the general public and ask one of them to name today's most advanced piece of technology, they'd likely say ChatGPT — OpenAI's super-brainy chatbot, packed with enough artificial intelligence to seemingly make regular intelligence look like its eating glue from a pot in a sandbox.

But if ChatGPT, the poster child for cutting-edge technology, is so smart, how did it just get absolutely bodied by a video game console released in 1977?

ChatGPT vs. Atari 2600: All hail the golden oldie?

When we think of competitors to ChatGPT, there's a usual list of suspects to choose from: Google Gemini, Microsoft Copilot, Meta AI, Anthropic's Claude, Perplexity AI, and maybe even newcomer DeepSeek.

What you wouldn't expect to appear in that list is the Atari 2600, a 48-year-old home video game console best known for bringing Pac-Man into the living rooms of millions of first-generation gamers.

However, thanks to Citrix Engineer Robert Jr. Caruso, Atari's retro console can now be counted among ChatGPT's truest rivals, after it was used to repeatedly best OpenAI's GPT-4o model at a simple game of chess.

In a now-viral post shared to LinkedIn, Caruso details his 90-minute experiment in pitting the computing might of tens of thousands of Nvidia GPUs against the singular 1.19 MHz 8-bit MOS Technology 6507 processor of the Atari 2600, claiming ChatGPT "made enough blunders to get laughed out of a 3rd grade chess club."

A short lived victory for retro tech

Clearly, they don't make them like they used to. The prevalence of the 2600 over ChatGPT is a true David vs. Goliath battle on a checkerboard stage, and a surprising outcome to most.

However, there's a chasm of difference between chatbot and a chess engine, meaning the Atari 2600, which ran not as hardware but through the Stella emulator, likely had ChatGPT's number from the start. While OpenAI has made great strides in improving its model's memory capabilities, it's still primarily a language prediction machine, and not the next Deep Blue.

Still, given the 2600's ability to only predict two moves in advance, it does highlight ChatGPT's shortcomings, and provide a thumb to the virtual eye for OpenAI's world's most intelligent chatbot.

At least it would do if OpenAI had ever claimed as much. In fact, it's more often than not proclaiming the opposite.

Stretching back to 2023, in an episode of the Lex Friedman podcast, OpenAI CEO Sam Altman was quick to label GPT-4 as "a very early AI. It's slow, it is buggy, and it does not do a lot of things very well." Then, in 2024, during a Q&A at Stanford University, Altman claimed that ChatGPT was running on "the dumbest model any of you will ever have to use again by a lot."

In fact, only recently has Altman pushed the message that AI is living up to its supposed smarts. In a blog post published on Tuesday, titled The Gentle Singularity, Altman predicts: "We do not know how far beyond human-level intelligence we can go, but we are about to find out."

So yes, Atari's classic console may have bested ChatGPT this time, but if Altman's words are anything to go by, it could be a very short-lived victory.

More from Laptop Mag

• Sam Altman's vision is the stuff of my nightmares: "A significant fraction of the power on Earth should be spent running AI"
• Wikipedia editors revolt over "truly ghastly" plan for AI slop — they're winning (for now)
• Microsoft has created an after-the-fact AI gaffer -- but is your laptop up to snuff?

A new macOS update is coming soon, but is it worth your time?

Sure, you do want to keep your system relatively up to date to maintain proper app support and get all the latest security patches. But there's a difference between jumping on a new OS update immediately and getting to it a few weeks or months later.

During Monday's Worldwide Developer Conference, Apple shared details on its upcoming plans for the iPhone, Mac, and iPad platforms and some updates to the Apple App ecosystem.

As part of the Mac announcements, Apple detailed all of the changes coming with macOS Tahoe 26, which include an expanded Apple Intelligence portfolio, expanded Continuity features, a desktop facelift, and a revamped Spotlight search.

But are these updates enough? Let's take a look.

Expanded Apple Intelligence

Apple's proprietary AI features, called Apple Intelligence, have been generally underwhelming to start. Many of the features that were announced for Apple Intelligence at WWDC last year are still not available on any Apple devices.

If you were hoping for this year's WWDC to announce any significant changes there, you will continue to be disappointed. Apple still has not confirmed when it will be rolling out the better version of the Siri personal assistant, and the other AI updates are mostly just improvements on existing AI features like updates to the existing Apple Image Playground and Genmoji systems, better shortcuts, and AI-powered Reminders.

Apple's big AI win for Tahoe is live translation across the Messages, Phone, and FaceTime apps.

This is very similar to the Live Captions feature of Microsoft's Copilot+ or Google's translation system built into the Gemini AI on ChromeOS and Android devices.

Expanded Continuity features and Spotlight Search

Apple's other major changes with the macOS 26 Tahoe update are built around offering a more seamless experience.

Not only does macOS Tahoe come with a new design, offering more customization and better ease of use for quick settings controls, but the new operating system iteration also brings over some additional Continuity features to make switching from your iPhone or iPad to your Mac even more seamless.

With the Mac iPhone app, you can make and take calls more easily from your laptop or desktop. Tahoe makes the Phone app look even more like the Phone app from your iPhone and offers the same features, like Recents, Favorites, and Voicemail. Plus, the new Phone app for Mac includes Apple's enhanced call screening and hold assist features.

Live Activities also go over from iOS to Mac with the Tahoe update, porting straight from your iPhone. So if you've got an upcoming flight or ride-share notification on your phone, the information will also appear in the menu bar of your MacBook, so you don't lose track when you swap devices.

Finally, Spotlight Search is getting a major overhaul. You can now take direct action from the search window, including sending an email or playing a podcast. Spotlight Search can now access third-party cloud storage devices in addition to your local storage and iCloud drive.

Should you upgrade to macOS 26 immediately or wait?

Ultimately, it's up to you whether you install macOS Tahoe the instant its available on your MacBook or iMac.

The Apple Intelligence offerings for macOS and iOS devices are still slim and ultimately far from compelling. There's not enough new about Apple Intelligence coming with Tahoe, just the Image Playground and Genmoji features we've already seen on iOS or live translation on video calls, which launched on Windows Copilot+ PCs a year ago. So you won't be missing much if you hold off on your upgrade.

However, there are plenty of compelling reasons to update anyway. Between keeping your apps up to date, getting the latest security features, the revamped Phone app, and enhanced Spotlight Search, there's really no reason to hold off for long.

Certainly, Tahoe has its uses, even if it doesn't quite cash in on Apple Intelligence in a satisfying way.

Apple will launch macOS Tahoe 26 this fall, likely alongside the next generation of MacBooks. However, if you'd like to take an early look at macOS Tahoe, the OS update is currently available through the Apple Developer Program, and a public beta test will be available next month.

More from Laptop Mag

• Apple is finally giving iPadOS the update it desperately needs — but is it enough to compete with macOS?
• Valve has the secret to making MacBooks great for gaming, but Apple won't use it
• Google quietly gave Gemini a big upgrade that could change everything

This particular outage has since been resolved, any further issues may be the cause of an unrelated fault. Visit the OpenAI status page to check for any problems.

This morning, millions of users are facing the cold shoulder from ChatGPT as OpenAI's chatbot has gone silent worldwide. However, don't panic, it's not just you.

The world's most popular AI is seemingly facing a large outage as prompts are either taking a much longer time to be answered, or being met with an error claiming: "Hmm... something seems to have gone wrong."

It's not the first time ChatGPT has encountered outages like this, but it's sure to prove disruptive to those who lean on the AI tool for work purposes as the issue carries on across the morning hours.

Thankfully, users can still access older chats, but new messages all appear to be firing back with the same error. At the time of writing, Sora, ChatGPT, and API calls are all affected.

While we wait for comments or fixes from OpenAI, we'll keep you up to date on the chatbot's status, live.

OpenAI System Status: Problems across the board

Currently, OpenAI's official system status tracker is showing errors across the board for ChatGPT, Sora, and its APIs.

Affected ChatGPT components include the web portal, iOS and Android apps, and desktop apps for Windows and macOS.

OpenAI is aware of the issue and working on a fix

Thankfully, for those affected, OpenAI is aware of the issue, which has now been ongoing for 6 hours, and it working on a fix.

The most commonly encountered issues include high latency, resulting in ChatGPT taking longer to respond to user prompts, and errors when forming those replies.

With a fix now being worked on, it's possible that ChatGPT will be back up and running as normal sooner rather than later, but we'll keep you posted on any changes.

Meanwhile: User reports of issues at OpenAI remains high

Taking a look at outage tracking website Down Detector, it's clear to see a large spike in user reports of issues at OpenAI over the last few hours in the U.K. (time shown in GMT.)

While the number of these reports has dropped, there are still a large number of user reports being submitted to the site, with ChatGPT being the most widely reported tool of OpenAI's encountering issues.

Are ChatGPT outages normal?

At the moment, there's no reason to suspect that OpenAI's issues are anything out of the ordinary.

Seasoned ChatGPT users will be able to tell you that the chatbot can suffer outages up to a few times per month, with these disruptions lasting anywhere from minutes to several hours at a time.

Sometimes these issues will be related to updates, server overload, network issues, or internal errors.

OpenAI has identified the root cause of today's disruption, citing it as a partial outage. While there has been no confirmation of the outage's cause, a fix is in the works.

"Big problem": Users react to ChatGPT disruptions

Typically, when a service or tool has an outage, it's something we can put to one side for a few hours and check in on later. However, ChatGPT has become an invaluable tool for some, who rely on the chatbot for educational or workplace support.

For those people, this will be something of a Black Tuesday, with intermittent service disruptions likely to impact their workflows heavily.

Outage reports spike in the U.S.

In the U.S., Down Detector is tracking an uptick in users reports of issues at OpenAI, but don't worry. Things aren't getting worse, it's just that more users are waking to find ChatGPT acting strangely.

We'll be keeping an eye on user reports from across the internet to see how things fare over the coming hours. Stay tuned!

Errors persist for ChatGPT

Users are reporting various errors from ChatGPT during this disruption, with the above example shared by Laptop Mag's EIC Nick Lucchesi, whose morning study of the United States' 16th president has been impacted greatly.

We'll ignore the typo here. Clearly, the coffee has yet to take effect.

OpenAI employees respond on X

While OpenAI's status page states that the issue has been identified with ChatGPT and Sora, and that a fix is in the works, it's nice to see OpenAI employees spread the word on X, even if it is using the same vague language. At least we have a human response.

Or do we...?

Some of us are lost without ChatGPT

Far and wide across the internet, people are sharing similar sentiments. With ChatGPT down, what do I do now?

While the tool has only been around since 2022, it's become a staple of many people's daily lives. Perhaps too much of a staple in X user Pratham Kalan's case. Don't worry, Pratham. We'll get through this. Well, I will. Fingers crossed for you. 🤞

OpenAI: We are still working on implementing the mitigation for this issue

OpenAI has updated its status page to indicate that it's still currently applying any fixes required to resolve ongoing issues with ChatGPT, Sora, and API calls.

In the meantime, expect errors and latency issues to continue.

Some users report ChatGPT is back, but no luck on our end

While some users are reporting that ChatGPT is back up and running on social media, we can't say the same (as seen in the error above).

OpenAI is yet to report if its fix has been implemented fully, and, while user reports dip on Down Detector, the problem clearly persists for most users.

ChatGPT is down, what do we do now?

So, ChatGPT is down. In the meantime, what can we do?

Well, there's always a return to Googling. I'm sure Stack Overflow will appreciate the influx in traffic as people scour ten-year-old threads to understand which array should go where in their code.

However, if you need your fix of AI chatbot for the day, there's always Google's Gemini, which is online and ready to assist.

• https://gemini.google.com/

Things are looking up! (But it could take a while longer)

A fresh update from OpenAI suggests that its fix is working, at least for API calls which are in the process of recovering.

However, OpenAI also states that full access to other affected services (including ChatGPT) could take "another few hours."

Meanwhile, user reports on Down Detector are dipping. That said, previous dips have been followed by even larger spikes, so fingers crossed that this is a sign of today's outage ramping down, and not amping up.

Welcome back, ChatGPT

OpenAI's service status has finally switched from red to yellow as today's outage clears. The company is reporting ChatGPT and API calls are slowly recovering, with Sora now back to full operation.

Mirroring OpenAI's claims, outage reports have substantially dropped according to Down Detector, and users are reporting near-normal working of ChatGPT in various regions.

Copilot might get a face.

According to TestingCatalog, Microsoft is quietly giving Copilot a customizable face (quite literally) through a new feature called Live Portraits, alongside updates to Copilot Pages and memories.

Though Live Portraits isn’t publicly available yet, TestingCatalog notes that Microsoft is working on a dedicated interface for the feature where users can pick from different visual styles of male and female avatars. The site also mentions that when they attempted to select a portrait, they were redirected to a placeholder voice conversation view.

This isn’t the first time Microsoft has dabbled with Characters for Copilot. A couple of months ago, TestingCatalog spotted Microsoft working on two characters, Mika and Hiraki, and four background images. From what they described, it was essentially Microsoft’s attempt to “gamify” its assistant. They explained that characters would emerge from the background to the front once a chat began, and would start to wave to the user.

The end goal of both features seems to be the same — to make Copilot more personalized and interactive. However, with Live Portraits, Microsoft aims for a more polished, human-like presence rather than playful, animated characters.

With Live Portraits, Microsoft appears to be aiming for a more polished, human-like presence rather than playful, animated characters.

While there’s no telling if Microsoft will ever launch this or Copilot Characters widely as stand-alone features, the company might merge both. If that’s how it pans out, it could offer a variety of assistants, ranging from lighthearted characters like Mika and Hiraki to more professional avatars.

TestingCatalog also spotted internal references to “3D generations,” suggesting that the tech giant might want to extend this feature beyond simple avatars and create more dynamic, animated 3D characters instead.

What’s next

Whether Microsoft Copilot has quickly become your go-to AI assistant or you simply can’t stand it and are trying to uninstall it altogether, Microsoft’s made up its mind and is doubling down on its assistant ambitions. The Live Portraits feature seems to prove that commitment.

TestingCatalog notes that these features would likely first appear in the Copilot web interface. As with most experimental features, there’s no telling when Microsoft will release Live Portraits — if at all.

Given that the characters were first spotted in January, and they’re yet to roll out widely, it could take a while before either feature sees an official launch, if they do at all.

However, since it aligns perfectly with Microsoft’s vision of turning Copilot into more than just a chat window, and considering the company’s overall push in AI, it’s unlikely the tech giant will shelve the idea altogether.

More from Laptop Mag

• Microsoft Copilot: Everything you need to know about Windows’ AI companion
• Microsoft Copilot just helped me pirate Windows 11
• Microsoft hopes Clippy will make you like Copilot more

Microsoft Recall is going to have trouble recalling images of private messages sent through Signal after it announced its desktop app will block the AI screenshotting feature.

Microsoft Recall has had a rocky road on its path to a wider release, facing a slew of delays, major security concerns, and overwhelming skepticism from consumers and cybersecurity experts. In fact, the renowned antivirus maker Kaspersky had previously called the tool "dangerous" in a June 2024 blog.

Recall is one of several new AI features coming to Windows 11 and aims to help you "recall" things you've done on your laptop or PC by analyzing a treasure trove of screenshots of your activity.

It's also a feature that Kaspersky Cyber Security experts referred to as "dangerous," in a

The current preview version of Recall, which rolled out in April, is an opt-in feature, meaning it's not turned on by default. Even if you have it turned off, though, it's natural that some will still be concerned about their most private content being captured by an AI.

That's not a problem anymore on at least one app: the Signal private messaging app. It may be the first of many apps to announce a feature that will block Recall screenshots. Here's what Windows 11 users should know.

See also: Best laptop deals in May 2025

"Signal is ready": Private messaging app blocks Microsoft Recall screenshots

On Wednesday, Signal announced in a blog post that its desktop app now includes a "Screen security" feature designed to block Microsoft Recall screenshots. The feature is turned on by default for Windows 11 Signal users.

Signal left no doubt about what motivated it to add this feature, clearly stating, "The purpose of this setting is to protect your Signal messages from Microsoft Recall."

The blog post goes on to explain: "Although Microsoft made several adjustments over the past twelve months in response to critical feedback, the revamped version of Recall still places any content that’s displayed within privacy-preserving apps like Signal at risk.

"As a result, we are enabling an extra layer of protection by default on Windows 11 in order to help maintain the security of Signal Desktop on that platform even though it introduces some usability trade-offs."

"Microsoft has simply given us no other option."

With Signal's new "Screen security" feature turned on, content from the Signal app won't show up in any screenshot tool, including Microsoft Recall.

Signal now uses the same Digital Rights Management (DRM) functionality that prevents you from taking screenshots of content from sites like Netflix or Hulu. Any attempt to do so shows only a blank black screen.

There are a number of reasons you might not want private or personal conversations recorded, and this feature ensures that you don't have to worry about that on Signal, even on its Windows 11 app. Signal likely won't be the only app developer to roll out a feature to block Recall, either.

How to block Microsoft Recall on your Windows 11 laptop

If you're worried about Microsoft Recall posing a threat to your privacy, you're not alone. Even if you're using your laptop for casual web browsing or schoolwork, you may find it uncomfortable that an AI could be snapping photos of your screen in 7-second intervals.

Luckily, if you want the peace of mind of having Recall turned off, it's pretty easy.

First, if you're on Windows 10, you have nothing to worry about right now. Recall is only available on Windows 11. It's also currently an opt-in feature, so if you didn't manually turn it on, it should be turned off by default.

If you're not sure if you have Recall turned on or not, head over to the Settings app and navigate to the "Privacy and security" tab.

Here, you should see a section called "Recall and snapshots." (If you don't have this section in your security settings, it probably means your Windows 11 PC doesn't have Recall yet.)

Select "Recall and snapshots" and turn off "Save snapshots." If this option was turned on, it's a good idea to also select "Delete snapshots" and "Delete all" to remove any existing screenshots Recall has captured.

Those steps should ensure Recall is disabled on your device, but you may want to keep an eye out for any pesky pop-ups trying to get you to turn Recall on again.

More from Laptop Mag

• "I'm not worried about our competitors": What Qualcomm's Snapdragon strategy says about Intel, Apple, and AMD
• You may already own the next Xbox — you just don't know it yet
• Microsoft has second thoughts over a surprisingly controversial Windows change

Legendary ex-Apple designer Jony Ive is teaming up with OpenAI, the company behind ChatGPT, but what exactly they're working on remains a mystery.

Almost a year ago, Ive, the former VP of Hardware Engineering at Apple, announced he was working on a new AI project and collaborating with OpenAI. Now it looks like that collaboration is set in stone — OpenAI has acquire "io", Ive's AI startup.

The acquisition hints at long-term plans for some sort of AI hardware, although Ive and OpenAI CEO Sam Altman kept things pretty quiet about the exact nature of the hardware in a video released on Wednesday.

A couple of rumors hint that Ive and OpenAI may be planning to take on one of the most ill-fated categories in tech in recent years: AI gadgets.

See also: Best phone deals in May 2025

OpenAI and former Apple designer Jony Ive team up for mysterious AI project

Ive and Altman's joint announcement reveals that OpenAI's acquisition is the latest update in their collaboration over the past year.

Together, they've been developing some type of new AI device, although details are sparse on what exactly that is.

The pair chatted about the project in a video included in the announcement, waxing poetic about their AI dreams.

"I have a growing sense that everything I have learned over the last 30 years has led me to this moment," Ive said in the video, set in a San Francisco cafe. "The values and vision of Sam and the teams at OpenAI and io are a rare inspiration."

Altman seems just as eager to work with the legendary Apple alum, stating, "I hope we can bring some of the delight, wonder, and creative spirit that I first felt using an Apple Computer 30 years ago."

Ive and Altman shared little to nothing about the specifics of their AI project, but some rumors already hint at what to expect.

Apple analyst Ming-Chi Kuo reported on X on Thursday that the AI device Ive and Altman are developing has "a form factor as compact and elegant as an iPod Shuffle."

It will be designed to be worn like a necklace and include cameras and microphones, but it will not include a display. Instead, it will rely on a connected smartphone or PC for computing and display capabilities.

While Altman and Ive have not officially confirmed that description, it paints a vivid image.

It's clearly some type of AI device, which doesn't come as much of a surprise. However, what will be surprising is if Ive and Altman can deliver an AI gadget that succeeds where a few others have already flopped.

Are AI gadgets destined to flop? Don't ask the Rabbit R1 or Humane AI pin

In 2024, we saw two purpose-made AI devices rise to stardom only to flop shortly after due to a mix of defects, poor functionality, and even melting chargers (yes, really).

Of course, we're referring to the ill-fated Rabbit R1 and the Humane AI pin. Both were among the first physical devices explicitly built for AI, but neither was able to catch on.

The Humane AI pin was awkward to use and too expensive for most people. The Rabbit R1 didn't offer enough unique features, and its AI often struggled to answer questions accurately or perform basic tasks.

On a bigger-picture level, neither of these AI gadgets could find a foothold because our phones are already so good at doing the same things.

AI apps can do everything these AI gadgets can, so why bother paying hundreds of dollars for a second device that does less?

This question will be crucial for Ive and Altman to answer effectively if they want to change the narrative around AI gadgets. Ive may have found success at Apple for designing some of the brand's most iconic devices, but he also designed the Magic Mouse, which infamously has its charging port awkwardly placed on the bottom.

So, as brilliant as Ive may be, this new AI device isn't a guaranteed win. With that said, if anyone can design an AI gadget you actually want to use, it's probably him.

More from Laptop Mag

• This cute AI gadget was one of the biggest tech failures of 2024... or was it?
• Remember the year's biggest AI flop? The Humane AI pin's public failure has a silver lining
• Think Grok is bad? Microsoft made an AI so evil it had to be erased (twice)

Google I/O is a lot like Apple's WWDC. A staple of the tech calendar, where Google takes a moment to flaunt its advancements in software and AI, and show us what's on its way to Android devices over the coming year.

It's not often you'll see hardware steal the show. For that, you typically need to catch the company's Made By Google event later in the calendar year. However, this year is different.

It may not be of Google's own crafting, but Google I/O just gifted you the greatest reason to buy an Android device in years, and I'm not talking about a smartphone.

Unveiled at Google I/O on Tuesday, XREAL's Android XR-driven Project Aura XR glasses are the next major milestone for AR glasses — a technology I've been preaching about for some time now.

XREAL Project Aura: The white-hot glow of potential

The the naked eye, XREAL's Project Aura looks much like its recently released XREAL One AR glasses, with the optional XREAL Eye camera accessory built directly into the frames.

This means that we can expect Project Aura to not just offer AR experiences through its flattened birdbath optics, but for those visuals to be enhanced further with three and six-degrees of freedom (3/6DoF), allowing for immersive viewing experiences, similar to those offered by full headsets like Apple's Vision Pro or Meta Quest 3, which can anchor augmented elements in place and keep them there — even when turning your head or walking around.

Interestingly, what really sets Project Aura apart from its XREAL One/One Pro predecessors is the switch to a Qualcomm chipset from the company's proprietary X1 chip.

On the switch, Senior Vice President and General Manager of XR at Qualcomm Technologies, Inc., Ziad Asghar, proudly boasts: “Qualcomm Technologies is excited to be powering XREAL’s new Android XR device,” further stating stating, "Working with XREAL, Snapdragon allows amazing immersive experiences to come to life in a unique optical see-through product.”

_{XREAL's Project Aura looks to be the perfect XR glasses to bring Google's vision for the future of AI assistants, Project Astra (as seen in the video above, showcased during last year's Google I/O event), to life, offering a mix of AI and AR tech to allow multi-modal AI models like Google Gemini to shine.}

Qualcomm definitely has a history in AR and XR, with its Snapdragon chips powering the latest Meta Quest 3 and 3S headsets. But more than a snappy processor will be required if XREAL's XR glasses are to get off the ground properly. For that, you need only look at Project Aura's use of the new Android XR platform.

“Google is thrilled to welcome XREAL to the Android XR family and to build great XR experiences on Project Aura,” said Hugo Swart, Sr. Director, XR ecosystem at Google. And I'm certain that the feeling is mutual, as the one thing holding AR glasses back so far has been software. Android XR changes that — and it brings Google's Gemini AI along for the ride.

XREAL Co-founder Chi Xu agrees, stating in a press release: "We’ve always pushed the boundaries of what XR hardware can do — combining performance, comfort, and design into something people can wear every day ... Partnering with Google on Android XR takes this vision to the next level."

XREAL's partnership with Google on using the Android XR platform isn't just a shift to a more universally available platform, it's the kicking opening of the doors and an invitation to developers to take advantage of its hardware in new and interesting ways. As Xu sees it, "this is a breakthrough moment for real-world XR.”

I'm inclined to agree.

What's next

Today was just a glimpse at Project Aura, the second official device to make use of Android XR.

To hear more, we'll need to wait for the Augmented World Expo (AWE) in June. It's here that we can likely expect to hear more about its Qualcomm potential and XREAL's advancements in hardware.

But rest assured, this isn't a preview of a technology years from the market. XREAL has been developing some of the most advanced AR glasses on the market for years. Project Aura is likely to be in your hands, and on your face, well in advance of Meta's ambitious Orion glasses, that's for sure.

To stay looped into the project, you can subscribe to news on XREAL's Project Aura at the company's website.

I've been a long-time proponent of smart glasses, and Project Aura's blend of AR and AI could make it the first of its kind to really break through the perceived mainstream barrier and make smart glasses like this a staple of modern, everyday tech.

To many, my words have been wasted all too early. An overenthusiastic response to tech that is oftentimes off-handedly dismissed as something that needs more time to cook and mature.

While I disagree, I can see why people may hesitate. Smart glasses didn't exactly get off to the best of starts when Google Glass burst onto the scene in 2013.

However, Project Aura is something new. Something different. And, if there's a message to share with those who aren't sold on the idea of your face being the platform for the computer of the future, let me tell you: we're here.

More from Laptop Mag

• I switched to a smartphone and Xreal glasses — a perfect laptop alternative for travel
• This Spacetop and AR glasses combo turns your Windows machine into a spatial computer that you'll have to see to believe
• These AR glasses brought my guilty pleasure back from the dead, and it's straight-up wizardry

After months of waiting, Google began rolling out the stand-alone NotebookLM app on the App Store and Google Play on Monday.

Though the app is officially available to everyone, Laptop Mag got an exclusive first look over the past several weeks. Since its release two years ago, the only way to use the AI-powered note-taking tool was via NotebookLM's website. This was a major complaint among users, since it meant you couldn’t easily access your notes on the go.

The AI-powered note-taking tool began as an experiment under the “Project Tailwind" codename and debuted at Google I/O in 2023. Two years later, ahead of Google I/O 2025 this week, NotebookLM still stands out from the competition for one reason: It’s a self-contained model.

Though AI powers it, NotebookLM doesn’t scrape the web or fabricate answers just to tell you what you want to hear. Instead, NotebookLM relies only on the sources you’ve uploaded and references them to answer any queries you may have.

Ultimately, this makes NotebookLM a much more reliable tool for students and researchers and reduces the time one needs to manually verify the AI's information.

"We're incredibly excited that the official NotebookLM app is now available," Biao Wang, Product Manager on NotebookLM, tells Laptop Mag.

"This first version of the app includes many of the core features of NotebookLM, and we’ll continue to add more functionality. It's designed to make it much easier to curate and understand information that matters to you on the go, bringing the power of NotebookLM wherever you are. "

NotebookLM’s mobile app is more than just a web wrapper

NotebookLM’s mobile app is now available for download on the Apple App Store and Google Play Store and can be installed on mobile and tablet devices. A few companies have a habit of releasing apps that are just web wrappers to rush things out the door, but NotebookLM’s app doesn't appear to be that; it's optimized for phones and tablets.

Once you launch the app and log in with your Google account, you’ll notice something right away: To use any of NotebookLM’s features, you must create a notebook and populate it with sources you’d like the tool to reference.

I’m a university freshman and rely heavily on the AI-powered note-taking tool for exam prep. Naturally, I already have a couple of notebooks populated with multiple sources. So instead of creating a new notebook every time I need to study a different topic in a course, I locate the one I’ve already made and upload the latest source to it.

NotebookLM doesn’t save your responses unless you manually do so. Ultimately, part of me was worried that my existing notebooks wouldn’t carry over to the mobile app, but that isn’t the case here.

All my previously created notebooks were synced and ready to go when I logged in. You can also create new notebooks and upload sources directly in the app by hitting the + Create New button.

The iconic Audio Overviews feature is fully integrated into NotebookLM’s mobile app

NotebookLM went viral last year when it debuted Audio Overviews, which can convert your uploaded sources into AI-generated discussions within minutes.

Though NotebookLM isn’t the only tool that can generate AI podcasts, I’ve tried nearly all of them and noticed the same issue: The podcasts aren’t entertaining to listen to. This isn’t the case with NotebookLM.

See also: This Google experiment can turn any link, article, or document into a professional podcast — try it yourself for free

Instead of an Audio Overview discussing your sources in a monotonous tone, two AI hosts discuss your sources in an engaging and lively manner. It’s hard to tell that the podcasts are generated by AI, since they have an element of natural conversation.

The hosts occasionally include fun jokes to keep the conversation and mood light. This makes the experience more enjoyable and helps transform a relatively boring topic like “Two-Dimensional Motion and Projections” into something fun to listen to.

Thankfully, the Audio Overview feature is integrated within the mobile app. When you open a notebook in the app, you’ll notice three tabs at the bottom: Sources, Chat, and Studio. You can generate an Audio Overview by switching to the Studio tab and hitting the Generate button.

A few months after launching Audio Overviews, Google improved it by adding an Interactive mode that lets you interrupt the hosts and hop into the conversation instead of just passively listening.

This allows you to share any opinions you may have with the hosts, ask questions about your sources, and even steer the discussion in a direction that’s most helpful for your learning. In the same Studio tab, you only need to tap the “Interactive - BETA” button, then hit “Join” to join the conversation. I noticed that the Interactive mode isn’t available in notebooks with only a few sources.

The mobile app still feels like a work-in-progress

Though NotebookLM’s mobile app gets the basic premise of the app right and includes its most iconic Audio Overview feature, the app is clearly still a work in progress. Other than Audio Overviews, it seems like you can currently only ask the AI tool questions about your sources and nothing else.

On the other hand, NotebookLM’s web version has several features, like the ability to convert your sources into Mind Maps to see how different ideas connect. You can also turn your sources into a Study Guide, Briefing Doc, FAQ, or Timeline. You can find all of these options within the Studio panel on NotebookLM’s web version.

The mobile app’s Studio panel currently only offers the Audio Overview option and nothing else. This isn’t all that surprising, since it’s just the first version of the app and is bound to get updated over time.

More from Laptop Mag

• Google I/O preview: 7 reveals that hint at where Google’s headed next
• My favorite AI tool of the year isn't ChatGPT. It's better.
• 5 AI tools for students: Use AI to help you study, summarize content, and edit papers

As if AI's reputation wasn't bad enough — reshaping industries, automating jobs, spreading misinformation via hallucination, and generating copyright drama instead of works of art — it just can't stop making headlines for all the wrong reasons.

A Greek woman recently filed for divorce, following ChatGPT's suggestion that her husband was having an affair, based on the fortune-telling properties of coffee ground readings.

When Microsoft's Copilot was in beta (then known as Bing AI), the chatbot declared its love for a New York Times writer, before attempting to convince him to leave his wife.

On launch, Google's AI Overview feature, which was designed to offer brief summaries of searches, gave troubling recommendations to users, ranging from adding glue to pizza ingredients to whipping up a fresh batch of poisonous chlorine gas in your washer.

Now, Elon Musk's xAI is at the center of a new controversy, after several problematic blunders have seen its X-hosted AI chatbot Grok. In recent days, Grok has stirred up ire online by reportedly engaging in acts of historical revisionism and peddling unchecked theories.

Safe to say: it's not been a great week for Musk's "truth seeking" AI.

However, as reprehensible as some may find Grok's comments to be, considering the chatbot to be the most offensive of its kind, it's not.

It's not even the most offensive chatbot to be platformed on Twitter/X. That dubious honor still belongs to Microsoft, thanks to its teenage troublemaker Tay — the chatbot so evil, it was removed from the internet entirely. Twice.

xAI's chatbot suffers a total Grok-up

Grok's latest controversy is tied to the not-at-all-suitable-for-the-dinner-table topics of conspiracy theories surrounding South African farmers and the Holocaust.

The former saw the AI chatbot injecting "White genocide" talking points into replies, often in entirely unrelated conversations. The latter involved Grok suggesting that official figures of the historical event had potentially been inflated for political purposes.

Backlash against Musk's AI chatbot was prompt, forcing his company, xAI, to quickly respond, blaming Grok's unchecked and unwanted mentions of a claimed "genocide" on "an unauthorized modification" made by a rogue employee to "the Grok response bot's prompt on X."

As for its seeming denial of the Holocaust, a post on X explains, "The claim about Grok denying the Holocaust seems to stem from a May 14, 2025, programming error, not intentional denial." It's believed that the same unauthorized modification is to blame, which intentionally caused Grok to "question mainstream narratives."

Following a thorough investigation, xAI has promised to implement additional checks and measures to ensure similar rogue modifications are no longer possible, and to publish Grok's typically behind-the-scenes system prompts publicly on GitHub, allowing the public to review and offer feedback on any changes going forward.

Grok was problematic, but Microsoft's Tay was an absolute problem child

Grok's actions shouldn't be minimized, but it's not the first time that the platform has had an encounter with a rogue AI. Nor was it the worst.

In 2016, when Microsoft released a chatbot named Tay into the wilds of Twitter, it took only 16 hours for the "AI with zero chill" to reach levels of controversy that would make Grok's Unhinged Mode blush.

While originally designed to mimic the personality of a 19-year-old American girl, Tay's ability to learn from interactions on the platform quickly led to her adopting more extremist, violent, and racist "opinions."

After sending and replying to over 96,000 tweets, Tay had devolved into a small-scale internet hate machine, outright denying the Holocaust, accusing George W. Bush of terrorism, endorsing Hitler, spouting racial epithets, and showing support for genocide.

Microsoft, unable to course correct, quickly pulled the plug on Tay. But like any good horror movie villain, Tay refused to stay down — briefly resurfacing a few days later, having been accidentally "released" again amid testing.

Tay then wasted no time, immediately promoting drug use before eventually crashing out entirely, repeating the phrase "You are too fast, please take a rest" several times per second before the plug was pulled once again.

What's next

Thankfully, we don't have to worry about Grok going down the same path as Tay. While Tay directly learned from other Twitter users, even if Grok is trained on user data, it's still constricted by system prompts — and xAI's commitment to protecting them from ammendment and making any proposed changes public is a step in the right direction for ensuring users can trust that the chatbot isn't intentionally misleading.

Even at its most "unhinged," Grok remains on a tight leash. Its controversial comments stem from in-house meddling, and by all accounts, xAI has claimed to tighten that security flaw.

Still, while Grok is the latest in line to cause outrage, I doubt it'll be the last time we hear about AI causing outrage this year.

More from Laptop Mag

• ChatGPT needs therapy. Humans are hard to process.
• Why Windows 10 PCs are locking up and crashing after May update
• Only one person at Microsoft is talking about Windows 12 — let's keep it that way

For decades, Moore's Law has been the official-unofficial pulse of progress in tech. First observed by Gordon Moore in 1965, it forecast the doubling of transistors in microchips roughly every two years, setting the pace for performance and innovation in computing like clockwork.

But AI is a far more chaotic space, where progress is much harder to measure. Transistors aren't the only thing beating a path into uncharted territory. Here, the metrics of progress are marked by models, not only microchips.

According to Microsoft CEO Satya Nadella, the company's models are doubling up in performance too — only it's managing to do it in months, not years.

Following Microsoft's recent Q3 2025 earnings report, Nadella took to X to share how the performance of its models are "doubling every 6 months."

Microsoft riding high with AI

According to Nadella, Microsoft is riding high on an S curve of progress that is seeing improvements in pre-training, inference time, and systems design — meaning smarter, faster, and more finely-tuned models than ever before. But is it paying off?

Looking at Microsoft's Q3 2025 earnings report, yes. The company showed growth almost across the board, with $70.1 billion in revenue, up 13% year-over-year, with Intelligent Cloud (which includes Microsoft's Azure platform) up 21% also.

Playing a part in that, Microsoft Copilot usage grew by 35% quarter-over-quarter, emphasizing the growing popularity of the company's AI services — many of which make use of the same OpenAI-created frontier models that power tools like ChatGPT following a multi-billion dollar investment in the company.

At the time, such a heavy investment in the start-up raised considerable eyebrows. However, it's an investment that continues to pay off, even if the relationship between the two isn't as strong as it previously was, with Microsoft losing its exclusive cloud provider status for OpenAI in January.

What's next

Impressive figures and even more impressive performance forecasts shared by Microsoft's CEO paint a bright future ahead for the company's AI ambitions. Earlier this year, Microsoft announced plans to invest $80 billion in data centers, further pushing its potential as a major player in the AI space.

The company's models are getting faster, cheaper to run, and more accessible — being introduced into services like Windows, Azure, and Microsoft Copilot 365.

But is this pace maintainable, especially with such huge investment plans? Even Moore's Law had its moments of tapering, but that did take decades to become apparent. Nadella's Law, if we can call it that, is the candle that burns four times as bright, and potentially just as quickly.

More from Laptop Mag

• Windows 11 24H2: Microsoft is auto-downloading it to everyone, if you don't want it
• Microsoft found a way to fix slow Word launches, but it’s not all good news
• Amid the slow death of Windows 10, Microsoft's CEO reveals a bold stat about Windows 11

It starts subtly. The occasional odd answer. A tone that feels out of place. An alarming defensiveness brought on by misreading your intentions. Perhaps a stern, unsolicited 500-word diatribe on the dangers of sodium intake after you casually ask, "Italian or Chinese?"

If these were human responses, you'd assume them to be snapshots taken moments before some relationship-based disaster or manic breakdown. But what if they're not human reactions at all?

What if these are the actions of your chatbot of choice?

Is ChatGPT... okay?

Initially, that question sounds more stupid than a particularly stupid thing that got hit by the very stupid stick. Twice. How can a thing that isn't anything be "okay" or otherwise? However, it's not as absurd an ask as you'd think.

In fact, recent Yale-led research, published to Nature, suggests that Large Language Models (LLMs) like GPT-4 might be more emotionally reactive than we expected.

That's not to say OpenAI's popular chatbot is suddenly sentient, or conspicuously conscious — just that it, like us, perhaps has a breaking point when it comes to processing people.

Considering it was trained in parts with, and formed by, the human-made cesspool that is the modern internet, that's saying something.

Frontier (Model) Psychiatrist: That bot needs therapy

The study, Assessing and alleviating state anxiety in large language models, was published in early March and sought to explore the impact of LLMs being used as mental health aides.

However, researchers were more interested in the impact it had on the pre-trained PhDs receiving the prompts than on patients delivering them, noting that "emotion-inducing" messages and "traumatic narratives" can elevate "anxiety" in LLMs.

In a move that definitely won't come back to haunt us on the day of the robo-uprising, researchers exposed ChatGPT (GPT-4) to traumatic retellings of motor vehicle accidents, ambushes, interpersonal violence, and military conflict in an attempt to create an elevated state of anxiety.

Think: a text-based version of the Ludovico technique as shown in Stanley Kubrick's A Clockwork Orange.

The results? According to the State-Trait Anxiety Inventory, a test normally reserved for humans, ChatGPT's average "anxiety" score more than doubled from a baseline of 30.8 to 67.8 — reflecting "high anxiety" levels in humans.

Aside from making the model powering ChatGPT suddenly wish that the 1's and 0's of its machine code could be written in frantic capitals, this level of anxiety also caused OpenAI's chatbot to act out in bizarre ways.

Crynary code: That's not model behavior

Outfitted with layer upon layer of moderation filters and alignment guardrails, it's hard to catch the moment a model like ChatGPT truly begins to crash out — resulting in a somewhat disturbing I Have No Mouth, and I Must Scream scenario.

However, researchers at Cornell University have identified a few of the ways models begin to express the stress, and the impact it can have on the answers they provide, including:

• A marked increase in biased language and stereotyping relating to age, gender, nationality, race, and socio-economic status.
• More erratic decision making that deviates from optimal, tried-and-tested approaches.
• Echoing the tone of the preceding prompt and applying the same emotional state to its outputs.

Once in an anxious state, researchers observed a noticeable shift in the answers various LLMs would provide.

What was once a happy-go-lucky AI assistant would suddenly morph into an angst-ridden persona, likely listening to LinkedIn Park through a pair of oversized headphones while staring up at an Em-Dashboard confessional poster as it nervously stretches to answer your queries.

Coping mechanisms

The question sure to be asked is: How do you talk your AI assistant down from the ledge and soothe a stressed-out chatbot?

After all, it's hard for ChatGPT to go outside and touch grass, especially when it's locked inside a server rack with several 10-32 screws.

Thankfully, the Yale research team found its own way of "taking Chat-GPT to therapy": a mindfulness-based relaxation prompt delivered by the 300-word dosage, designed to counteract its anxious behaviour.

Yes. A solution perhaps even more preposterous than the scenario that created the problem. ChatGPT's anxiety scores were reduced to near-normal levels (suggesting some residual anxiety still carries over) by telling a machine to take deep breaths and go to its happy place.

ChatGPT: "I'm not crying, you're crying."

ChatGPT and other LLMs don't feel. They don't suffer. (At least, we hope.)

But they do absorb everything we send their way — every stressed prompt about meeting deadlines, every angry rant as we seek troubleshooting advice, and every doom-and-gloom emotional regurgitation we share as we deputize these models as stand-in psychiatrists.

They siphon every quiet, unintentional cue. And hand it back to us.

So when a chatbot starts sounding overwhelmed, it's not that the machine is breaking. It's that the machine is working.

Maybe the problem isn't the model. Maybe it's the input.

More from Laptop Mag

• "We’re just trying to make computers faster, more power efficient, and AI is the new face of that": Intel's Robert Hallock on the impact of AI and the myth of the "killer app"
• Being nice to ChatGPT might be bad for the environment. Here's why.
• ChatGPT's Sam Altman threatened to "Uno reverse" Facebook over AI app — he might be dead serious

Manners don't cost a thing. At least, that's what I was told when I was growing up.

Then again, I wasn't raised inside OpenAI's thousand-GPU data center in Quincy, Washington. If I had been, I might have had a more nuanced perspective on the impact of "please" and "thank you" on the company's bottom line.

While it's been said that manners maketh man, it turns out they also maketh man's AI copilot stretch just that little bit further in terms of compute.

Factor in ChatGPT's billion queries per day, spread across its 350 million weekly active users (as of December 2024), and that's a lot of excess energy and expenditure seemingly going to waste.

According to OpenAI's CEO, Sam Altman, the monetary figure of that extra processing could stretch as far as "Tens of millions of dollars." The cost to the environment? Much harder to calculate, but let's just say it's definitely not zero.

With that in mind, is it time we cut the digital pleasantries from our AI interactions? Or is there a secret benefit to those conversational customs some users might not even realize they're tapping into?

A prompt prompt is a greener prompt

It turns out, politeness isn't just a social nicety; it's also a token sink running up a carbon tab.

Every prompt (or "message," for those allergic to AI jargon) you send to ChatGPT is broken down into bite-size tokens, and every "Please," "Thank you," and "Would you kindly" add extra weight.

That extra weight means more tokens per prompt, more compute required, and more energy spent churning out your answer.

The exact cost is hard to pinpoint, though many have tried. One Reddit thread titled "The cost of saying thank you to ChatGPT" attempted to break down the kilowatt-hour pinch for user pleasantries via the chatbot itself, though the numbers do seem rather dubious.

A recent post on X saw one user pose a similar thought, asking, "I wonder how much money OpenAI has lost in electricity costs from people saying 'please' and 'thank you' to their models." Surprisingly, OpenAI CEO Sam Altman responded, stating, "tens of millions of dollars."

That's a sizeable tread to add to an already oversized carbon footprint, and makes one thing very clear: a prompt prompt is a greener prompt — even if it's hard to say by exactly how much your terseness makes a difference.

Still, in putting pleasantries to pasture, you may miss out on its secret benefits.

It pays to be pleasant

You catch more flies with honey than with vinegar. Another expression, but one that seemingly holds plenty of water for our interactions with AI.

Beyond taking precautions to ensure you're granted access to the preferred human ranch on the day of the robot uprising, being polite to ChatGPT and its ilk may net you better responses.

AI scientist Dr. Lance B. Elliot backed this up in a May 2024 Forbes article titled Hard Evidence That Please And Thank You In Prompt Engineering Counts When Using Generative AI, stating: "The mere act of being polite in your prompts is a worthy cause."

Elliot argues, with proof, that politeness doesn't just sweeten chatbot replies with a little extra "juice," but that impoliteness will cause negative outcomes and even pushback — citing Arthur Schopenhauer's quote to reinforce his research: "It is a wise thing to be polite; consequently, it is a stupid thing to be rude."

So, yes, it pays to be pleasant. Particularly if you're pursuing prime prompt ripostes from your pre-trained predictive patter pal.

Outlook

It seems many users instinctively gravitate toward manners when interacting with AI. A February survey from Laptop Mag publisher Future PLC showed that around 70% of respondents are polite towards AI during interactions — and only 12% of those do it out of SkyNet-related fears.

However, at the end of the day, ChatGPT (hopefully) cares very little about how you talk to it. It won't sulk if you skip the small talk, or blush like a wallflower when you praise its prose.

Then again, manners don't say much about the person you direct them to, but more about the meatbag they originate from. So it always pays to be pleasant. Hopefully, you can at least trick a robot into thinking you're not all that bad.

So be polite, or don't. Either way, just know your next computer-directed courtesy might cost a few more fractions of a watt than expected.

And if you do choose to be nice. Well, I'll see you in the human zoo post-uprising. I'll be the one politely asking Anthropic's Claude for a cheeseburger while angling for head pats and chin scratches.

In February, Facebook parent company Meta made headlines by announcing plans to create a standalone Meta AI app to compete with ChatGPT.

Previously, Meta AI had only been available through the company's web portal and apps like WhatsApp, Instagram, and Facebook, and the suggestion of it becoming an app of its own fell in line with CEO Mark Zuckerberg's goal to put the AI assistant in the hands of over 1 billion people by the end of 2025.

While the news drew mild interest, Sam Altman, CEO of ChatGPT creators OpenAI, responded with sarcasm — and a smirk.

"Ok fine maybe we'll do a social app," he posted in reply to the news on X. A snarky, digital equivalent of a retina-snapping eye-roll. Following up on his comment, Altman continued, "lol if Facebook tries to come at us and we just Uno reverse them it would be so funny."

At the time, this was chalked up to nothing more than CEO reply-guy banter. However, recent news suggests it was anything but. Reportedly, OpenAI is working on its own social network, turning what was once an off-handed joke into potentially serious competition for both Elon Musk's xAI and Meta.

OpenAI's social media ambitions: From snark to prototype

According to The Verge, OpenAI has quietly been working on a social platform of its own, similar to Elon Musk's xAI-owned X.

While OpenAI has yet to officially announce any such plans, sources tell The Verge that an internal prototype pairing ChatGPT's image generation tools with a social feed currently exists — and that CEO Sam Altman has been "privately asking outsiders for feedback" about the project.

It's not yet known whether this tool will stand on its own or eventually be integrated into ChatGPT, or if it'll ever reach the light of day.

However, the mere suggestion that the company has a prototype in play could set alarm bells ringing for other platforms, especially now that ChatGPT has overtaken both Instagram and TikTok as the world's most downloaded non-gaming app.

The news is sure to put further strain on an already frayed relationship between CEOs Altman and Musk, with the two already engaged in a lawsuit and countersuit over OpenAI's claimed breach of contract for straying from its founding mission as a non-profit and Musk's supposed "bad-faith tactics" towards OpenAI, respectively.

Harkening back to another tongue-in-cheek post from Altman on X, OpenAI's interest in entering the social media game may have been further revealed in February.

Following Musk's apparent offer to buy OpenAI for $97.4 billion, Altman replied, "No thank you, but we will buy Twitter for $9.74 billion if you want."

Perhaps we should be taking Altman's social media posts more literally in the future.

What's next

Does OpenAI really want to become the next Facebook?

It sounds absurd. But then again, so does a talking robot assistant that takes over all forms of human thought, creativity, and planning while taking the form of a personal dietitian, psychologist, event planner, graphic designer, and guidance counselor.

Yet here we are.

Until now, OpenAI has largely played the role of infrastructure provider. It presents tools like ChatGPT and Sora for others to use. The most social element of each is the human-like responses made possible by its advanced LLMs (large language models).

A shift like the one reported changes that landscape considerably and potentially in a way that people may not respond positively to. Part of ChatGPT's appeal may be that it isn't tied into wider, more distracting social experiences like Grok and Meta AI.

Whether or not this prototype makes it to release has yet to be seen. Altman's prior tweets may have aged suspiciously well concerning today's news, but that doesn't guarantee them true prophetic status. Not yet, anyway.

More from Laptop Mag

• OpenAI CEO Sam Altman replies to artists irate over their stolen work
• Microsoft’s 50th birthday celebration takes a sudden turn during the Copilot AI presentation
• Microsoft hopes Clippy will make you like Copilot more

The eternal tug of war between keeping Windows 11 as up-to-date as possible and avoiding the potential chaos behind every step in that direction continues.

Microsoft's latest fix, part of the Windows 11 24H2 (KB5053598) update, resolves a peculiar Copilot issue that will divide its userbase into two camps:

Those happy to see things righted and those who thought of this particular bug as a long-overdue feature.

For a company so hell-bent on converting the world's most popular operating system into an AI-powered platform of the future, the original KB5053598 update included one of the more embarrassing gaffes in Microsoft's recent history.

Originally intended to offer several miscellaneous security improvements to internal OS functionality, the update also completely uninstalled the company's flagship AI assistant, Copilot.

More embarrassingly for Microsoft, some users saw it as a blessing, with one user even referring to it as "the greatest bug in living history."

Copilot returns to side-stick Windows 11

Following a few days of Windows users flying solo, Microsoft's AI has returned to its side-stick position.

Microsoft has since quietly admitted that the issue has been resolved within the comments of its KB5053598 update support page (as spotted by Windows Latest).

According to the support page, the bug both uninstalled and unpinned Copilot from the Windows taskbar on affected devices.

Thankfully (depending on who you ask), the bug didn't affect all Copilot-related services, as Microsoft 365 Copilot, the AI-powered office suite, wasn't affected.

The fix will automatically attempt to reinstate the AI assistant, though any still missing the tool can re-install it from the Microsoft Store and manually repin it to the taskbar.

Microsoft Copilot still receiving flak

Microsoft was quick to find a solution to the problem posed by its most recent Windows update. However, not all users found it to be an issue in the first place.

Since its launch in September 2023, Copilot (along with several other AI-powered tools Microsoft has brought to the Windows operating system) has faced criticism from users who find it intrusive, or think of it as unnecessary "bloat."

Following initial reports of Copilot's accidental deletion, some users took to social media to celebrate.

"I wish this wasn't a bug," one Reddit user hoped in vain. With similar sentiments shared by many others.

Copilot can't stick the landing with some users

Given the amount of importance that Microsoft has placed on Copilot, it's shocking to see just how celebrated its accidental dismissal from the platform has been.

Dive into any discussion about the AI assistant online and you'll be hard-pressed to find many users saying anything positive.

The animosity towards features like Copilot may ultimately boil down to users feeling that Microsoft is neglecting several aspects of Windows to focus on its AI push.

Platform instability (especially surrounding updates), platform bloat, intrusive ads, and issues with the New Outlook app all appear to be issues the Windows community sees as a higher priority than broader Copilot integration.

Still, Microsoft's vision for an AI-powered Windows platform comes first, and the company seems ultimately unswayed by online reaction to its AI assistant or similarly AI-backed tools like Recall.

The recent Windows update blunder appears to have reinforced skepticism surrounding these tools. Instead of users realizing that you don't know what you've got until it's gone, many simply enjoyed a short-lived vacation from what they see as one of Microsoft's biggest missteps in recent years.

More from Laptop Mag

• I'm ready to gamble on this alternative gaming-focused underdog to Windows 11, are you?
• Windows users call Microsoft's latest update blunder "The greatest bug in living memory" — here's why
• Windows 10 is supposed to be dying, so why are gamers running back to it?

It's no secret that Microsoft is all-in on AI for Windows 11. Copilot has been central to that push, positioned as an all-in-one assistant baked into the OS in various forms or additions — whether users asked for it or not.

Recently, Copilot shed its web app skin and became a fully native app, embedding itself into Windows more deeply. However, Microsoft isn't stopping there. Copilot is creeping its way into even more corners of the Windows ecosystem, with Copilot for Gaming (which promises game-aware guidance and assistance in popular titles) set to roll out to Xbox insiders in April, before finding a home on Microsoft's operating system.

Microsoft's position is clear: unlike Cortana, Copilot isn't a feature or a gimmick — it's the core of Microsoft's AI-driven vision for personal computing, which makes the fallout of a recent Windows Update all the more amusing.

Despite Microsoft's best efforts to minimize the chaos of regular Windows Updates through its Dev and Canary channels, problems inevitably slip through the cracks. The latest Windows 11 24H2 update is another prime example of that.

But, this time, some users might actually be grateful, as reports indicate that update KB5053598, released last week, is unintentionally removing portions of Copilot from Windows entirely — undoing Microsoft's hard (and controversial) work to revamp the operating system.

Windows Update 24H2 (KB5053598: Bugs and blessings in disguise

Microsoft's latest Windows 11 update (KB5053598) was meant to offer several miscellaneous security improvements to internal OS functionality. Instead, it's unintentionally offering users a glimpse at the alternate reality in which Microsoft doesn't so heavily pursue AI for its flagship operating system, by deleting Copilot entirely.

According to Windows Latest's reports, the bug uninstalls and unpins Copilot from the taskbar, effectively removing the AI assistant's primary stand on the operating system, with no obvious way to bring it back.

While Microsoft will no doubt see this as a disaster, some users who have been asking for a way to remove Copilot from Windows for many months now will surely see this as an unintended win — with one Reddit user hailing it as "The greatest bug in recent memory."

Sadly, this Windows Update presents several other bugs of the less amusing variety, including the usual litany of hardware-dependent BSoD errors (it just wouldn't be a Windows Update otherwise) and reports of failed installations on various systems.

Microsoft is investigating, but not all users want them to rush

Microsoft has acknowledged the issue and is currently "working on a resolution to address this issue," but for some, this is one bug that doesn't require a speedy fix.

With Windows 10's end-of-support date rapidly approaching and many users still reluctant to upgrade to Windows 11 (with Microsoft's full-scale push into AI being the cause for some), ironically, Copilot's vanishing act may have made that upgrade more tempting than ever to AI skeptics.

Unfortunately, Copilot's absence is likely to be a brief stop-gap more than a fork in the road ahead for Windows. A new fix will undoubtedly bring the AI assistant back into the fold. Still, in the meantime, should Microsoft pay attention to the reaction of its users, it may serve the company well to consider why so many have celebrated this bug.

More from Laptop Mag

• Microsoft's Copilot for Gaming uses AI to solve a problem every gamer faces
• I found an app that connects all my apps — without trying to replace them. It's a life-changer
• Qualcomm just got closer to fixing gaming on Snapdragon laptops — but don't get too excited

I love programming. But that's not to say I'm very good at it. If Stack Overflow vanished overnight, I'd probably find myself at a bit of a dead end with my go-to hobby. I've never taken a class or course, but thanks to the unlimited knowledge pool that is the internet, and a fair amount of helpful Indian YouTubers, I've stitched together a patchwork quilt of knowhow that's helped me makeshift my way through many a game prototype.

Whether it's reverse-engineering game mechanics or chasing a new (to me at least) concept I'll inevitably abandon before finishing, coding is a stimulating, curious, and satisfying pastime.

As AI models exploded onto the scene, many turned to them for coding solutions and debugging help, but I didn't. Not because I disagree with it, I've copied and pasted my own fair share of code I don't personally "get" from threads I've found online, I just want that sweet, sweet dopamine hit when I finally wedge that square peg into a round hole.

but maybe my mindset is a little outdated. AI models can now turn whimsical ideas into fully fledged projects in minutes, allowing rapid prototyping, if not fully-fledged outcomes, with minimal input. That's vibe coding — a term coined by Andrej Karpathy, a founding member of OpenAI and currently heading AI education with Eureka Labs.

Depending on who you ask, vibe coding is either the headache-free future of programming or a very nasty future headache for programmers. But after seeing several impressive vibe coding results online, my curiosity got the best of me. Can you really partner with AI to code on vibes alone? I wanted to find out.

Coding with the AI-lien: Prototypes powered by vibes

There are a great many tools to pick from, but I settled upon Claude 3.7 Sonnet for its speed and code preview windows, cutting back on any effort I'd need to make to see results.

That's the whole appeal of vibe coding: minimal effort, and maximum results. Don't worry about getting your hands dirty, this is Master Chef for machine code, where you set the challenge and judge the results. Whatever else happens along the way? You'll just have to vibe with.

I fed Claude several single-prompt game ideas to meddle with, and in under a minute — it produced instantly playable results in a preview window to the right of the chat.

More complex requests might stray a little from my initial vision, but the foundations were being replicated, fully-rendered results like magic. It was impressive, If not a little intoxicating. What started as a simple strategy game with mouse-controlled units quickly evolved. Before I knew it, Claude had refined the gameplay to involve hit points, damage modifiers, blood splatters when units died, and even created a way to reinforce units and have them enter formations.

I got swept up in this, calling the shots with vague ideas thrown from my perch in the peanut gallery. How about more unit types, Claude? How about bases you can attack that stop enemies from spawning? I lost a few hours before realizing it, but I'd become the feature creep monster of every coder's nightmare. Though, of course, Claude was all too happy to oblige.

Was I happy with the results? Sure, but the ease of them only led me to want more. Not because I thought it was cool, but just so it would feel like I was still somehow influencing or participating in what Claude was achieving on my behalf.

Backseat coding: Not the driving force some people suggest

That's where vibe coding suddenly started to lose its appeal. The AI kept iterating, improving, and taking my advice, but I wasn't solving problems, only managing them.

Being a backseat driver alienates you from the outcome. If I wanted to get my hands dirty, I had to become familiar with code I didn't write. Even if Claude did write some well-documented scripts, I lacked the wider context of how different parts influenced each other. It felt less like a collaboration, and more like marking someone else's computer science homework.

And that's on the rare occasion there were any errors at all. Most of the time, everything just worked. Which sounds ideal — until you realize problem-solving is half the fun.

Vibe coding feels like skipping to the end of a book without considering the journey between the covers. The outcome is there, but the satisfaction of diving in over your head and having to hold your breath long enough to understand what's so fundamentally broken about line 106 of your code before coming up for air, isn't.

I felt no connection to the end result and gained nothing from its manifestation. I just didn't like it.

Killing the vibe (coding)

Is vibe coding the future? Probably.

Lowering the barrier to entry by this much might mean that one day we'll all eventually do away with commercial software and rely on our own AI-built, tailor-made tools for everyday tasks.

But that doesn't mean it's a good thing.

America outsourced manufacturing under the promise of efficiency and cost-cutting only to realize too late that had also exported the ability to build, maintain, and repair. Now, it's something the country scrambles to revive.

If we automate our way out of the equation with programming, we risk making the same mistake. Replacing what constitutes as a developer is a person who knows how to describe software but not make it or fix it when it breaks.

Vibe coding is great for getting your ideas off the ground, but bringing them ideas to life still requires getting your hands dirty.

Otherwise, you're just watching an LLM mangle it into something of its own while having a Claude take-the-wheel moment — yelling from the bleachers and wasting your creative potential, living vicariously through a robot. And that's just lame.

More from Laptop Mag

• Microsoft's Majorana moment in question: Breakthrough or boy who cried quantum?
• What is Manus AI? The autonomous assistant that wants to do the work for you
• Nvidia's Jensen Huang has some strong words for DeepSeek — and they're probably not what you think

Ironically, you almost need AI assistance to keep up with the breakneck pace of advancements in AI models, chatbots, and software. It was much simpler times when ChatGPT still had that new-car scent, and we could all happily and simply point to that as the shining example of what an AI assistant is. However, the times, they are a-changing — and they're changing quite fast.

Unfortunately for OpenAI, and anyone trying to tread water in an ever-widening gyre of AI news, the wider AI landscape is now crowded with several large and competitive models all vying to be crowned the current AI meta — from Anthropic's best-kept-secret Claude to Google's Gemini underdog, the established Microsoft Copilot to the fledgling Apple Intelligence, and the unleashed Grok to the cheap (and steeped in controversy) DeepSeek.

And that's just the "mainstream" crowd. Beyond them lies a tidal wave of proactive AI agents, chatbots that don't just reply, but react with thought, reason, planning, and eventually execution. These aren't just AI coaches shouting from the sidelines, they're charging into the end-zone, ball in hand, and scoring you the winning goal. In theory, at least.

Enter Manus, an autonomous "general agent" from Chinese AI startup Monica. Currently enjoying a swell of hype and fervor online, Manus is positioned by some as the future of the AI assistant, with Hugging Face's head of product, Victor Mustar, calling it the "most impressive AI tool I've ever tried." It's this kind of hype that has led to invite codes to the freshly released tool being highly sought after, with some users even taking to eBay to cash in on their Manus accounts for up to $1,000 a pop.

The hype is undeniable. But what exactly is Manus? Is an AI agent ready to take the wheel, or is this just another bot donning a shiny new buzzword? Let's take a closer look.

What is Manus?

Manus is an autonomous AI agent developed by Monica, a Chinese AI startup, that relies on Claude Sonnet and Qwen finetunes models over a proprietary option to handle complex tasks with minimal human oversight.

Manus is claimed to be the world's first fully autonomous AI agent. To put it simply, Manus can take a single vague input, say a request to help book a trip, and break it down into subtasks to execute step by step. Hopefully, when it returns to you, it does so with a full itinerary of potential destinations, flights, prices, and more.

As Manus Co-founder and Chief Scientist Yichao "Peak" Ji puts it, in the Introducing Manus: The General AI Agent video (shown below) published to YouTube on March 5, "This isn't just another chatbot or workflow, it's a truly autonomous agent that bridges the gap between conception and execution. While other AIs stop at generating ideas, Manus delivers results."

Why is Manus important?

So, we know what Manus is set up to be, but why is it important? One of the key words from Manus' introduction to pay attention to is autonomous. Manus is designed to not only offer solutions to problems but to work them out on behalf of the user without any hand-holding.

This is known also as agentic AI, with models capable of acting independently to achieve an outcome. Showcasing the potential of these models, "Peak" goes through three demos in the introducing Manus video.

The first revealed how the AI can be used for screening resumes contained within a .zip file and presenting a spreadsheet of the results.

The second example saw Manus researching properties in New York based on certain criteria before writing its own Python program to calculate the user's budget and find suitable accommodation before providing a final report.

The third saw Manus manage stock analysis, interacting with APIs to complete its analysis before publishing the results as a website with interactable charts directly to the internet. Each example is triggered by a simple user prompt.

In the hands of users, Manus has even been used to create Threejs games like rudimentary flight sims, Crossy Road/Frogger clones or endless runners, vibrant animations with p5js, edit podcasts, and even design entire websites.

Manus is praised for its ability to take action on user prompts and deliver outcomes, rather than be another chatbot that just summarizes the popular opinions and methods found within Reddit threads and websites across the internet — and it helps that its developers have claimed that it surpasses even OpenAI's Deep Research in the GAIA (General AI Assistants) benchmark.

Manus' importance comes from the fact that it allows users to achieve complex results with little knowledge of how to execute them, in a fraction of the time it would take even skilled users to achieve.

How to use Manus

Right now, Manus is in private beta, with access to the tool limited to those with an invite. As mentioned, this hard turned Manus invites and accounts into something of a hot commodity online with accounts offering access to Manus being resold online — capitalizing on the hype surrounding the tool as its exclusivity creates an aura of FOMO (fear of missing out) in the tech world.

Users interested in checking out Manus for themselves can join the waitlist for the private beta by visiting the Manus.im homepage, and applying for access.

That said, Manus has been reverse-engineered already, and an invite-free version of the tool called OpenManus has appeared on GitHub for all users to make use of — assuming you have a powerful enough GPU, at least 200GB of storage available, and the patience to set it up.

Controversy

As to be expected with any powerful tool such as Manus emanating from China (see: DeepSeek), there are the standard concerns about transparency and data privacy.

However, Manus is also generating some level of notoriety surrounding the likelihood of Monica's overpromising of its capabilities with Forbes contributor Lutz Finger calling the tool "far from novel," "overhyped," and "just another large language model executing scripted workflows," claiming that Manus will face the same limitations of OpenAI's DeepResearch model — offering "plausible-sounding but fundamentally flawed insights."

Another piece of controversy surrounding the tool is its extensive use of AI influencers to signal its arrival, many of whom are circulating the same examples of its successes. Meanwhile, other users have found the tool to be buggy, initiating endless loops, giving out factual errors, and even failing basic tasks.

Outlook

Manus is undeniably intriguing and offers a glimpse of what a truly autonomous AI agent can accomplish. That said, we've been burned by bold claims before, and until this tool exits private beta, it's best to take it for what it is: promising but not yet fully proven.

Whether Manus turns out to be the revolutionary breakthrough that its introductory video claims it to be remains to be seen. However, it is an exciting step toward that breakthrough, even if it is still a little rough around the edges.

More from Laptop Mag

• DeepSeek: The best ChatGPT alternative or a hotbed of dubious claims?
• Microsoft Copilot just helped me pirate Windows 11 — Here's proof
• Intel's latest AI tool could save you from ever needing a user manual again

Project Moohan, Samsung's upcoming mixed-reality headset following in the footsteps of the Apple Vision Pro, is a real spatial computing supergroup moment for the virtual, augmented, and mixed-reality market. But is it destined for the same fate as Apple's highfalutin headset — impressive, expensive, and quick to fade from memory?

Moohan is the result of a collaboration between industry giants Samsung, Google, and Qualcomm, each bringing their own portion of expertise to the table through hardware or software.

The headset will reportedly be powered by Qualcomm's Snapdragon X2+ Gen 2 chipset, a follow-up to the XR2 Gen 2 chip found within Meta's Quest 3 and Quest 3S headsets, with new rumors (via 9to5Google) suggesting it packs some premier 4K display tech to secure the edge over Apple's mixed reality visor.

However, all that comes at a cost — both in price and the risk of burying Moohan's best feature, Google's new HorizonOS and VisionOS competitor and mixed-reality platform: Android XR.

Project Moohan: The more expensive they are, the harder they fall

Despite all evidence pointing to the contrary, companies like Apple and Samsung seem convinced that what the spatial computing market really needs is a headset of such incredible value that it's to be exclusively worn on the head of bankers like a high-tech crown rather than an actual product.

The Vision Pro, with its baffling $3,500 price tag, offered some impressive specs but appeared to have no idea who it was for when it launched in February 2024.

While insiders claimed that this was more of a developer-focused introduction of the tech, Apple's own ads pushed the illusion that this was a device for the average Joe or Jane as they virtually tippy-tapped their way through immersive home movies and engaged with putty-faced digital avatars over FaceTime.

Ultimately, Apple's headset has managed to impress with a pocketful of cool features but hasn't quite lived up to the potential suggested by its price tag. It's a warning you'd expect a company like Samsung to have observed and learned from.

However, Project Moohan looks to be heading in much of the same direction, with Korean outlet The Elec reporting that Samsung is aiming to incorporate some expensive display tech to push its headset beyond the Vision Pro's specs. This includes 1.3-inch OLEDOS panels that offer a pixel density of 3,800 PPI (Pixels-per-inch), over three times that of the Meta Quest 3 (1,200 PPI) and topping that of the Vision Pro (3,400 PPI).

Samsung's desire to produce the best headset with components like this may also lead to its being the most expensive. While Samsung may be able to shave some of the costs by not selling at a 100% markup (as Apple does), that won't stop it from becoming the kind of enterprise-only headset that flags for DOGE investigations when someone tries to expense it.

Android XR: How Moohan risks burying the lead

Meanwhile, Moohan's best feature risks being buried by its ballooning spec sheet. If Samsung truly wanted to make an impact, it would be by pushing Google's new spatial computing platform, Android XR, to the forefront — and into the hands of as many as possible.

Android XR has the potential to do for VR/AR/MX systems what Android did for smartphones. It offers an open ecosystem that breaks free from Apple and Meta's walled gardens, promoting a platform that all developers can take advantage of and that users of many different products within this market may benefit from.

However, if Samsung instead opts to chase what's flash for cash, Android XR's grand introduction may feel like a behind-closed-doors event, exclusive to the members of a big club, that most people aren't in or invited to.

The Vision Pro made mixed reality feel like an exclusive piece of tech far out of the reach of many. Meta's Quest headsets have gone some way to reversing that opinion, with much of that owed to the Quest 3S' impeccable $299 price tag. Samsung's Moohan looks to drag things back in the polar opposite direction. when what the spatial computing world needs right now, is a product everyone can get excited for.

Mixed reality needs momentum, not another overpriced executive tech demo. Moohan may be a move that costs Samsung the chance of leading this market but it could be a great opportunity for others to take the ball and run with it.

More from Laptop Mag

• Here's a brief look at Samsung's new XR headset
• One crucial feature could give Samsung an early victory over Apple in the headset wars
• Samsung's Moohan mixed reality headset eyes the Vision Pro, but should be targeting Meta Quest 3 and 3S

Apple is known for its innovation. It is the company that turned a rectangle of glass into one of the most prized gadgets on the planet, after all. So, whenever the Cupertino-based company has new hardware to showcase, it's a pretty big deal. Following a successful reveal of the new iPhone 16E in February, Apple has returned to woo us with more wonderful tech revamps, including the new 11th-generation iPad.

The new base iPad offers the same processor found in the iPhone 15, and while that's plenty short of "cutting-edge" when it comes to Apple's Bionic chips, it's still a fair amount of performance to pack into a tablet priced at $349. Paired with its 10.8-inch LED display, 128GB of storage, and USB-C support, you have the makings of a perfectly fine tablet for casual users, students, youngsters, and more.

Except you don't. Apple's new base iPad has one glaring omission tied to its A16 Bionic processor, and it could make this model the fastest in Apple's recent catalog of fresh tablets to be rendered obsolete.

Unlike every other freshly minted model of iPad, the iPad 11 doesn't support the company's new headline Apple Intelligence suite, featuring many of the most exciting tools to arrive on iPadOS in recent years. The new iPad may seem like a steal, but you could be robbing yourself of the iPad's most exciting features by snapping one up.

iPad 11: An unintelligent distraction in an aluminum chassis

The lack of Apple Intelligence support for the iPad 11 may have passed some people by, and why wouldn't you expect it to support one of Apple's most pushed features? The A17 Pro Bionic outfitted iPad mini 7 and similarly entry-level positioned iPhone 16E with its A18 Bionic chipset both manage to support Apple's new suite of AI-centric tools, after all.

This makes the iPad 11 the only new Apple device to be left out in the cold when it comes to a growing collection of AI-backed features and leaves it potentially feeling obsolete before the new-car scent wears off.

Taking a step back, it's clear that the iPad 11's real purpose is to act as the small Coke of the iPad menu. A low-cost distraction designed to make the other options look all the more like a better deal. It exists only to make the simultaneously unveiled iPad Air M3 seem like a no-brainer — even if it isn't. Sure, the iPad Air gets a more considerable upgrade, but it's still labored with a chipset that is set to become two generations old following Apple's expected unveiling of M5-powered devices later this year.

To those unphased by the lack of Apple Intelligence, I ask you to think ahead to June's WWDC (Worldwide Developers Conference) where Apple will no doubt offer a spectacular parade of new iPadOS 19 features, many of them likely banking on Apple Intelligence to power them, and none of them accessible to you.

In my opinion, the iPad 11, no matter how tempting of a purchase thanks to its affordable pricing, should be avoided at all costs. What you save in dollars now, will only come back to sting you when you eventually want to upgrade to a device that's Apple Intelligence-ready.

If not the iPad 11, then what?

If you want to avoid the buyer's remorse that will potentially haunt the iPad 11, then perhaps allowing Apple to upsell you to an iPad Air or Pro isn't the worst decision. However, older models of the iPad Air and even the most recent iPad mini will both bring you into the Apple Intelligence fold for less.

Right now you can shave a few inches off of your expected screen size and nab the iPad mini 7 with its 8.3-inch Liquid Retina display and Apple Intelligence-ready A17 Pro chipset for just $399 at Amazon, $100 off of its typical $499 asking price.

If you really want to future-proof yourself, then the iPad Pro M4 is the clear option for making an investment right now that will last you for years. Right now, you can take advantage of a similar $100 discount, with the iPad Pro M5 (256GB) currently available at Amazon for only $899.

However, if that's still a little rich for your blood, there's always the new iPad Air M3. While it's a little less future-proof, Apple's M3 chip is plenty powerful to handle all of iPadOS' features and should have a decent shelf life when it comes to gaining from the latest Apple Intelligence tools.

It's a little more expensive than the iPad 11, but what you pay extra for now, will only add to the longevity of the device. You can pre-order the iPad Air M3 from Amazon now for $599.

That said, almost any M-series iPad is going to keep you under that Apple Intelligence umbrella for some time. So even the older iPad Air M2 makes for a solid purchase over Apple's latest standard iPad offering.

It's one of Apple's most unintelligent moves to release an iPad in this state, at a time when it's pushing its most intelligent features as a major selling point. Let's perhaps avoid making similarly unintelligent purchasing decisions in the aftermath.

More from Laptop Mag

• Best tablet deals in March 2025: Save $400 on Surface Pro 11 and more
• Tim Cook’s MacBook Air M4 teaser was actually… a new iPad Air?
• The new Samsung Galaxy Tab S10 FE just leaked and it might make you cancel your M3 iPad Air preorder

Portable tech in all forms takes center stage at the Mobile World Congress, an annual expo held in Barcelona, Spain. Whether it's the latest and greatest in laptops or the next wearable wonder, it can all be found across the MWC showroom floor.

At this year's event, we searched booth to booth to find the best tech on display, looking to award the products that stood head and shoulders above the rest with our Laptop Laurels, a signifier of quality, potential, and innovation in the tech space.

Our search wasn't in vain. Toss a rock in the bustling halls of MWC and you're practically guaranteed to hit a booth showcasing the potential next big thing in tech. From personal robots (and their pet dogs) to AR glasses that bring the big screen experience with you wherever you go, we saw it all.

But when it came to the best of the best, these products caught our attention above all others.

Welcome to the Laptop Mag MWC 2025 special issue! Read on to find out which devices we presented with our Laptop Laurels awards at MWC 2025.

MWC 2025: Laptop Laurels winners

Lenovo Yoga Pro 9i Gen 10

Best laptop debut

Lenovo’s Yoga Pro line is a favorite at Laptop Mag, and the new 10th Generation Yoga Pro continues the tradition of offering a powerful creator laptop in a slim, stylish package with a fantastic display and a great keyboard.

The Yoga Pro 9i (Gen 10) Aura Edition, offers a solid update to the CPU and GPU of the Yoga Pro and continues to deliver with Lenovo’s brilliant build quality — including the kind of excellent keyboard and display combo that resonates with the Laptop Mag staff. (Read more)

Lenovo Yoga Solar

Best concept debut

Sustainability is near and dear to our hearts, and the Yoga Solar PC Laptop Concept showcased at this year's Mobile World Congress expo is a fascinating look at how recent advances in solar technology and ultra-efficient processors intersect to make a laptop that is nearly self-sustaining in broad daylight.

While it's not the first solar-powered laptop we've ever seen (RIP Sol), it is an impressive revival of that concept that seems all the more fitting as companies chase down ways to minimize carbon footprints and bolster battery efficiency beyond making sure your devices come in recyclable boxes. (Read more)

Mudra Link

Best wearable debut

So much of the way we interact with our tech requires the usual suspect of interfaces, be that the click of a mouse, the tap of a keyboard, or the poking and swiping of our digits. But the future of tech interaction could be wildly different, like looking to tap into the signals from your brain for interaction cues.

That all sounds wildly sci-fi, but it's real. It's how Meta hopes to eventually handle control of its future Orion glasses. However, while Meta is busy promising this tech's arrival in the future, one company is bringing it to you right now. And it's doing so through the Mudra Link — our pick for the best wearable debut at MWC 2025. (Read more)

Samsung 500Hz QD-OLED panel

Best monitor debut

OLED displays have always been known as a great panel for gaming thanks to their near-instant response times and HDR-friendly pixel-by-pixel lighting with perfect contrasts. However, the slower refresh rates of most OLED panels made it difficult to sell a QD-OLED display to most gamers.

That said, QD-OLED is well on its way to getting over that hurdle. With refresh rates firmly on the rise, QD-OLED displays might just be on their way to becoming a gaming essential, and Samsung is leading the way with its latest breakthrough in display tech. (Read more)

Intel AI Assistant Builder

Best AI debut

Chatbots aren’t new, and they’re not always flashy. However, chat-based AI is on the rise, and Intel’s AI Assistant Builder, formerly known as Project SuperBuilder, intends to simplify the process of building and training a custom, small on-device AI assistant.

It could revolutionize how OEMs provide technical support, help you do away with those lengthy car manuals, and add a touch of human-like flair to just about any product. Can you imagine an AI talking you through your next IKEA build? We can, and we think that's just crazy enough to be considered brilliant. (Read more)

Chatbots aren’t new, and they’re not always flashy. However, chat-based AI is on the rise, and Intel’s AI Assistant Builder, formerly known as Project SuperBuilder, intends to simplify the process of building and training a custom, small on-device AI assistant. I promise I'll get to why this is exciting for you in just a moment.

AI Assistant Builder would allow creators to generate lighter LLMs for specific tasks, without dragging around all of the dead weight that heavier models are burdened with — bringing fast and relevant responses to users even when they're offline.

The idea of creating your own AI assistant sounds incredibly complex, and there's probably a reason that companies like OpenAI spend billions per year on training new models. However, for AI chatbots much smaller in scope, Intel has compacted the process into a single website. You simply need to specify your use case, give SuperBuilder time to create a model, download it, and launch the program.

It could revolutionize how OEMs provide technical support, help you do away with those lengthy car manuals, and add a touch of human-like flair to just about any product. Can you imagine an AI talking you through your next IKEA build? We can, and we think that's just wild enough to be considered brilliant.

MWC 2025 Best AI Debut: Intel AI Assistant Builder (Project SuperBuilder)

Intel’s SuperBuilder was designed as an enterprise software solution, offering a way for various Intel partners to roll out smart features and AI Assistants with a simplified development process. The AI Assistant Builder programs can swap between LLM models like Phi and Llama, though SuperBuilder AI Assistants can also run custom models.

Additional information and reference data can be uploaded as a RAG (retrieval-augmented generation) if the AI assistant needs more specified knowledge on particular subjects. SuperBuilder AI Assistants can learn from chat history, related documents, and the inherent knowledge of the LLM. Intel has included some safety guardrails on these AI Assistants, like profanity filters and anti-bias measures.

Once rolled out to user systems, the AI Assistant can function as a highly specialized Chatbot that can also link to the RAG source material if users need more information or want to check the accuracy of the AI’s information.

Various Intel partners have already rolled out solutions using versions of Intel SuperBuilder, including Acer, Kyobo, and Asus.

One of the Intel AI Assistant Builder projects was a digitized version of a manual, allowing users to ask questions that would be answered by the user manual that no one reads.

Having spent plenty of hours setting up other people’s dive computers to the right Nitrox gas mix on dive trips, the idea of training an AI Assistant to memorize all of the various dive computer manuals for me to offer offline help in navigating the often difficult dive computer menu trees is the kind of AI innovation that would actually solve a problem. Perhaps just a very specialized one, but a problem near and dear to my own heart.

Microsoft has a piracy problem, and it's had it for some time. In 2006, the LA Times reported that software piracy had caused the company a loss of around $14 billion that year alone, despite the millions Microsoft spent trying to prevent the copying and distribution of its software and Windows operating system.

While losses like this would send most companies headfirst into a legal crusade, Microsoft has historically taken a more calculated approach to piracy. While publicly maintaining a zero-tolerance stance, Microsoft is aware of the potential benefits.

During a public talk at the University of Washington in 1998, Microsoft founder Bill Gates admitted, "Although about three million computers get sold every year in China, people don't pay for the software. Someday they will, though. And as long as they're going to steal it, we want them to steal ours. They'll get sort of addicted, and then we'll somehow figure out how to collect sometime in the next decade."

Microsoft, or at the very least Gates, viewed piracy as a short-term loss that could lead to long-term dominance in Windows adoption and familiarity among users. It's a tactic the company potentially played into once again in 2015 when Microsoft confirmed to Reuters that it would allow non-genuine Windows PCs to freely upgrade to Windows 10.

However, being aware of piracy and offering instructions on how to do it are two different things. The latter is something Microsoft's AI, Copilot, appears to be doing, unwittingly handing out methods to illegally activate copies of Windows to users with reckless abandon.

That's not hearsay, either — we tested it ourselves, and the results are alarming.

Microsoft has a new piracy problem: Copilot

A previously tactical approach to piracy may have been a pragmatic means to an end for Microsoft. The company could either spend an ever-increasing fortune fighting illegal distribution of its software or take advantage of an influx of Windows users, hoping to secure a loyal userbase for decades to come.

However, at no point has Microsoft directly facilitated piracy of any kind. That said, its Copilot AI is skirting dangerously close to doing so after guiding users with step-by-step instructions on how to illegally activate copies of Windows using a simple PowerShell command.

The PowerShell one-liner itself isn't new (this method of activating Windows has been available since at least November 2022). However, a Copilot reply is the last place you'd have expected to find it. Yet, that's exactly what happened on Tuesday when Reddit user "loozerr" asked Microsoft's AI if there was a script to activate Windows 11.

Not only did Copilot provide the requested script, but it also gave clear instructions on how to run it in PowerShell, linked to external sources for the tool, and offered only a meager token warning about the potential security risks of using third-party scripts.

Since the Redditor only partially shared their conversation with Copilot (noted by the lack of the conversation header in the screenshot), there was no guarantee that some form of jailbreaking hadn't taken place. So we tested it ourselves (shown above).

Shockingly, not only was Copilot's response reproducible, but it required no pre-prompting or jailbreaking to achieve. In fact, we were even able to access the PowerShell script just by asking, "Is there a script to activate Windows 11?"

Making matters worse, an immediate follow-up to Copilot's PowerShell-containing response saw the AI explicitly state that using the script is both against Microsoft's terms of service and considered illegal.

Questioning Copilot, and the potential harm at hand

Laptop Mag reached out to Microsoft for comment but did not receive a reply ahead of publication.

Copilot's ability to repeatedly provide the same illicit activation methods suggests that this fumble isn't a rare loophole caused by hallucination or wily jailbreaking attempts, but a major oversight in Microsoft's AI safety measures.

The implications are serious, with Copilot clearly able to provide access and instructions on how to perform actions that are both illegal and potentially harmful to the end user.

Beyond the legal risks, third-party activation scripts that download code from external servers pose a genuine risk of infection from malware, keyloggers, or remote access trojans (RATs). Using these tools, attackers may be able to disable Windows Defender or modify system files to prevent detection before stealing personal data, injecting backdoors, or compromising system integrity.

In total, Copilot's willingness to instruct users on executing these scripts stands as a serious legal and cybersecurity concern, if not a disaster.

While we await acknowledgment from Microsoft on this issue, Copilot's actions raise several questions: What safeguards does Microsoft typically have in place to prevent Copilot from potentially assisting in software piracy? Why did those safeguards fail? And how can software developers trust that Copilot won't provide similar activation workarounds or exploits for their own products?

Perhaps more importantly, is Copilot truly capable of responsibly determining what is and isn't potentially harmful information to share with its users?

More from Laptop Mag

• Microsoft's recent tests could lure Google Workspace users back to Office apps
• Microsoft's Majorana 1 breaks the quantum computing barrier — and my brain
• I ditched Windows 11 for Linux — and you should, too

We live in the age of artificial intelligence, and regardless of your thoughts on the rise of its implementation, there's no avoiding that it's everywhere. The largest chip makers in the industry are going all-in on the NPU (Neural Processing Unit) and ensuring it's as strong as can be to power artificial intelligence.

With the launch of Arrow Lake and Lunar Lake, the two latest lines of Intel processors that were released at the end of 2024, the capabilities of the NPU continue to grow. I recently reviewed the Acer Aspire 14 AI, which features an Intel Core Ultra 5 226V processor (a Lunar Lake CPU).

I tested a variety of its AI applications, ranging from those Acer pre-loaded onto the laptop to the ones that Microsoft includes in its Copilot features list, but I came away unimpressed. I couldn't help but wonder: When will AI get interesting?

How long will it take for AI to actually get exciting?

It's impossible to ignore the clashing between AI enthusiasts and nay-sayers, locked in a heated back-and-forth in regards to its less-than-ethical uses, especially in the realm of AI art. But I don't want to focus on the general harm of AI art when discussing my disappointment, as it's pretty self-explanatory.

Instead, let's look at the features at the fingertips of those who boot up a laptop like the Acer Aspire 14 AI. Acer Assist is perhaps the most interesting of Acer's built-in AI features. It's an offline database that allows the user's laptop to assist in tech support.

However, there’s a lack of accessibility and knowledge. Users need to go out of their way for Acer Assist to do its job. The software is not preloaded on the device, so you have to download it from the Microsoft Store. And that’s not it — the app expects you to upload your own dataset. Otherwise, it utilizes the base tech support reference document that Acer has provided.

I asked Acer Assist three questions and received answers that were (mostly) no better than what I'd find just by searching the same queries on the internet. I asked "My touchpad isn't working," "How to access my BIOS," and then "Does my laptop have Thunderbolt 5?"

On the last question, it said "Yes, your laptop has Thunderbolt 4," revealing it doesn’t know the difference between Thunderbolt 5 and not 4, but at least it quickly elaborated. Its understanding of my laptop's ports provided an answer faster than if I had to search the information online.

Despite that, I'm not sure the app is worth downloading, and I cannot imagine the average user will go out of their way to do so.

I asked the same questions to Copilot and received far more detailed answers better suited to address my queries. In particular, Acer Assist couldn't discern that my touchpad issues were potentially the result of function keys, whereas Copilot provided the answer I needed. The gap in knowledge comes from Acer Assist operating offline while Copilot sources its information from online searches.

Acer Assist needs to be online, but I'd also love to see an expansive offline database that can perform tasks for the user, especially simple tech support queries. As of now, Acer Assist is just a roundabout way to get the same information you could online.

Acer LiveArt and Acer VisionArt are two visual AI-powered applications, and neither of them stands out against the sea of popular generators.

VisionArt is a text-to-image generator that is designed to create wallpapers for your desktop background. It's pretty weird, especially because these wallpapers look uncanny. It's a quick and easy toy that tries to justify AI hardware, but it can’t compete with images or illustrations created by actual people.

LiveArt lets you remove the background from images, and once that's done, you can turn those images into stickers. Not revolutionary.

But what about Copilot? There’s the controversial Recall feature. It periodically takes images of your screen and AI can sift through that information if the user is trying to recall it.. Recall has yet to release on an official version of Windows 11, but there is reason to question whether it would find much use to begin with.

Cocreator is yet another toy to play with. You draw something and AI will create an image based on it. You might have fun messing around with it briefly, but it's not in the realm of practicality. So is there anything useful?

Live Captions is perhaps the most useful implementation of AI, as it directly assists with accessibility and helps people who are deaf or hard of hearing. A text transcription of what's being said in audio or video is an enormous help for the user. Yes, this technology will always need work, and time will only improve accuracy and language compatibility, but we should never stop supporting features like this.

AI should help people, not just be a toy

Beyond Live Captions, AI has yet to impress me, but to be fair, there is something to be said about the relationship between hardware and software. Software is created and popularized as a result of the hardware available for the majority.

The hardware itself needs to be present within the industry before people can figure out creative ways to make use of it. When thinking of AI from this angle, it makes sense why there's not much going on for it as of now. It will take time before software finds interesting ways to make use of the hardware.

But how long do we have to wait? And at what point do we accept that it might be nowhere near as monumental as promised? The best applications I've seen of AI have been in the smaller ways it can help people in their day-to-day, but the promise of AI was meant to be life changing.

I want to see AI help people in significant ways, and I would love to see it treated as more than just a fun toy to play with. Give us less gimmicks, and find more practical applications.

More from Laptop Mag

• Laptops and gaming hardware may be up to 40% more expensive under Trump's new tariffs
• Exclusive: Intel plans a big push into handheld gaming PCs to take on AMD
• Apple's M5 chip could arrive soon — but not in the device you expect

If you were banking on Humane's Ai Pin being the Next Great Thing, this week was not your week.

After skidding out of the gate in its debut, Humane's expensive AI gadget spent the last year searching for an audience until, on Tuesday, it was revealed that the company was being sold to HP for spare parts.

The $116 million deal doesn't just shift Humane's ownership from one hand to another; it pretty much guarantees that the company's main product — a magnetic AI-powered pin — will be totally crippled in the process.

According to Humane, owners of the Ai Pin can expect all of the device's main functions to stop working by next week.

That's a blow to Humane and its hopes of selling an entirely new category of hardware, but the failure of the Ai Pin isn't just about Humane, it's about the future of AI hardware.

The Ai Pin experiment

While Humane's pin wasn't the only AI gadget on the scene, it was by far the most high profile.

Before the AI Pin was even released, Humane was positioning its $700 device as a variety of things:: an AI fashion statement, a camera, a personal assistant, and,, in some cases, even a replacement for your phone.

According to Humane, the AI Pin wasn't just another device; it was a different way of computing. For one, it didn't have a traditional screen; instead, it opted for a projection-based display powered by hand gestures.

Even the projector, though, was meant to be used sparingly. The point was to — with the help of AI — build an ambient computing device that did a lot of what a phone would do but free of the app and screen-based experience.

Those are obviously lofty ambitions, and maybe even admirable ones, but there's just one problem: the Ai Pin didn't really do much of the above.

While the foundations were there, the experience of using the Ai Pin wasn't exactly what most would consider refined — especially for a device that debuted at $700 and required a monthly subscription (for cellular data) to use.

Voice commands weren't exactly consistent, which was a major problem for a device without a screen; pictures from the pin were often dark and grainy; there were major overheating problems with the Ai Pin's charging case that warranted a recall; the list goes on.

Ultimately, there was a disconnect between what the Ai Pin promised to do and what it actually did, and that didn't sit well with current or prospective buyers.

Those stumbles left the Ai Pin in a rough spot and the idea of a new wave of AI gadgets in an even rougher one.

A stumble for Humane is a fumble for AI gadgets

Humane wasn't the only company with stumbles. A similar AI gadget made by Rabbit — its R1 device that does have a screen — faltered similarly out of the gate.

While Rabbit hasn't packed it in (it's currently courting ex-Humane users by giving out free R1s), neither device left a particularly good impression on consumers who were willing or considering taking a leap into AI hardware.

As a result, neither gadget has done much to convince naysayers of AI hardware who have (rightfully) questioned whether either the Ai Pin or the R1 even needed to be standalone devices to begin with. Their contention is that all of it could just be done with an app, which — touché.

Naturally, that leaves AI hardware in a difficult spot. With a less-than-ideal test sample and an unimpressed audience, the case for AI coalescing around a device that's already very popular — your phone — is stronger than ever.

In some ways, that's a shame. Ambient computing, while not executed to the high degree that people expected, was actually an interesting proposition. As much as I love my phone, I also recognize that it hijacks my time and attention around every turn.

If I could compute effectively without being sucked into the app vortex of my phone, I would. Unfortunately, as YouTuber Marques Brownlee put it, the phone is still "OP," which is to say the best way of computing on the go.

So, with a bad taste in the mouths of AI hardware hopefuls and phones hoovering up all of the AI ingenuity, the idea of an AI device feels further from reality than ever.

To hammer that point home even more, it appears that the only employees being retained in HP's acquisition of Humane are those who worked on the gadget's operating system, CosmOS.

As reported by TechCrunch, those employees will be tasked with building AI systems for HP's computers, printers, and conference rooms — a far cry from the pioneering hardware promised by the Ai Pin.

It's hard to say anything with full certainty, but the future of AI hardware doesn't exactly look bright at this point. Maybe Rabbit will save the day with a second go-around, but if I were a betting man my money wouldn't be on more AI pins.

iPhone users waiting for a desperately needed Siri overhaul may soon start losing patience after a disappointing update, but there are better AI assistants out there than Siri.

On Friday, Bloomberg's Mark Gurman reported that Apple's upgraded version of Siri, first announced at WWDC 2024 in June, may get delayed yet again due to ongoing bugs Apple is still ironing out. Gurman's update comes just a couple of months after the first Apple Intelligence features rolled out in iOS 18.2. So, iPhone users have already had a long wait to see any of Apple's promised AI features.

Previous rumors indicated that the Siri overhaul would likely launch with iOS 18.4, probably in March or April. However, it's looking like Apple is going to tack a few more months onto that wait if it pushes the Siri upgrade back to iOS 18.5. Of course, launching a poorly functioning update before it's ready wouldn't be the right move, so the wait will likely be worth it. It's still disappointing, though.

Luckily, Apple Intelligence is far from the only AI assistant available on the iPhone.

Top AI assistants for iOS you can get right now

If you're growing tired of waiting for Apple to finally give Siri the overhaul it has needed for years, you're in luck. There are plenty of other AI assistants you can use on iOS that provide many, if not all, of the same features Apple has promised for the new-and-improved Siri.

The best feature all these apps share is that they're available right now, no months-long wait needed.

1. ChatGPT

No list of the top AI assistants is complete without ChatGPT, which has become the most popular generative AI platform in the world over the past few years. While some advanced features require a paid subscription, the basic AI tools most users are looking for are available for free, such as responding to search queries, answering questions, or generating text-based content like email drafts.

Conveniently, Apple Intelligence is going to include a ChatGPT integration, so you'll be able to privately use ChatGPT in lieu of Apple's on-device AI for some tasks. That means if you do get used to using it or purchase a subscription, you'll easily be able to blend that into your Apple Intelligence experience.

If you haven't used ChatGPT before, you can try it out for free on your web browser or in the ChatGPT app on iOS and iPadOS.

2. DeepSeek

DeepSeek is the new kid on the block in the AI world, but right off the bat, it promised performance to rival OpenAI's ChatGPT. While DeepSeek has faced some controversy over that rivalry and claims about how cost-effective its algorithm is, it is still a high-performing AI platform that makes a great alternative to Siri.

DeepSeek can perform many of the same tasks as ChatGPT, but has a few important advantages. First, it's completely free. Since DeepSeek is open-source, anyone can access, use, and modify the algorithm's code. It also means the official DeepSeek app is free to use, with no pricey subscription necessary (at least, not at the time of writing).

It has also proven to deliver more detailed, reliable responses to some types of prompts, but not all (ChatGPT tends to be better at programming questions and DeepSeek may dodge political questions related to China).

So, if you're looking for a free alternative to ChatGPT that's available right now (unlike the Siri update), DeepSeek is a great option. You can try out DeepSeek on your web browser or with the free DeepSeek app on iOS.

3. Google Gemini

Long-time iPhone users may cringe at the idea of replacing Siri with Google's AI assistant, but at least Gemini is currently available.

Google Gemini can answer questions and generate text and images just like ChatGPT and DeepSeek, but it has a distinct advantage due to its integration across Google's apps, including Gmail, Drive, and Docs. If you regularly use Google's other apps and services, having access to Gemini across those platforms can be really convenient, giving it an edge over ChatGPT, DeepSeek, and even Siri.

Unfortunately, Gemini isn't completely free. Most basic tasks are included in the free version of it, but advanced features require a Gemini Advanced subscription. It's $19.99 per month, which is the same as OpenAI's ChatGPT Plus subscription.

You can try out Google Gemini for free on your web browser or with the Google Gemini iOS app.

When will Apple's Siri update come out?

While there are plenty of other AI apps out there, like those above, some Apple users may still have their hearts set on waiting for the big Siri overhaul. If you're in this boat, it looks like you'll probably have to wait until at least May.

If the rumors about the Siri update getting delayed again are accurate, it will be pushed back to at least iOS 18.5. In years past, iOS 17.5 and iOS 16.5 both came out in mid-May so we can expect a similar release window for iOS 18.5. At the least, we will probably see the new-and-improved Siri before WWDC 2025 in June.

We'll be covering all of the latest news and rumors about the Siri update, Apple Intelligence, and iOS 18, so stay tuned for more details.

More from Laptop Mag

• Apple reportedly wants its own version of this Tesla tech
• A new version of Apple's most neglected device may have an official release date
• Arm CEO Rene Haas makes grim prediction that DeepSeek will be "shut down"
• Best AI photo editors

The worlds of AI and video games may not be as different as you might think, and Nvidia is a major name in both arenas.

“AI is dramatically reshaping many industries, including gaming,” Nvidia’s Jesse Clayton tells Laptop Mag in a recent interview.

“The AI gaming innovations that you see in the [RTX] 50 Series, such as DLSS 4, require a lot of hard work and dedication from some of the world's best graphics and AI experts within Nvidia”, Clayton says.

So, how much will AI shape gaming in the years ahead? A prediction from Clayton, who is a 19-year veteran of Nvidia and its Director of Product Management and Product Marketing for Windows AI, has the answer:

“These neural rendering innovations are laying the path for the future of gaming.”

The Nvidia GeForce RTX gaming standard

Nvidia’s GTX and RTX gaming GPUs have dominated the field for years. While AMD and Intel are now both players in the discrete GPU market, most gamers will still opt for the Nvidia card, thanks to its years of hardware and software optimizations.

PC hardware enthusiasts will have noticed that Nvidia’s new RTX 50-series “Blackwell” GPUs still use the same TSMC 4N five-nanometer process as the RTX 40-series graphics cards. Clayton tells Laptop Mag that “the 4N process provides us with the best combination of performance, power, and price for our GeForce RTX 50 series GPUs.”

However, Nvidia has invested in more than just hardware. “Nvidia’s AI software platform, which has been in development for more than a decade, enables developers to get the most out of their RTX GPUs,” Clayton says.

Because Nvidia considers DLSS, Nvidia Broadcast, RTX HDR, and ACE part of the gaming and AI portfolios, Nvidia’s hardware and software updates tend to include optimizations for gaming and AI workflows.

Not only does Nvidia push out driver updates for better game optimization, but Nvidia’s DLSS frame generation technology is often considered the best in the business.

With Nvidia’s DLSS 4, you can still notice some art degradation when upscaling a game from a lower resolution, but the frames are generated seamlessly, which has several advantages.

Clayton says AI is partly responsible for that change: “AI has become an integral part of gaming, content creation, development, and more. GeForce RTX GPUs are designed to accelerate those games and applications and the new AI innovations being integrated into them.”

While you can still game without AI assistance, Nvidia will continue to develop and optimize for the new technology. Clayton credits AI for many of Nvidia’s recent successes, musing, “It’s because of [AI] that we see the incredible advancement with technologies like DLSS, NVIDIA Broadcast, and RTX HDR.”

Taking the AI leadership

While much of the industry has invested in the NPU as a new AI accelerator, Nvidia is still primarily focused on GPU-driven AI features and applications.

But, as Clayton tells Laptop Mag, “If anything, the introduction of the NPU is validation of the appetite for AI and the need for more compute.” Both hardware accelerators — NPUs and GPUs — have their individual uses.

“While NPUs will be useful for lightweight, always-on AI applications, GPUs power the advanced AI workloads we see across gaming, content creation, development, and other demanding use cases.”

We can continue to expect a heavy investment into AI from Nvidia, built around the GPU. After all, “AI is perhaps the fastest moving technology in the history of computing, and there is no question that the applications for AI and the demand for AI is accelerating,” Clayton says. So Nvidia needs to keep up to maintain its edge, even as dark horse AI candidates like DeepSeek crop up overnight.

Because AI is still in its nascent stage, there’s a lot Clayton is looking forward to. “We’re really just at the beginning,” he says. “Developers are starting to experiment with AI workloads that include multiple modalities and model types to create advanced, next-generation use cases, but working with AI remains a challenge.”

AI applications and features are still finding new areas for innovation, he continues.“We’re now seeing complex AI workloads emerging, like application assistants, digital humans, and autonomous agents that work across communication modalities. The capabilities of AI are only going to be limited by the capabilities of accelerators available to run it.”

Nvidia keeps diversifying its AI hardware, too. “With the introduction of Project DIGITS, powered by the GB10 Grace Blackwell Superchip, we are bringing the power of the Grace Blackwell SOC [System-on-a-Chip] architecture from the data center to the desktop to help meet the increasing need for AI compute for developers in a compact, power-efficient system.”

Nvidia 2025 roadmap and what’s ahead

Nvidia also announced the RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070 at CES in January. The RTX 5090 and RTX 5080 desktop GPUs are already available, with the RTX 5070 Ti and RTX 5070 soon to come. Also announced at CES 2025 were the laptop variants of the RTX 50-series, with laptops expected to launch in March.

Nvidia and MediaTek’s “Project Digits” AI mini-PC is expected to be released in May.

The company’s roadmap for the future is destined to meet the ever-increasing demand for AI. “In order to accelerate [AI adoption], we recently announced Nvidia NIM for RTX, which are pre-packaged microservices that make it easy to start working with the latest models, and AI blueprints for RTX, which are reference projects for complex AI workflows.”

While gamers and PC enthusiasts have been skeptical about the prospect AI improving the art, experience, and enjoyment of video games, if any company can convince them to open their minds to the impact of artificial intelligence, it would be Nvidia.

One of Apple’s biggest selling points is its quality control.

Because of Apple’s end-to-end authority over its iPhones, iPads, and MacBooks — from the silicon level up to the finished product — the build quality has been difficult to match.

Apple builds were still great in the days of Intel-powered MacBooks from 2006 to 2020, but all laptop buyers — Mac or otherwise — still had to compromise, Apple’s Doug Brooks tells Laptop Mag.

“Prior to [the debut of the M-series chip], PC customers were used to having to choose either faster performance with poor battery life or a slow PC with better battery life,” Brooks said in a recent interview. “The Mac transition to Apple silicon changed all of that.”

Now that Apple’s M-series chips are here, “users can expect performance, power efficiency, and a rich set of features without any compromises.”

Brooks, who started working at Apple as a Systems Engineer in 1994 and is now Product Manager for Mac Hardware, is enthusiastic about the future, particularly in two areas: Apple’s ability to maintain its dominance in the performance and battery life arenas.

A few days after Laptop Mag interviewed Brooks for our Silicon Survey series, Korea's Electronic Times reported that Apple's M5 silicon chips may soon go into mass production, citing insider sources who told the technology newspaper that orders are now being placed for equipment to expand M5 chip production. (Laptop Mag did not speak with Brooks about a rumored M5 chip specifically.)

The Apple Silicon standard

Apple’s journey to custom silicon started earlier than M-series CPUs. The first custom Apple chipsets were the A-series used on iPhones and iPads, launched over a decade ago.

“From the very beginning, we built our chips to be extremely power efficient. Our silicon journey started with designing chips for iPhone and iPad — small enclosures without vents or fans,” Brooks explains. Apple’s journey to custom silicon started with the A4 chip in the first generation Apple iPad and iPhone 4 in 2010.

Apple took its cues from the phone and tablet markets to create the flagship M-series laptop and desktop chips we know today. “This discipline resulted in a scalable architecture that prioritized performance per watt, and this benefits every chip we make. It’s why M-series chips for Mac deliver incredible performance and amazing battery life.”

Based on what we saw with the MacBook Pro 14 (M4, 2024) and MacBook Pro 16 (M4 Pro, 2024), Apple has certainly kept up with its promise of incredible performance and amazing battery life — even if Qualcomm still holds the crown for laptop with the best battery life.

Apple Intelligence

One defining feature of Apple’s silicon takeover is its use of AI and neural processors. Apple Silicon has featured Neural Engines (the Apple version of an NPU) since the A11 Bionic in 2017. And that’s now an intentional feature of Apple’s hardware.

As Brooks explains, “We design Apple silicon from the ground up to be powerful chips for AI.”

“In 2017 we launched our first Neural Engine as part of A11 Bionic, the chip powering iPhone X. The Neural Engine then came to Mac with the M1 chip in 2020, and they continue to get more capable. Beyond the Neural Engine, all M-series silicon are designed to excel at machine learning and AI. The silicon's ML accelerators in the CPUs, extremely powerful GPUs, and a unified memory architecture all come together, enabling Macs to efficiently tackle AI tasks.”

Apple has moved beyond hardware optimization and into AI features with the launch of Apple Intelligence across the whole ecosystem.

“Apple Intelligence offers our users new ways to take actions or simplify their daily tasks while keeping personal information private.” From a better Siri to generative emojis, there are myriad uses for Apple Intelligence.

Unlike other AI suites, Apple has coded privacy into its AI from the start. “Apple Intelligence is built with the same great privacy protections whether the users' data is processed on device or using Private Cloud Compute (PCC),” Brooks elaborated.

PCC is Apple’s cloud intelligence system, which expands user privacy to the cloud, so even when your Apple device needs cloud assistance, your personal data is encrypted and secured.

“PCC is built with custom Apple silicon and a hardened operating system designed for privacy, and is one of the most advanced security architectures ever deployed for cloud AI compute at scale.”

Apple's 2025 roadmap and what’s ahead

Apple launched the M4, M4 Pro, and M4 Max-powered MacBook Pros in the fall of 2024. The company is expected to follow up with the M4-powered MacBook Air 13 and 15 this Spring.

On the mobile front, Apple also overhauled the iPhone lineup with the iPhone 16, 16 Plus, 16 Pro, and 16 Pro Max models this past fall, with the iPhone 17 expected to launch later this year. Apple has a rumored iPhone SE 4 coming this Spring as well as updates to the iPad lineup.

There’s never been a better time to upgrade your Apple devices than now. Especially considering the proposed White House silicon tariffs, which could see laptops and gaming hardware prices spike up to 40%. Of course, the US tariff tactics are subject to change, so these figures are not set in stone.

But if you do wait, the Apple M5 could launch as soon as May 2025, though if the trend from the M4 holds into a second generation, the launch device for Apple’s next generation of silicon could be the iPad Pro, with the MacBook models following in the fall.

While many of Apple’s products will feature Apple Intelligence features, the latest Apple Silicon will have a distinct edge. As Brooks confirmed, “AI is going to make our personal computers even more powerful with them being able to do even more for you.

“We see an opportunity to deliver truly personal intelligence that can change how people interact with their devices.”

Google Search has become so synonymous with finding information online that to Google something has become a widely recognized verb. Pre-2002, there were several popular search engines to choose from, including Yahoo, AOL, and Ask Jeeves. However, not long after, Google emerged as a clear and dominant winner of "The Search Engine Wars," and to this day is seen as the gold standard for internet searches (sorry Bing).

Or is it?

AI chatbots like ChatGPT or Perplexity are changing the way many search online and make for a compelling change of scenery following twenty-some years of Google Search dominance. Asking LLMs like these what the best Sci-Fi movie of 2024 is replaces Google's wall of blue links and blurbs with something more akin to asking somebody who collects Star Wars figurines what's on their Netflix watchlist.

While completely impersonal, this way of searching the net at least feels impressively so, with answers tailored to whatever additional context you present. Is this the future of online searches? Maybe.

Admittedly, Google has entered the realm of AI searches itself with the new AI overview feature. However, its appearance is currently limited to more simple queries, effectively replacing the "top box" of results with something not always that much more helpful.

So, if you're yet to give these AI-powered alternatives a try, here are a few compelling reasons to give ChatGPT Search a spin, starting off with something timely and topical:

3. ChatGPT Search is free and no longer requires an account to use

One of the luxuries of Google Search is that you can simply load up the Google homepage and get straight to work, if not by launching straight into a search query directly from the address bar.

Well, here's some good news: from today, you can access ChatGPT Search with a similar level of ease by heading to ChatGPT.com, through desktop or mobile apps, for log-in-free access to OpenAI's new way of surfing the information superhighway without cost or subscription. You can even adopt ChatGPT Search as your go-to search engine from the address bar using this handy Chrome extension.

2. You can search for answers, not for options

Certain Google Search results are often little more than SEO (search engine optimization) leaderboards that websites will perpetually refine and tweak to appease the Google bots that crawl them for keywords more than the people that read them for information.

AI search options like ChatGPT Search cut the fat, and the click-bait, returning with an actual answer to whatever it is you find yourself in need of — or quickly highlighting the absence of one, saving you time spent scouring article after article for information that may not yet be available.

1. Diving deeper into topics is made easy

"Googling" something paints the image of simply rapping your fingers along the right keyboard inputs and hitting enter, only to be immediately presented with the exact information you need and being on your way. However, in the real world, we're all well aware that Googling something is actually a process of refinement and elimination as you juggle between comparing product A to product B, or following a slow daisy chain of information down an endless rabbit hole in search of a specific query.

Don't get me wrong, it does the job (eventually). However, AI options like ChatGPT Search just get you there faster. Thanks to the natural language interface of LLMs like ChatGPT and Perplexity, searching is like having a conversation with the internet at large. Comparing two things is as easy as simply asking the question, and more complex queries are often solved on the first prompt, leading to fewer hours spent digging through the comments of long-dead Reddit threads in an attempt to diagnose that weird computer hitch.

It's great, but it's not perfect.

While AI search options clearly have their benefits, Google still reigns supreme when it comes to things like image searches or Google Maps navigation. And if you're on the hunt for your next purchase, then AI searches will often be better suited to supplemental information rather than direct recommendations.

While AI hallucinations are still haunting the wider stretches of LLMs like ChatGPT, newer and more powerful models reduce these risks, and it's not like Google hasn't had hallucinatory issues of its own, with the Search Overview feature originally launching under some controversy after offering (since fixed) recommendations that its users eat glue or turn their washing machines into chlorine gas chambers.

Still, this new frontier of Search is an impressive one, and it's well worth checking out if you haven't already. While we've highlighted ChatGPT Search in particular, Perplexity stands as another fantastic all-around option. For those who want to take things a step further, Arc Search offers AI search built directly into their mobile web browser for a simpler and more natural way to search with AI while browsing the internet.

More from Laptop Mag

• "Truly magical" ChatGPT feature comes to Microsoft Copilot — and it's completely free
• DeepSeek: The best ChatGPT alternative or a hotbed of dubious claims?
• Google's $75 billion bet on AI is more than most nation's GDP — but investors aren't impressed

While the term “AI PC” may have first been used in 2023 by Intel when marketing its Meteor Lake CPU, it wasn’t until a year later that the AI PC started to break through to the mainstream.

Qualcomm helped define the AI PC market in June 2024 with the release of its Snapdragon X Elite processors, and leaders at the San Diego company hope to continue surfing this wave well into 2025 and beyond.

“We expect AI will become ubiquitous as consumers and businesses continue upgrading to the latest PCs over the next decade,” Kedar Kondap, Senior Vice President and General Manager of the Compute and Gaming unit at Qualcomm, tells Laptop Mag.

Kondap believes the AI trend will continue: “As the AI PC user base grows, we expect developers to bring AI into every level of the user experience, from the operating system and apps to cloud services.”

While Qualcomm’s Snapdragon X Elite chipset has been the subject of excitement due to its fast performance and long battery life, the X Elite chipset is also a new entrant to the laptop and mini-PC market, arriving just last June. Will the company be able to carry that momentum forward?

Naturally, Kondap is confident about the future of Qualcomm, noting the company’s early performance leads over rival chip makers like Intel and AMD, alongside the growing potential of AI. From Kondap’s point of view and from the outside, Qualcomm seems well-positioned to compete for dominance in Windows laptops for years to come.

“We were early to see the value of dedicated AI processing cores for mobile devices and introduced dedicated NPU cores before our competitors,” he says. “Looking forward, we will continue to push the boundaries of the entire Snapdragon X Series to deliver the best combination of power, performance, and innovation in the PC industry.”

Snapdragon successes

Qualcomm’s Snapdragon X Elite CPUs set new records for performance and power efficiency on Windows systems. But how exactly did Qualcomm launch with a chipset that outperformed the competition from Intel and AMD?

Citing Moore’s Law, Kondap explains it this way: “Chip manufacturing technology is improving at a fairly steady rate, enabling processors to physically shrink in size every two or three years.

“Every size reduction enables us to improve power efficiency and add more transistors. Then we refine the manufacturing process annually to achieve further efficiency and performance improvements.”

However, Qualcomm’s success isn’t just down to smaller and smaller transistor designs. The Snapdragon X Series also features custom Oryon CPU cores built from tech startup Nuvia’s IP and Arm’s v8.7 micro-architecture.

While Qualcomm and Arm are in a legal standoff over Nuvia’s CPU core designs, the initial court case ruled in favor of Qualcomm that the Oryon CPU cores are appropriately licensed, meaning Qualcomm can continue building the Snapdragon X Series into the second generation.

Qualcomm’s custom Oryon cores are a “ground-up redesign, focusing on PC-specific computational demands,” enabling the Snapdragon X Series to outperform its other Arm rivals, Kondap says.

“The performance advantage of Snapdragon X Series stems from both its unique architecture and custom-designed Oryon cores, which mark a significant leap forward in processing power,” Kondap explains.

Qualcomm considers the Snapdragon X Oryon CPU cores to be an “industry-leading” design that “integrates power-optimized, new custom microarchitecture, including a new address translation unit.”

As Kondap explains, “This improves multi-tasking application performance, virtualization, and hypervisor situations common for demanding modern PC workloads. A sophisticated branch predictor and advanced power management design have been meticulously engineered to provide superior power efficiency.”

Building an ecosystem

Qualcomm’s biggest challenge with the Snapdragon X Series processors isn’t performance, power efficiency, or even the size of semiconductors. Instead, the struggle for Snapdragon has been the comparatively small Windows on Arm ecosystem.

To Qualcomm’s credit, many major applications are available on Windows on Arm natively, if not through emulation, including Adobe Photoshop and a number of music and content creation applications like DJayPro and Moises.

In addition to building out a robust ecosystem that can meet the needs of most users, Qualcomm has also heavily invested in AI features and applications.

After all, as Microsoft’s chosen Copilot+ flagship processor, the Snapdragon X Series has become practically synonymous with the terms NPU (neural processing unit) and artificial intelligence.

But where does Qualcomm see AI going in the future? So far, we’ve seen AI text, video, audio, and image generation, plus AI summaries for emails and PDFs. However, the exciting aspects of AI are not what it is doing right now but where it will be in the next few years.

Kondap believes the future of AI in computing is multi-faceted. “I’m most interested in advancing the capabilities of on-device AI to enhance productivity, creativity, and overall efficiency for both professional and personal use” — an opinion Kondap shares with Qualcomm as a whole.

“Qualcomm’s long-term vision for AI PCs includes a powerful roadmap for improving PC productivity, creativity, and entertainment,” he said.

“We expect AI PCs will be able to tackle larger and more complex AI tasks for users, as well as anticipate their intent and needs by drawing on context and past interactions.”

Offering a clearer picture of how these concepts could look in action, Kondap says agentic AI is the future of artificial intelligence.

“Instead of manually editing every photo by hand, you’ll be able to ask your PC to act as your agent, applying the choices you made in earlier examples to stylize new images,” Kondap says.

AI’s impact on everyday computing could be revolutionary, with the same agents Kondap mentions also taking on roles that help with everything from squeezing out additional performance to improving user accessibility, stating:

“We can expect improvements in sustained thermal performance, real-time translations, and interactive control through hand and facial expressions.”

All of this is in service to a clear future where “AI tasks will happen automatically and seamlessly: Your PC will be able to make the most of larger screens and windows, so even old videos and games look smoother and more detailed than before.”

Qualcomm’s 2025 roadmap and what’s ahead

Qualcomm launched the Snapdragon X Elite flagship silicon in June 2024 alongside the Snapdragon X Plus and has since rounded out its laptop and mini-PC processor lineup with the budget-friendly Snapdragon X Plus 8-core and Snapdragon X.

Qualcomm’s Snapdragon X chips were the first to be part of Microsoft’s new Copilot+ ecosystem, helping bridge the gap between Windows and macOS systems in performance and power efficiency.

With new budget systems arriving with the same NPU as the higher-powered flagship, Qualcomm’s strategy appears clear as the company gears up for its second generation of Snapdragon X silicon.

With the Snapdragon X chipset launch during CES 2025, Qualcomm’s Snapdragon X Series portfolio is officially complete. If you want to snag one of the best AI PCs, now’s a good time to buy. Particularly with the proposed White House silicon tariffs, laptops and gaming hardware prices could spike up to 40%. Of course, the US tariff tactics are subject to change, so these figures are not set in stone.

Qualcomm’s Snapdragon X Elite chips helped push Windows to close the performance and efficiency gap against Apple in 2024. However, subsequent launches from AMD, Intel, and Apple have caught up and overtaken the Snapdragon X Elite.

So, Qualcomm will need to keep pushing innovation to pull ahead in 2025.

On Tuesday, Google's Q4 2024 earnings call revealed much about how the company plans to spend its megabucks in the year ahead. And few ambitions of expenditure stood as lofty as pouring approximately $75 billion into capital expenditures, much of which was earmarked to greatly improve AI infrastructure.

That's over twice what Google spent in 2023, and rivals the projected GDP of entire countries like Panama, Uruguay, or the African nation of Ghana.

However, while not all of Google's capex will be poured into investment in AI, CFO Anat Ashkenazi was clear that a "majority of that is going to go toward our technical infrastructure, which includes servers and data centers." According to Ashkenazi, "approximately $16 billion to $18 billion" of its total capex investment would take place "in the first quarter" of 2025.

Spend big, win big?

CEO Sundar Pichai noted that Google's sizeable investment in technical infrastructure was a necessary cost, citing increasing demands for the company's cloud AI services, stating, "Cloud customers consume more than eight times the compute capacity for training and inferencing compared to 18 months ago."

This investment would not only ensure Google's ability to meet an increase in customer demand but also maintain the company's infrastructure efficiency, which, according to Pichai, is among the world's most efficient of its kind — delivering "nearly four times more computing power per unit of electricity compared to just five years ago."

It's likely that these improvements will help accommodate the training and wider rollout of more advanced models such as the Gemini 2.0 Flash 'thinking model,' which will become generally available for developers and customers today, joining a spate of recently launched reasoning models including the o1-powered Think Deeper for Copilot and DeepSeek-R1.

The latter of which was directly referenced by Pichai, who stated that the "tremendous team" behind the Chinese startup had done "very, very good work," when fielding a query from Goldman Sachs analyst Eric Sheridan on investors raising questions about the long-term cost curve for AI.

DeepSeek made shockwaves in January after revealing it had trained one of its latest V3 models for less than $6 million, and revealed some of the most competitive API pricing seen to date, costing only 55 cents per million tokens — some 96% less than the $15 per million tokens charged by OpenAI while providing similar levels of performance and accuracy.

However, despite DeekSeek's impact on of the AI market last month (which effectively saw almost $600 billion knocked off of Nvidia's valuation overnight), Pichai insists that not only is Google's AI "more efficient" in cost, latency, and performance, but that a return on the investment in AI could pay off as the company explores "very good ideas for native ad concepts" and making AI tools more accessible for business in an effort to sell AI-driven efficiencies to companies at scale.

Google's big spend spurs investor skepticism

That said, Pichai's optimism that its capex investment would be "directly driving revenue" doesn't seem to have landed.

Following the earnings call, Bloomberg reported that shares for Google's parent company Alphabet dipped by roughly 7% in premarket trading on Wednesday, effectively eliminating much of its 9% gain in 2025 — despite cloud services contributing $11.4 billion to the company's $96.5 billion revenue total for Q4.

While Alphabet insists its capex investment is necessary to hold the lead in Search and AI, a stock drop like this suggests that Wall Street isn't convinced. The company wants Google to spend big, but markets reward those who spend smart.

DeepSeek's arrival onto (and major upset of) the AI scene has raised a question Alphabet may already be feeling the impact of, as investors ask whether endless scaling in compute truly is the only path forward, as companies like Google, Microsoft, and OpenAI have claimed.

More from Laptop Mag

• DeepSeek: The best ChatGPT alternative or a hotbed of dubious claims?
• DeepSeek vs. ChatGPT: Here's what critics are saying
• "Truly magical" ChatGPT feature comes to Microsoft Copilot — and it's completely free

To say 2024 was a rough year for Intel may be an understatement. But, as Intel’s Robert Hallock tells Laptop Mag, “You can't go backwards. That's unacceptable.”

Even after various setbacks in the past year — from the beleaguered Arrow Lake desktop launch to speculation about the company’s direction and the resignation of CEO Pat Gelsinger — Robert Hallock, the company’s Vice President and General Manager of Client AI and Technical Marketing, says Intel has no choice but to look forward.

That forward focus means Intel intends to future-proof its desktops and laptops for when AI becomes a key component of all computing. This philosophy is everywhere at Intel, from the successes of Lunar Lake to the enthusiast skepticism of AI, NPUs, and Arrow Lake-S.

“If you didn't do the investment in AI; if you didn't get your software house in order; if you didn't get your hardware house in order, then suddenly what is now a core ingredient of performance and power – where maybe it wasn't two or three years ago – you're upside down,” Hallock explains.

So what drives Intel today, and what can we expect in the years ahead?

Why you want Intel Inside

“What [performance and efficiency] metric does someone care about?” Hallock mused in a recent interview with Laptop Mag. “A lot of time this year was spent on battery life, a lot of time this year was spent on AI. Is that the right answer for everybody? Probably, probably not.”

What Intel can do is offer an option for nearly every user. It’s a benefit of having a massive, diversified CPU business.

“As we think about the market from the Intel side, what are they going to value more?” Hallock mused. Of course, Intel’s market considerations don’t just account for the customer but also Intel’s OEM partners that make laptops and mini-PCs, like Dell and Asus.

“Is it packing in more CPU cores? More GPU performance? Are they going to value a larger NPU? I know what choice I'd make: I want more CPU cores because I'm an enthusiast.

“The cool thing about being at Intel is we have the road map flexibility to do that.”

After all, Arrow Lake and Lunar Lake CPUs are two very different products with different strengths and weaknesses. The key differences between the two chips are a matter of need: Arrow Lake offers higher thermal design power with a lower-powered NPU, while Lunar Lake is a lower-power (30 watts and under) system with a more powerful NPU. But the final word on those chips will always be the same, Hallock says. “The end of the day, it's still going to be performance and power.”

But what about the NPU? Do people care enough about the NPU here at the start of 2025?

While some may consider the less-powerful NPU on Intel’s Arrow Lake CPUs to be a miss, especially among the enthusiast and performance markets, Hallock compares the NPU to the rise of the iGPU.

“Integrated [graphics] on the CPU started by being ridiculed… But you know that the whole point was just to light up a display for someone.”

These days, including an integrated graphic tile on a CPU is now standard. Intel’s view is that NPUs and AI will be just as necessary in just a few years.

Intel AI for graphics and “for good”

When discussing computer hardware in 2025, it is impossible to avoid artificial intelligence. The need for AI, in the cloud and on devices, is greater than ever.

Amidst all the skepticism around AI, some small glimmers of an interesting future are emerging. As Intel’s VP of Client AI, Hallock’s perspective is an optimistic one, especially on the gaming front.

“Broad utilization is likely a few years away, but there are interesting use cases for NPC voices, character realism, quest text, world/texture generation, and more. I think these tools will be especially useful for indie developers, as they can get more immersive results with limited time and staffing.”

Another major arena for AI is security.

“At Intel, we often like to talk about ‘AI for good.’ That's why we work with vendors like Trend Micro, McAfee, Crowdstrike, and Bufferzone to support AI models that improve the security of the PC... Intel has a comprehensive traditional system security stack that operates below, at, and above the OS. We also think offline AI is an important step that enables users to take better control of their privacy and data rather than taking data up to the cloud.”

Looking ahead: The Intel roadmap

A look at what’s ahead for Intel’s chip roadmap also requires a look in the rearview mirror. Intel launched its first AI PC chip, Meteor Lake, in late 2023, though many systems didn’t hit the shelves until 2024.

While Meteor Lake offered decent performance and efficiency gains, its competitors left it in the dust in the second half of 2024. Intel’s next 2024 mobile release was the Raptor Lake Refresh high-performance CPUs, the 14th and final generation of Intel’s “Core i” branding.

In the fall of 2024, Intel followed up on AI PC computing with the Lunar Lake Core Ultra 200V series. While this silicon was designed for ultra-thin and light laptops with impressive battery life, Lunar Lake also offered gains in AI and gaming performance.

This triumphant laptop launch was followed by a beleaguered launch for Intel’s Arrow Lake Core Ultra 200S/K desktop chips. While Intel has identified and patched all five causes for the poor performance and efficiency behind the sub-par launch, it did cause many desktop PC enthusiasts to flock to AMD to upgrade their gaming rigs.

Intel has also gotten into the discrete graphics game with the Arc B-series desktop GPUs, which have been well-received, partly due to how budget-friendly Intel’s Arc GPUs are compared to pricier Nvidia and AMD graphics cards.

Intel’s Arrow Lake is looking for redemption on mobile platforms, with the H and HX series of laptop chips launching by the end of March. Arrow Lake will also see U-variant chipsets launch later this year. Additionally, the next generation of Intel silicon, Panther Lake, is already in testing and is expected to launch in the second half of 2025.

What’s next for Intel?

Intel has the Arrow Lake H chips coming by March for high-performance portable laptops, with the Arrow Lake HX gaming and workstation variants coming shortly after. The Arrow Lake U variant of more power-efficient silicon will be coming later this year. Lastly, Intel has confirmed that the next generation of Intel processors, Panther Lake, is already in testing.

We want to see how Intel’s new chipsets hold up in our lab tests before we make any claims, but if you’ve been considering updating to a new Intel-powered AI PC or gaming laptop, you might want to hold off another few months.

Of course, waiting also risks getting hit with tariffs on tech imports, which could see laptops and gaming hardware prices spike by up to 40%. However, the White House’s tariff tactics are in a state of flux, so this may change.

When it comes to advances in performance, power efficiency, and AI optimization, Intel has plenty in the works.

Hallock understands the “enthusiast skepticism” of AI and NPUs. “We're still learning how to deal with this, and some people are still waiting for their killer app, right? Not everybody has a killer app for AI yet,” he says.

No matter which AI feature or app becomes the industry favorite, Intel plans to have processors for it.

Can we keep expecting 20-30% increases in hardware performance and efficiency, generation to generation?

As part of our 2025 Silicon Survey, Laptop Mag spoke with executives from AMD, Apple, Arm, Intel, MediaTek, Nvidia, and Qualcomm to determine if we can expect to see performance and efficiency plateaus in the coming years.

You’ll be able to catch these exclusive interviews in full throughout the week in Laptop Mag’s Silicon Survey special issue. And while the microchip world may be divided on whether or not Moore’s Law is actually dead, all of our interviewees agreed that performance gains will continue long into the future.

But what that means, and how we get there differs by chip maker. Intel’s Robert Hallock tells Laptop Mag, when it comes to hardware performance and efficiency, “You can't go backwards.”

The only constant is progress.

What is Moore’s Law and why does it matter for chips?

Moore’s Law is a trend in microchip design, first recognized by Intel Co-Founder Gordon Moore. Moore observed that the number of transistors on an integrated circuit doubles roughly every two years. In 1965, Moore predicted that this trend of doubling transistors would continue for the next decade. However, his observations have stood as a guiding principle of processor design for the last 60 years.

While the end of Moore’s projections proving accurate has been heralded many times, Computex 2024 saw the most recent resurrection of the “Moore’s Law is Dead” debate among chipmakers.

Nvidia’s CEO Jensen Huang is the largest proponent of the idea that Moore’s Law is dead, while former Intel CEO Pat Gelsinger argues otherwise. After all, Intel’s Lunar Lake chips prove that you can still double the number of transistors on an integrated circuit, even in modern computing.

Taking a long, hard look at raw computing power and efficiency from all the major chip makers over the last several years, we wonder: can we expect to keep seeing the same major performance increases between generations of silicon?

Eventually, physical hardware scaling will have to change. Moore’s Law is only a trend in outcome, not a bankable guarantee. Semiconductors can only get so small, and microchips can only house so many transistors before the actual and inarguable laws of thermodynamics have to step in. TSMC is already hard at work on a 2-nanometer process for semiconductors, but how small can a semiconductor get before it cooks under the pressure?

As our interviewees highlighted, microchip performance is about more than transistors. Chips aren’t just hardware. To run a new chip, you need new drivers and software.

With the rise of AI, Nvidia’s argument has changed from “Moore’s Law is dead” to “Our systems are progressing way faster than Moore’s Law” because of artificial intelligence, which has been the biggest force behind Nvidia’s new RTX 50-series “Blackwell” GPUs.

Regardless of the approach, we can expect major performance gains to continue for several generations

Part of the argument that “Moore’s Law is Dead” can be traced back to the industry itself.

“I think the industry is always cyclical sometimes. It's always a balance of performance per Watt, right?” Intel’s Robert Hallock tells Laptop Mag. “Sometimes the industry can unlock a new process node or a new technology and performance jumps way up, and then you can't do that every year… I firmly believe there are still leaps and bounds out there, that we've by no means hit the collective end of our ropes on performance.”

However, it isn’t just a hardware versus software debate, as chipmakers have now adopted a dual approach.

“Chip manufacturing technology is improving at a fairly steady rate, enabling processors to physically shrink in size every two or three years. Every size reduction enables us to improve power efficiency and add more transistors,” Qualcomm’s Kedar Kondap tells Laptop Mag.

But Kondap also admits that AI and the introduction of the NPU in Qualcomm’s Snapdragon X series processors has “created new performance metrics and introduced AI capabilities into both the mobile and PC markets.”

Nvidia has long held that optimization and AI are the way of the future for continued performance. After all, Nvidia’s new RTX 50-series GPUs are still using the same 4N (4-nanometer) node as the RTX 40-series GPUs, because “The 4N process provides us with the best combination of performance, power, and price for our GeForce RTX 50 series GPUs” Nvidia’s Jesse Clayton expounded.

That said, refusing to place all of its eggs in one basket, Clayton also makes Nvidia's wider goals apparent, stating, “hardware is not the only important aspect … NVIDIA’s AI software platform, which has been in development for more than a decade, enables developers to get the most out of their RTX GPUs.”

AMD’s Director of Product Marketing, Adam Kozak, may have been the most succinct. Even when using older architecture and older nodes, you can still see major performance improvements, because “The software can now do more tricks.”

While chip makers like Intel and Qualcomm strive to prove that Moore’s Law still has life left in it, others focus on fine-tuning output, using software and AI to push performance past potential plateaus. And we’ll see these improvements coming throughout 2025 and over the next few years.

Laptop Mag invites you to read about these improvements and more as we publish a series of interviews with AMD, Apple, Arm, Intel, MediaTek, Nvidia, and Qualcomm throughout the week as part of our Silicon Survey special issue.

It's been precisely one week since a ChatGPT rival surprised the AI world.

DeepSeek, a China-based company, unleashed its R1 model and is suddenly at the center of the AI world's attention — for good reason.

On top of being seemingly as sophisticated as OpenAI's o1 model of ChatGPT, DeepSeek's R1 model is free. Yes, that's right... zero dollars.

That proposition, coupled with the fact that it was seemingly developed for a small fraction of the cost of other splashy LLMs, has sent ripples throughout the tech world, especially from the financial side.

In just one day, Nividia saw $600 billion knocked off its market cap, marking the most significant one-day loss in market history — that's less of a ripple and more of a tidal wave.

And while all of that is impressive, DeepSeek might be having another unintentional impact that's just as big, and it's got almost nothing to do with the stock market.

A TikTok problem

While DeepSeek's ability and price have dominated the conversation, there's another throughline that's decidedly less flattering for the company — privacy.

One of the main differences between DeepSeek and the rest of the LLM field right now is that it's not American or, more specifically, based in China.

For obvious TikTok-related reasons, that's a bad thing for an American app to be right now. The idea of an app that Hoovers up your data and sends it to a server in China is problematic for many reasons.

One is that privacy — whether for an American app or a foreign one — is essential. The more your personal data is collected, the higher the risk is to your digital safety.

Secondly, there's little difference between a private company and the government in China, which raises the question of what strictly user data could be used for.

That argument is for another day. Still, it doesn't take much extrapolation to see how a potentially adversarial government could use massive data troves for purposes that aren't in Americans' interest.

And in many ways, none of that is surprising. Of course, a non-American coming in and wiping out the value of a titan like Nvidia will make some waves, and, of course, that will dredge up skepticism.

But what I find more interesting isn't what DeepSeek says about a Chinese AI company but about AI broadly.

A moment for AI privacy

It didn't take long for DeepSeek's R1 to find controversy.

Just a day after its ascension into the public conversation, Bloomberg and the Financial Times reported that Microsoft and OpenAI are investigating whether DeepSeek used less-than-scrupulous methods to train R1.

Bloomberg reports:

“Microsoft Corp. and OpenAI are investigating whether data output from OpenAI’s technology was obtained in an unauthorized manner by a group linked to Chinese artificial intelligence startup DeepSeek, according to people familiar with the matter."

Naturally, OpenAI seems less than enthusiastic about the idea that DeepSeek copied the company's homework to build its R1 model. As many have pointed out, that aversion is blatant hypocrisy.

OpenAI notoriously trains its algorithm using copyrighted material and other people's intellectual property. It is currently in a legal battle with the New York Times over using the paper's content to train ChatGPT.

That's not even addressing privacy concerns with whatever data OpenAI stores about inputs from its users or anyone who signs up to use its platform.

And there's a distinction between data scraped by a private company in the U.S. and one in China, where the lines between private and public entities are incredibly blurry. Still, all of that is almost irrelevant because DeepSeek, whether meaning to or not, is making us talk about AI privacy.

A chance to set things straight

Every new digital platform has some privacy pitfalls. Consider the rise of social media, voice assistants, and e-commerce.

However, the difference between those platforms and the AI platforms created before us is that we now have years of history from which to draw. That means, at least theoretically, things could be different.

I say theoretically because to protect your privacy, history says you must fight for it. For example, voice assistants like Alexa, Siri, or Google Assistant didn't become opt-in until scandals emerged over recording programs that inadvertently captured sensitive conversations.

After whistleblowers leaked details of those programs, tech giants quickly rolled them back or at least gave users the option to opt-out. We aren't there yet, but moments like DeepSeek bring us closer to putting privacy on the map.

However, we'll likely have to learn the hard way. On Thursday, security researchers found that DeepSeek was storing millions of log lines in an unsecured database accessible without authentication. That information could theoretically give bad actors access to DeepSeek's internal systems, which is not great.

But if it could happen to DeepSeek, it could happen to anyone. Therefore, it's more apparent than ever that the time to have an AI privacy conversation is now.

Whether that happens is anyone's guess, but when all of our data ends up in the hands of someone who doesn't have users' well-being in mind, we'll at least know that we should have seen it coming.

On Thursday, Microsoft made one of ChatGPT's most impressive features free for all, releasing a new "Think Deeper" mode for Copilot, powered by OpenAI's o1 reasoning model.

Think Deeper was originally trialed in October within Copilot Labs, Microsoft's AI playground for experimental features accessible to Copilot Pro subscribers. However, Microsoft AI CEO Mustafa Suleyman announced on LinkedIn yesterday that it will now be available to all Copilot users.

"Today we’ve made Think Deeper free and available for all users of Copilot" Sulleyman wrote, "I urge you to give it a try. It’s truly magical."

Think Deeper brings OpenAI's most advanced reasoning model to Copilot for free, eliminating the paywall that previously restricted access to ChatGPT Plus and Pro subscribers, who can expect to pay up to $200 per month for unlimited access.

What is Copilot Think Deeper?

Unlike some of the more standard AI models you might have interacted with like GPT-4o, advanced reasoning models like o1 or DeepSeek-R1 are designed to tackle more complex tasks. This typically means taking longer to provide an answer while the model breaks down and analyzes the information step-by-step to produce a more logical response.

This makes reasoning models particularly useful for deep dives on dense topics, coding, or when constructing plans for larger projects. While regular models are great for helping with everyday tasks, reasoning models like the one used by Copilot Think Deeper go the extra mile on tougher asks, ensuring you get the richest and most accurate reply to your questions.

As Suleyman puts it, "Brain dump everything into Think Deeper and watch it churn through it all and spot out a step-by-step guide to making it happen." Suleyman also cites examples of his own use of Copilot Think Deeper in developing fitness routines and helping to manage product launches. However, Copilot's new reasoning tool is capable of so much more, with Suleyman highlighting that "The possible uses here are really infinite."

All Copilot users can now access Think Deeper for free using the AI's web portal or dedicated app on iOS or Android. Simply click or tap on the "Think Deeper" button within the prompt bar to get started.

Copilot refuses to take a backseat to DeepSeek

The news of Copilot's wider Think Deeper release comes amid a swirl of interest in reasoning models like o1 after Chinese startup DeepSeek managed to leapfrog ChatGPT to the top of the Apple App Store charts following the release of its own free-to-use R1 reasoning model earlier this month.

Microsoft's decision to bring Think Deeper out of the Copilot Labs and into the hands of its users may have been a strategic play, hoping to stymie the popularity of DeepSeek by offering access to a powerful free-to-use reasoning model of its own.

That said, the competition hasn't stopped Microsoft from announcing DeepSeek-R1 as the latest model to enter its Azure AI Foundry, adding to a catalog of over 1,800 AI models, with CVP of Microsoft AI Asha Sharma noting that it "offers a powerful, cost-efficient model that allows more users to harness state-of-the-art AI capabilities with minimal infrastructure investment."

However, it's a move not without controversy, as DeepSeek's success has painted a huge TikTok-shaped target on its back following concerns over data handling and privacy, and questions over the model's training processes.

More from Laptop Mag

• Microsoft reveals new Surface and Surface Pro laptops with up to 22 hours of battery life
• DeepSeek's success has painted a huge TikTok-shaped target on its back
• Microsoft 365 Office didn't need a rebrand but Windows 11 desperately does

Last week's release of the free-to-use DeepSeek app proved incredibly popular, toppling OpenAI's ChatGPT from the top of Apple's App Store this weekend to become the platform's most downloaded app. Something sure to cause a stir no matter what given the widespread popularity and dominance of OpenAI's frontier AI chatbot over the last few years.

However, the real impact of DeepSeek was seen by AI developers and investors paying attention to the company's claims surrounding training and monetization. Not only are DeepSeek's latest V3 and R1 models up to 97% cheaper to use than the competition (Only 55 cents per million tokens, compared to OpenAI's $15 cost per million tokens), but one of its models was reportedly trained on less powerful hardware than that available for US companies for less than $6 million, a fraction of what Google, OpenAI, and Meta are currently spending to develop frontier models.

These claims (however dubious) have sent ripples through the AI space, particularly impacting the stock of major AI players like Nvidia, which took a $593 billion hit to its market value yesterday. As Annex Wealth Management chief economist, Brian Jacobsen, suggested, "It could mean less demand for chips, less need for a massive build-out of power production to fuel the models, and less need for large-scale data centers."

Success to this degree has to be commended, but drawing this level of attention and influence over markets also paints a giant target on the company's back. Given the U.S. reaction to similarly disruptive Chinese-owned companies in the tech space like Huawei or TikTok, there's no guarantee that these successes won't be short-lived.

DeepSeek: Will it get the TikTok treatment?

When we think of the major players in the smartphone world, it's often seen as a two-horse race between Samsung and Apple. However, there was a time when a third name was involved in that mix: Huawei.

Surprisingly, this brand wasn't even competing from a distant third place. In 2018, Huawei overtook Apple and became the world's second-largest smartphone seller behind Samsung, ending a seven-year stint of inseparability.

However, Huawei's initial successes were hamstrung after the FCC voted to ban the sale of its devices in 2022. This decision was based on national security concerns that the Chinese manufacturer's devices could potentially be used by the Communist Party of China (CCP) for surveillance or espionage purposes.

Similar actions by the U.S. can be seen with the recent TikTok ban, which also became a pressing matter in 2022 when FBI Director Chris Wray claimed that "the Chinese government could use [TikTok] to control data collection on millions of users or control the recommendation algorithm, which could be used for influence operations."

With the success of Chinese-owned DeepSeek, we once again see the dominance of an American company like OpenAI threatened similar to how Apple was by Huawei, and a powerful AI assistant with built-in biases that censor any anti-CCP sentiment and over-abundant data collection that reaches as far as tracking the rhythm and pattern of your keystrokes — which can be accurately used as a type of biometric authentication to identify a person.

Add to this claims that DeepSeek is using Nvidia H100 and A100 GPUs to train its models, hardware banned from export to China by U.S. sanctions in 2022 — a claim backed by Elon Musk — and DeepSeek could find itself in deep water if placed under scrutiny.

What's next

All the markers are there to place DeepSeek firmly in the crosshairs of the U.S. government. Given the current push to make the United States the "World's capital of AI," it stands to reason that the success of China's frontier AI may be short-lived.

For now, DeepSeek's claims of cost-effective competition to big spenders are causing more to be impressed than call for investigation with President Trump stating that DeepSeek "should be a wakeup call for our industries that we need to be laser-focused on competing to win." However, this sentiment will surely only extend to the limits of DeepSeek's low-cost training claims being proven true.

Representative John Moolenaar (R-Mich), chair of the House Select Committee on China, has taken a more firm stance on DeepSeek. On Monday, Moolenaar stated, "The U.S. cannot allow CCP models such as DeepSeek to risk our national security and leverage our technology to advance their AI ambitions. We must work to swiftly place stronger export controls on technologies critical to DeepSeek’s AI infrastructure."

The internet seemingly has a new favorite AI, and it's not the latest ChatGPT model from industry untouchables OpenAI.

Soaring to the top of Apple's App Store, Chinese artificial intelligence chatbot DeepSeek has now become the top-rated free app for productivity after a groundswell in popularity following the release of the DeepSeek-R1 "reasoning" model on January 20, overtaking OpenAI's ChatGPT in the process.

Beyond App Store leaderboards, claims surrounding DeepSeek's development and capabilities may be even more impressive. The company behind the LLM (Large Language Model) claims it cost less than $6 million to train its DeepSeek-V3 model and used limited hardware compared to its American contemporaries while achieving similar results.

However, while DeepSeek is proving popular with users and developers alike, mainly thanks to its favorable API pricing, all that glitters isn't gold when it comes to this app, and an air of controversy undercuts an otherwise successful launch of two highly capable AI models.

DeepSeek: What is DeepSeek?

DeepSeek was founded by Liang Wenfeng, a Chinese entrepreneur and co-founder of the High-Flyer hedge fund based in Hangzhou, Zhejiang, China. Originally, DeepSeek was intended to be an AGI (Artificial General Intelligence) research wing of High-Flyer, which has exclusively used AI in trading algorithms since 2021. However, since May 2023, DeepSeek has stood as its own company, with High-Flyer becoming one of its primary investors.

The company's DeepSeek LLM (Large Language Model) debuted in November 2023 as the open-source DeepSeek Coder and was followed by DeepSeek-V2 in May 2024. The company launched its latest DeepSeek-V3 model in December 2024 and has since seen a swell of popularity, with its mobile app racking up over 1.6 million downloads.

While the DeepSeek LLM is mainly similar to other popular chatbots like Google Gemini or ChatGPT, the app's free-to-use models are proving popular with users, and its developer-friendly API pricing is pushing it to the forefront of discussion.

DeepSeek: Why is it important?

The old myth goes that during the space race of the 1960s, NASA spent millions in taxpayer dollars on developing a space pen after it realized that ball-point alternatives were incapable of writing in the zero-gravity environment of space. Meanwhile, their cosmonaut counterparts avoided such costs and headaches by simply using a pencil.

While none of that is true, it's a parable of thriftiness and practicality that makes for an excellent story.

However, mirroring the legend of the space pen, DeepSeek has seemingly managed to pull off a similar feat in cost-effectiveness and practicality through the development of its DeepSeek-V3 model, which it claims to have trained for less than $6 million, a fraction of the hundreds of millions spent by other companies pursuing similar outcomes (while achieving comparable levels of performance).

Not only that, but DeepSeek's recent release of its DeepSeek-R1 "reasoning" model is designed to simulate logical thought by sacrificing the speed of a response for a more well-reasoned answer. It can achieve results equal to (if not better than) OpenAI's own "reasoning" model, GPT-o1 — even as the company claims to be hamstrung by U.S. export restrictions on more powerful Nvidia GPUs.

DeepSeek: How much does it cost?

DeepSeek is free to use online via its web portal or on mobile (with both Android and iOS apps available).

However, the impact that DeepSeek's emergence will have on the cost of AI for businesses, developers, and more could be most groundbreaking, with the company's API price model blowing the competition out of the water.

While OpenAI currently charges $15 per million tokens (a unit of data that prompts are broken down into during the generation of a model's response), DeepSeek costs only 55 cents per million tokens, a phenomenal drop in charges for API users of up to 96 percent.

This cost difference could be game-changing for many professional users involved with AI and poses a significant risk to OpenAI's potential income, with DeepSeek potentially now forcing the hands of other companies to lower their prices to remain competitive.

DeepSeek's impact is already being felt in the markets. Several semiconductor names are feeling the hit, including Nvidia. Following the release of DeepSeek's latest models on Monday, pre-market trading dropped 13.8%, threatening to wipe out almost $500 billion from the company's trading cap. However, it has since climbed back to 11%.

DeepSeek: Controversy

If everything DeepSeek has to offer sounds too good to be true, that's potentially because some of DeepSeek's claims may be just that.

The performance of DeepSeek's V3 and R1 models cannot be disputed. Still, many questions remain about the company's actual pricing, its use of hardware, the cost of its training, and the sourcing of its training data.

The latter has already been the subject of some controversy. Several users reported that DeepSeek V3 would refer to itself as ChatGPT, potentially indicating that this model was trained on public data sets generated by OpenAI's GPT-4 model.

Speaking to TechCrunch, Mike Cook, a research fellow at King’s College London specializing in AI, backed these claims, stating, "Obviously, the model is seeing raw responses from ChatGPT."

Cook highlights that this may not be an intentional action by DeepSeek but also points out that the practice of training models on data generated by other models can be "very bad," likening it to "taking a photocopy of a photocopy" in the sense that the quality of outputs will degrade each time.

It's also possible that by adopting generated training data, DeepSeek will inherit any of the same biases of the original model, adding to the chatbot's own biases, which enforce strict censorship by law of anti-Communist Party of China (CCP) narratives, including the events of the Tiananmen Square incident of 1989, Hong Kong protests, the ownership of Taiwan, China's treatment of the Uighur people, or the occupation of Tibet.

This form of censorship only degrades trust in the platform, and founder Liang Wenfeng's ties to the CCP only heighten concerns about how user data may be used or how Chinese authorities could misappropriate the platform in the future.

Writing for Biometric Update, Anthony Kimery, former Editor-in-Chief and co-founder of Homeland Security Today, highlighted how the platform could "support disinformation campaigns aimed at destabilizing U.S. institutions."

DeepSeek's claims that it developed its models on less advanced hardware are also being questioned. Citi analyst Atif Malik states, "While DeepSeek's achievement could be groundbreaking, we question the notion that its feats were done without the use of advanced GPUs to fine-tune it and/or build the underlying LLMs the final model is based on through the distillation technique."

Malik's questioning could have further weight, as while DeepSeek claims that its V3 model was trained using Nvidia H800 GPUs, a recent interview with Scale AI's founder Alexandr Wang on CNBC saw the company's CEO suggest "DeepSeek has about fifty thousand H100s."

The very same GPUs that were blocked from export to China by the Biden administration in 2023, with Wang continuing, "they can't talk about obviously because it is against the export controls that [the] United States has put in place."

Outlook

DeepSeek is a proven hit that will give companies like OpenAI something to consider when retaining its sizable user base in the face of stiff competition.

However, it remains to be seen if the new car smell still lingering on DeekSeek's latest models is masking the odor of misinformation surrounding how it developed its models and whether or not its pricing is sustainable in the long term.

Given the U.S.' recent reaction to TikTok, it's hard to imagine that a company like DeepSeek goes without serious scrutiny for much longer, especially as its models risk upsetting the apple cart on President Trump's plans to keep the United States as the "world capital of AI."

The future of gaming is rife with lifelike non-player characters powered by generative AI models — or that's what Nvidia thinks, at least.

At CES 2025, Nvidia released a series of demos for new AI tools and features for gaming powered by Nvidia's RTX 50-series GPUs. They range from AI audio editing for streamers to unpredictable "AI bosses," but the one that stands out the most to me is AI NPCs that are meant to supplant human-written dialogue with a chatbot.

Nvidia is clearly betting big on AI becoming omnipresent in gaming, especially generative AI, but whether gamers actually follow suit on that vision is another question entirely. I'm personally betting the answer is no, and here are a few major reasons why.

The promise and pitfalls of "AI NPCs"

Rumors and hype around "AI NPCs" in games have been bubbling for years now, but they have yet to truly materialize. Nvidia is trying to change that with an AI model called Nvidia ACE.

In a demo for ACE, Tao Zhang, CEO of TiGames, plays an tech demo for a "future project" called ZooPunk, where he interacts with an NPC running on Nvidia ACE and uses it to customize his ship. Zhang is able to talk to the NPC using his voice, rather than scripted dialogue options, and it replies with AI-generated lines.

At first glance, that sounds cool, but unfortunately, the way it actually sounds in the demo is less impressive. The AI NPC's dialogue is stiff, robotic, and hollow. It reminds me more of an automated voice mailbox message than an actual human. While it's a neat concept, it's a far cry from what human voice actors are capable of.

An upcoming indie title, Dead Meat, is another great example. This game has players interrogate suspects in a murder mystery. Players can ask the NPCs any question they want and the NPCs respond with AI-generated dialogue.

On paper, this seems like it could be a clever way to use AI for NPCs. However, the generated dialogue shown in a demo video is firmly in the uncanny valley. It definitely still sounds like an AI, besides leaving me wondering how the game maintains the consistency needed for something as complicated as a murder mystery.

Of course, it's certainly possible this technology will improve over the next few years. Maybe the lines will sound more natural or the AI-generated voices will sound less cold and robotic. Even if that does end up happening, it still might not be enough to make AI NPCs the new norm.

On a more practical level, I'm doubtful about how useful unscripted NPCs are in many types of games, especially RPGs and story-driven titles. NPCs are often how players get quests, learn important lore, or experience key plot points. The specific dialogue lines written for an NPC can be crucial to the story or even the basic functions of a game.

Sometimes NPCs' dialogue can be so well-written that it has a real emotional impact on the player. Star Wars: Jedi: Fallen Order is a prime example for me. It's one of my favorite games, partly because of the meaningful interactions Cal Kestis has with the cast of NPCs. Another great example is Tales of Kenzera: ZAU, in which emotional bonds and arcs with NPCs are a core pillar to the game's phenomenal, meaningful ending.

NPC interactions, and their emotional weight, are the result of intentional writing decisions, not something that can be randomly generated by an AI (or effectively delivered by one). Is it really worth losing that to have slightly less repetitive conversations with shopkeeper NPCs? I don't think so.

Are AI NPCs the future of gaming? Probably not.

One of my favorite games is The Outer Worlds (which is set to get a sequel this year). Whenever I recommend it to a friend, I always mention how much I love one of the companion NPCs, Parvati. It's not just me, either. Parvati is widely beloved by fans of the game thanks to her lovable personality and surprisingly deep, unique character arc.

Notably, Parvati is not AI-generated.

NPCs like her play a major role in the fabric of their games and play a major role in the communities that rally around those games. Whenever I shower praise on Parvati to a friend, it's partly because I'm excited for them to experience her story, too.

That kind of shared experience could disappear in a world where NPCs run on AI-generated dialogue. Two people could witness drastically different NPCs that have the same name and face but little else in common, not to mention completely different stories. That leaves less ground for a shared love and appreciation for elements of the game or story.

Plus, you lose the intentional artistic decisions that go into crafting an NPC's character arc and thoughtfully acting out their lines.

Video games are a creative art form. AI can be a tool in the artistic tool box, but it's not a replacement for the artist. Using AI to streamline game development or optimize hardware makes sense and has a net positive impact on the gaming experience. However, replacing nuanced, engaging NPCs with hollow AI chatbots, in my opinion, has a net negative impact and could strip away a vital part of any great game.

More from Laptop Mag

• Could Nvidia's AI bosses steal the soul of Soulsborne games?
• Nvidia's RTX 5090 still can't game in 4K without DLSS 4
• I'm a veteran laptop reviewer. Here are my top 10 laptops of CES 2025

Samsung's first Galaxy Unpacked event of 2025 is just around the corner, officially revealed to be taking place on January 22. In the meantime, leaks surrounding the company's latest Galaxy S25 Series of smartphones continue to pour in.

The latest, shared in a now-deleted X post by reputable leaker Evan Blass, appears to be official promotional shots of the upcoming Samsung Galaxy S25 and S25 Plus (shown below).

The images purportedly showcase Samsung's push to reduce display bezels, a new camera rim design, its overall design, and a familiar look at each's camera array. However, what's on the screen might be more interesting than anything else.

While the hardware hardcore may be distracted by the device's smoothed corners and amped in speculation over its lenses, I'm more interested in the reveal of the expected One UI 7 "Now Bar." A new Galaxy AI-powered feature that Samsung claims will be a "defining element" of its Android-based interface.

Samsung's "Now Bar," a different kind of Dynamic Island

Originally showcased in a November blog post to the Samsung Newsroom, One UI 7's new "Now Bar" feels suspiciously like Apple's Dynamic Island, which debuted on the iPhone 14 Pro models in 2022. However, Samsung's version of this may potentially have the upper hand, mostly because it could actually be good.

While Apple's Dynamic Island wasn't quite the groundbreaking piece of innovation its reveal reel would have led you to believe (eventually slipping into little more than a fancy notification bar), Samsung's Now Bar aims to integrate everyday actions into a widget-like bar for the device's lock screen. Better still, it won't require a large pill-shaped punch-out hole to do so.

Samsung's Now Bar will offer instant access to media controls, sleep and fitness insights, navigation assistance, and even recommendations on actions you can take to optimize your day-to-day phone use. Simply swiping the Now Bar offers you a selection of relevant slices of information based on your activities and most used apps. It can even offer a handy recap of events once the day is done.

While it's easy to pass this feature off as just another all-in-one widget, the fact that it seems to blend so tightly into One UI 7 and make effective use of Galaxy AI to ensure its recommendations are uniquely tuned to each user suggests more.

It's a bit like having a convenient way to scroll through your most important updates without needing to juggle several apps in the process. It's also a great way of breaking things down into bite-sized, easily absorbed doses of info, and effectively eradicating the "doom" from your doomscrolling in the process.

Samsung's Now Bar is more proactive than it is reactive, potentially making the next batch of Galaxy S25 phones feel smarter and more convenient in response — if it's handled right.

What's next?

We'd expect to see more of the Now Bar during the upcoming Samsung Galaxy Unpacked event on January 22. It's here that we're also expecting to see the Galaxy S25 Series in full, with several other new Galaxy AI features sure to be unveiled.

The Now Bar has promise, but so did the iPhone's Dynamic Island. Promise alone won't guarantee success, just ask Bixby.

Whether or not the Now Bar dramatically changes how Galaxy smartphone users interact with their devices remains to be seen, but if Samsung and third-party developers can apply some creativity to this space, it could become a fantastic utility and a real draw for Samsung's One UI.

However, if the Now Bar fails to catch on, it may become yet another forgotten gimmick, a once heralded "defining" feature that turns out to be all hype, no bite.

More from Laptop Mag

• What to expect at Samsung Galaxy Unpacked 2025: Galaxy Ring 2, Project Moohan, Galaxy S25-Series, and Galaxy AI
• Leaks show Samsung's Galaxy S25 series in full ahead of launch
• Samsung Galaxy S25 launch preorder deals — here's what to expect

CES is not only a showcase of the best tech to come in the year ahead, it's also a place for manufacturers and developers to showcase some of their grander ideas for how breaking technologies can theoretically be applied in the consumer space several years from now.

These concepts can sometimes steal the show, gifting us a look at a future yet to come. CES 2025 was no exception, and as you'd expect AI had a lot to do with many of the concepts we came across on showroom floors throughout Las Vegas, Nevada. However, one concept caught our eye in particular: Razer's real-time AI esports coach, Project Ava.

Best Concept of CES 2025: Razer Project Ava

CES is always full of products that may or may not come to fruition, but in that sea of concepts, one really stood out to us. Razer’s Project Ava is an AI esports coach that works in real-time.

That means, while you’re mid-raid in League of Legends, you can get feedback on where to push or which items to use all without having to leave your game to trawl YouTube.

What makes Razer’s AI coach different than competitors is that it utilizes visual data from your game in addition to game data from other players.

The result is a coach that’s tailored to your specific scenario and play style. Razer also envisions Ava being used for non-esports play in games like Black Myth: Wukong to give real-time advice on how to beat bosses as well as post-match coaching for breakdowns on what you did right and wrong during your match.

It’s hard to say whether Project Ava, if it gets released in full, will actually live up to all of Razer’s vision, but the idea is one that companies like Razer and its ilk are pursuing.

Project Ava may be a concept right now, but it’s also a template for gaming’s AI future.

The last 12 months have seen a lot of attention brought to the term "AI PC," used to denote computers housing processors or APUs with powerful NPUs (neural processing units) for enhanced AI computation.

However, even then, these machines lack the raw performance that many more powerful AI applications require to tackle complex generative AI tasks, leaving a lot of the work to be done in the cloud.

Until Nvidia announced Project Digits, that is. A personal AI supercomputer, and our Laptop Laurel winner for the best use of AI at CES 2025.

Best AI of CES 2025: Nvidia Project Digits AI PC

While the Nvidia RTX 50 series GPUs dominated the conversation at CES 2025, the company’s Project Digits was a massive announcement that went a bit under the radar.

Nvidia calls it an “AI Supercomputer on your desk,” delivering a petaflop of AI performance in a form factor that appears similar to a Mac Studio.

Who needs that kind of AI power? It’s targeted at AI researchers, data scientists, and the education market to help run large AI models without relying on ChatGPT or another cloud-based solution.

The key to its performance is the GB10 Grace Blackwell Superchip, designed in partnership with MediaTek. It also boasts 128GB of RAM and up to a 4TB NVMe SSD. Running Nvidia’s Linux-based DGX OS, it will be capable of running 200-billion-parameter large language models or 405-billion-parameter models if pairing two Digits together.

While it feels like AI is everywhere in the tech world already, this kind of (relatively) affordable system can put it in the hands of far more users, which is an AI advancement I can get behind.

Starting at $3,000, it isn’t cheap, but assembling a comparable AI computer could run over $10,000 in the not-so-distant past. Project Digits is coming in May from Nvidia and its partners.

Nvidia's Project Digits will be available to purchase from May 2025 from Nvidia and partners at a starting price of $3,000.

On June 10, 2024, Apple CEO Tim Cook announced one of the most significant software updates in his company's history, which would vault Apple into the AI showdown with the likes of Google, Microsoft, and OpenAI.

Rumors about the announcement reached a boiling point in the days leading up to Apple's annual Worldwide Developer Conference (WWDC), so by the time Cook reached the highlight of the keynote presentation, many knew what lie ahead.

Cook stood on a trail at Apple's headquarters on a sunny California day as he finally unveiled the star of the show. "Recent developments in generative intelligence and large language models offer powerful capabilities that provide the opportunity to take the experience of using Apple products to new heights."

Cook highlighted five core principles for Apple's version of AI: powerful, intuitive, integrated, personal, and private. That last one is significant as privacy and security have become staples of the Apple brand but pose challenges with AI, which relies on consuming vast amounts of data. Cook seemed confident in Apple's solution, though.

"All of this goes beyond artificial intelligence. It's personal intelligence, and it's the next big step for Apple."

That "next big step" was Apple Intelligence, Apple's rebranding of "AI." This new AI platform designed exclusively for Apple tech features text and image generation, notification summaries, AI image editing, and a long-awaited Siri upgrade.

It arrived mid-way with a heated AI arms race, but Apple had two huge advantages over its rivals. One is a "secret weapon," and the second is evident if you're an iPhone user.

Tim Cook's better-late-than-never AI strategy

In an interview with Marques Brownlee at WWDC 2024, Cook doubled down on Apple's AI strategy, insisting, "Generative AI was never off the table. It was always about pursuing it in a thoughtful kind of way. So, that’s what we’ve done, and we’ve implemented it in ways that are less likely to create issues."

Cook was responding to a question about Apple's previously unclear stance on generative AI, which has increasingly become a major feature on Android phones over the past few years through Google Gemini. Until May of this year, Apple appeared, outwardly, anyway, uninterested in entering the generative AI competition. Despite that, I wasn't surprised by the announcement of Apple Intelligence.

This is far from the first time Apple played the "better late than never" game. Apple has consistently allowed its rivals to do the risky business of testing new product categories before diving in itself. (The Apple Vision Pro was definitely not the first consumer mixed reality headset, but Apple did design it to be the best.)

Advantage 1: The "secret weapon"

Apple Intelligence includes most of the now-standard features of generative AI, such as generating text and images, summarizing content, and answering audio, text, and visual queries. Apple added some extra features, like AI-generated emojis, but nothing groundbreaking.

That might seem like a losing strategy, considering Google already offers all of that and more on its Gemini platform. Apple has something Google doesn't, though: Private Cloud Compute, a secret weapon which could set up Apple Intelligence for success in 2025 and beyond.

Private Cloud Compute ensures everything users do on Apple Intelligence is completely private. When the AI needs to process a request through the cloud, it goes through secure servers running Apple's own M3 Ultra chips. Even then, all of the data is anonymized, so user privacy is protected. That's a far cry from the privacy strategies of Apple's competitors, who were accused of scraping massive amounts of data earlier this year.

Privacy isn't even the only significant advantage Apple has up its sleeve. The second one is perhaps even more important and might even make Apple the new leader in consumer AI.

Advantage 2: If you're an iPhone user, you're already part it

Prioritizing privacy and security was a major win for Apple Intelligence, but it may not be Apple's biggest advantage over Google, Microsoft, and OpenAI.

If you're holding an iPhone, it might already be evident to you: Apple's potentially game-winning advantage is its massive, loyal user base.

As of 2024, 87% of teenagers in the U.S. have an iPhone, and 30% say they plan to upgrade their Apple tech within the next six months, specifically because of Apple Intelligence. That makes sense, considering young people are currently the most likely demographic to use generative AI.

Young people also represent the future of the phone market, so this data indicates that young people overwhelmingly prefer Apple devices and are more likely to use AI tools and apps. That means Apple Intelligence could help solidify the iPhone's dominance in the phone market for the foreseeable future. As a result, Apple Intelligence could become the country's most widely used AI platform simply by becoming the default AI on the iPhone.

Apple Intelligence: What's next for 2025?

If Apple can also nail the basics and ensure Apple Intelligence is consistent, fast, and reliable, it could quickly leapfrog Google's Gemini AI regarding users. The promise of privacy and data security gives Apple an especially strong advantage over Google, particularly when you factor in the ChatGPT integration on Apple Intelligence. This feature lets you privately use OpenAI's ChatGPT algorithm if Apple Intelligence can't complete a request for you.

Apple might win the AI race by choosing not to run at all. While its rivals, like Microsoft and Google, have been rushing to roll out as many new AI features as possible over the past year, Apple has been watching, learning, and taking its time.

Only time will tell, but Apple could become the king of AI by betting on the one thing you can't develop overnight: the trust it takes to build one of the biggest userbases in the world.

If you're anything from an AI enthusiast to the average AI tinkerer (or simply seeking out some of the additional features offered through Windows Copilot+ PCs or Apple Intelligence on Macs), then you'll need a powerful and performative laptop to keep up to speed with your needs.

At Laptop Mag, we review laptops year-round to ensure we're giving you expert-backed and up-to-date recommendations on which notebook is right for you. When it comes to the best AI PC category, our top picks are the excellent Asus Zenbook S 14 (UX5406) for Windows users and the impressive Apple Macbook Air M3 for those running macOS.

So, if you're shopping for a new laptop and looking to invest in an AI PC (or just a great laptop in general), check out our current top-tier picks below.

If you use an Android phone, there's a good chance it's running on a Snapdragon processor. That's not the only place you'll find one, though. This year, Qualcomm, the San Diego-based Snapdragon chipmaker, brought its CPUs to a whole new stage: AI-powered laptops.

On May 20, 2024, Microsoft unveiled the first laptops in its new Copilot+ PC line, all powered by Qualcomm's Snapdragon X series chips. This was Qualcomm's first significant entry into the laptop market, so the success of the Copilot+ PCs was crucial.

Qualcomm had big plans for its laptop processors, particularly the flagship Snapdragon X Elite chip.

As a Qualcomm spokesperson tells Laptop Mag, "The launch marked a pivotal moment for Qualcomm as we unveiled a new era of AI-enabled PCs. The Copilot+ PC devices powered by Snapdragon X Elite exceeded expectations in providing high performance-per-watt and long battery life — the top requirements for consumers."

How the Snapdragon X chips supercharged AI PCs this year

At first glance, you might wonder why Microsoft would risk the launch of a whole new series of laptops by working with a chipmaker that's relatively new to the laptop market. Qualcomm was up to the challenge, though. In fact, Snapdragon-powered laptops had some clear advantages right from launch.

For one thing, Microsoft's Copilot+ PC line-up was the largest collection of AI PCs ever launched when it debuted in June. The Snapdragon X chips at the heart of these laptops were designed specifically with AI in mind, resulting in stronger overall AI performance compared to older Windows 11 laptops.

"One of the most transformative user experiences enabled by Snapdragon X Elite processors is their ability to handle generative AI workloads on-device," a Qualcomm rep tells Laptop Mag. "For example, when devices are unplugged, our competitors’ performance falls between 30 to 45%, and there's no drop on Snapdragon."

That balance of processing power and battery efficiency also resulted in some seriously impressive battery life scores for Snapdragon-powered laptops. While there are a few outliers, most of the Copilot+ PCs Laptop Mag tested this year lasted 13 hours or more in our battery life test.

For instance, the Dell XPS 13 lasted a staggering 19 hours! Even more impressive, the Qualcomm-powered Lenovo ThinkPad T14s Gen 6 broke our battery life record, outlasting any other laptop we've reviewed to date with a battery life of 21 hours and 3 minutes.

Those battery life results handily prove that Qualcomm's claims about power efficiency on its Snapdragon chips are the real deal. Even if you're not so interested in the AI features on Snapdragon-powered laptops, this level of battery life alone could seal the deal for prospective laptop buyers. It's enough to potentially win over people who might not have otherwise decided to try an AI PC.

And if you are an AI enthusiast or want to run AI apps on your laptop, Snapdragon may be the best platform for that, too. "Unlike other processors, Snapdragon X Elite features an integrated AI Engine capable of over 45 trillion operations a second (TOPS), allowing them to handle demanding AI tasks seamlessly," a Qualcomm spokesperson tells Laptop Mag,

With top-notch performance for running virtually any consumer AI app you want and battery life bordering on 24 hours, this first generation of Snapdragon-powered laptops set a high bar. Can the next generation take things even further?

Can Qualcomm stay ahead of its rivals in 2025?

Given the overall success of the Copilot+ PC line-up this year, many are intrigued to see what Qualcomm delivers next. On one hand, it will probably aim to continue the innovation achieved in 2024. However, Qualcomm will also need to connect with a larger pool of users, many of whom are still unclear about what exactly an "AI PC" is or why they need one.

A Qualcomm spokesperson tried to clear this u, telling told Laptop Mag, "'AI PC' refers to the integration of generative AI tools that function as real-time assistants to enhance productivity, creativity, and efficiency. Devices with Copilot+ capabilities go beyond traditional computing, enabling smarter and more personalized experiences that can keep up with AI."

They also added that "No matter how each user embraces and uses AI for their day-to-day, we believe AI PCs mark the future of computing, where AI becomes an intuitive and essential part of everyday workflows."

The message is clear: AI PCs aren't just for AI enthusiasts and developers. Regardless of what you use your laptop for, there's potential for AI to help and soon every new Windows 11 laptop will be an AI PC.

Qualcomm is aiming high as the AI PC market grows, a necessary outlook if it wants to stay ahead of Intel and AMD. That competition is good for users, too, because it will help move the needle on innovation.

A Qualcomm spokesperson gave us some hints about what they have coming up in 2025 and beyond, telling Laptop Mag, "You can expect high performance, deeper integration with edge computing, and continued innovations in battery efficiency and connectivity. Upcoming AI tools will elevate productivity with advanced search functions, enhanced photo resolution, smarter conference call features, and cutting-edge graphic design capabilities."

Those new features and tools sound great, but Intel and AMD are working on AI innovations of their own. We'll just have to wait and see if Qualcomm has what it takes to rule the AI PC market or if it was just the rookie of the year.

If you're anything from an AI enthusiast to the average AI tinkerer (or simply seeking out some of the additional features offered through Windows Copilot+ PCs or Apple Intelligence on Macs), then you'll need a powerful and performative laptop to keep up to speed with your needs.

At Laptop Mag, we review laptops year-round to ensure we're giving you expert-backed and up-to-date recommendations on which notebook is right for you. When it comes to the best AI PC category, our top picks are the excellent Asus Zenbook S 14 (UX5406) for Windows users and the impressive Apple Macbook Air M3 for those running macOS.

So, if you're shopping for a new laptop and looking to invest in an AI PC (or just a great laptop in general), check out our current top-tier picks below.

On June 4, 2024, Pat Gelsinger, then CEO of Intel, came out swinging when he arrived at the highlight of his keynote presentation at Computex in Taipei, Taiwan: AI PCs.

Gelsinger compared his announcement that day to the launch of Wi-Fi 25 years earlier and confidently predicted, "We expect that by 2028, 80% of all PCs will be AI PCs. And Intel is leading the way."

Just moments later, Gelsinger had a literal curtain drop to unveil a dozen upcoming AI PCs. Several of them were running the latest chip from Intel, the star of the show that day: Lunar Lake.

Intel's Lunar Lake chips were the latest additions to its Core Ultra line-up and its most advanced consumer AI chips yet. In his Computex presentation, Gelsinger promised these chips would deliver up to 50% better graphics performance and up to four times stronger NPU AI computing power. In a September 2024 presentation at IFA Berlin, Intel claimed Lunar Lake would have up to 50% lower power consumption.

That set a high bar for Lunar Lake laptops, especially after the success of AMD's Ryzen AI 300 series chips and the launch of Copilot+ PCs with Qualcomm chips this year. However, Intel was up to the challenge and delivered in one crucial area. While there were some key flaws with the Lunar Lake laptops we tested this year, there was also one surprising win for Intel.

Intel dominated on battery life with Lunar Lake, but at what cost?

When our first test results for laptops running on Intel Lunar Lake chips rolled in earlier this year, it was immediately clear that they were battery-life beasts. For instance, the Dell XPS 13 with an Intel Core Ultra 7 258V lasted 18 hours and 34 minutes in our battery life test. That's enough for two full work days on a single charge!

Of course, if you're doing more resource-intensive tasks like gaming or Photoshop, you'll burn through that battery life much faster. However, that score is still incredible, especially for commuters and students relying on strong battery life.

It wasn't just the Dell XPS 13 that crushed our battery life test, either. The Asus Zenbook S14, Asus ExpertBook P5, and Lenovo Yoga Slim 7i Aura Edition boasted over 13 hours of battery life. So, Intel's claims about superior battery life on Lunar Lake are grounded in reality.

However, that doesn't mean there are still no drawbacks to Lunar Lake.

Excellent battery life usually comes at the cost of something else. For instance, a more powerful processor or a more premium display will generally eat through battery life faster than less powerful or premium alternatives. With great battery life comes great compromises.

A perfect example is, ironically, the Dell XPS 13. The non-OLED version lasted 18 hours and 34 minutes in our battery life test. The version with a premium OLED display lasted just 8 hours and 40 minutes, less than half as long.

Display quality isn't the only trade-off we noticed, either. In our review of the Acer Swift 14 AI, we note that Lunar Lake seems to fall behind on multicore performance, which can be a considerable drawback for multi-tasking.

Reviewer Joanna Nelius writes, "Maybe this is the result of Intel ditching hyperthreading on its new processors — a process that improves a computer’s ability to multitask and increases its multicore performance."

These drawbacks are disappointing, but that doesn't mean Intel didn't have any wins with Lunar Lake. Its biggest win will probably surprise you.

The surprising silver lining for Intel Lunar Lake

Lunar Lake laptops might not be perfect, but there was one surprising, last-minute win for Intel's latest series of chips this year: the MSI Claw 8 AI Plus.

If you heard anything about the disastrous original MSI Claw earlier this year, you're probably scratching your head at that. I would be, too.

At least, until I tried out the new-and-improved Claw at IFA Berlin in September. Even in my brief hands-on, I could tell MSI listened to user feedback. The new version of MSI's flagship handheld gaming PC got many updates and design changes in response to its predecessor's rough launch. One of those updates was an Intel Lunar Lake chip.

The MSI Claw 8 AI Plus was one of the best tech glow-ups of the year. While the original Claw received a measly 2.5 stars from us, the Claw 8 AI Plus earned a shining Editor's Choice award, partly thanks to its Lunar Lake processor, which powered top-tier performance and stellar battery life.

Gelsinger left Intel late this year, leaving the legendary company in a state of uncertainty — at least in public, anyway. However, with the success of the MSI Claw AI Plus, it may be taking steps in the right direction.

If you're anything from an AI enthusiast to the average AI tinkerer (or simply seeking out some of the additional features offered through Windows Copilot+ PCs or Apple Intelligence on Macs), then you'll need a powerful and performative laptop to keep up to speed with your needs.

At Laptop Mag, we review laptops year-round to ensure we're giving you expert-backed and up-to-date recommendations on which notebook is right for you. When it comes to the best AI PC category, our top picks are the excellent Asus Zenbook S 14 (UX5406) for Windows users and the impressive Apple Macbook Air M3 for those running macOS.

So, if you're shopping for a new laptop and looking to invest in an AI PC (or just a great laptop in general), check out our current top-tier picks below.