From large language models (LLMs) to reasoning agents, today’s AI tools bring unprecedented computational demands. Trillion-parameter models, workloads running on-device, and swarms of agents collaborating to complete tasks all require a new paradigm of computing to become truly seamless and ubiquitous. Silicon’s mid-life crisis AI has evolved from classical ML to deep learning to generative AI. The most recent chapter, which took AI mainstream, hinges on two phases—training and inference—that are data and energy-intensive in terms of computation, data movement, and cooling. At the same time, Moore’s Law, which determines that the number of transistors on a chip doubles every two years, is reaching a physical and economic plateau. For the last 40 years, silicon chips and digital technology have nudged each other forward—every step ahead in processing capability frees the imagination of innovators to envision new products, which require yet more power to run. That is happening at light speed in the AI age. As models become more readily available, deployment at scale puts the spotlight on inference and the application of trained models for everyday use cases. This transition requires the appropriate hardware to handle inference tasks efficiently. Central processing units (CPUs) have managed general computing tasks for decades, but the broad adoption of ML introduced computational demands that stretched the capabilities of traditional CPUs. This has led to the adoption of graphics processing units (GPUs) and other accelerator chips for training complex neural networks, due to their parallel execution capabilities and high memory bandwidth that allow large-scale mathematical operations to be processed efficiently. But CPUs are already the most widely deployed and can be companions to processors like GPUs and tensor processing units (TPUs). AI developers are also hesitant to adapt software to fit specialized or bespoke hardware, and they favor the consistency and ubiquity of CPUs. Chip designers are unlocking performance gains through optimized software tooling, adding novel processing features and data types specifically to serve ML workloads, integrating specialized units and accelerators, and advancing silicon chip innovations, including custom silicon. AI itself is a helpful aid for chip design, creating a positive feedback loop in which AI helps optimize the chips that it needs to run. These enhancements and strong software support mean modern CPUs are a good choice to handle a range of inference tasks. Beyond silicon-based processors, disruptive technologies are emerging to address growing AI compute and data demands. The unicorn start-up Lightmatter, for instance, introduced photonic computing solutions that use light for data transmission to generate significant improvements in speed and energy efficiency. Quantum computing represents another promising area in AI hardware. While still years or even decades away, the integration of quantum computing with AI could further transform fields like drug discovery and genomics. The developments in ML theories and network architectures have significantly enhanced the efficiency and capabilities of AI models. Today, the industry is moving from monolithic models to agent-based systems characterized by smaller, specialized models that work together to complete tasks more efficiently at the edge—on devices like smartphones or modern vehicles. This allows them to extract increased performance gains, like faster model response times, from the same or even less compute. Researchers have developed techniques, including few-shot learning, to train AI models using smaller datasets and fewer training iterations. AI systems can learn new tasks from a limited number of examples to reduce dependency on large datasets and lower energy demands. Optimization techniques like quantization, which lower the memory requirements by selectively reducing precision, are helping reduce model sizes without sacrificing performance.  New system architectures, like retrieval-augmented generation (RAG), have streamlined data access during both training and inference to reduce computational costs and overhead. The DeepSeek R1, an open source LLM, is a compelling example of how more output can be extracted using the same hardware. By applying reinforcement learning techniques in novel ways, R1 has achieved advanced reasoning capabilities while using far fewer computational resources in some contexts. The integration of heterogeneous computing architectures, which combine various processing units like CPUs, GPUs, and specialized accelerators, has further optimized AI model performance. This approach allows for the efficient distribution of workloads across different hardware components to optimize computational throughput and energy efficiency based on the use case. As AI becomes an ambient capability humming in the background of many tasks and workflows, agents are taking charge and making decisions in real-world scenarios. These range from customer support to edge use cases, where multiple agents coordinate and handle localized tasks across devices. With AI increasingly used in daily life, the role of user experiences becomes critical for mass adoption. Features like predictive text in touch keyboards, and adaptive gearboxes in vehicles, offer glimpses of AI as a vital enabler to improve technology interactions for users. Edge processing is also accelerating the diffusion of AI into everyday applications, bringing computational capabilities closer to the source of data generation. Smart cameras, autonomous vehicles, and wearable technology now process information locally to reduce latency and improve efficiency. Advances in CPU design and energy-efficient chips have made it feasible to perform complex AI tasks on devices with limited power resources. This shift toward heterogeneous compute enhances the development of ambient intelligence, where interconnected devices create responsive environments that adapt to user needs. Seamless AI naturally requires common standards, frameworks, and platforms to bring the industry together. Contemporary AI brings new risks. For instance, by adding more complex software and personalized experiences to consumer devices, it expands the attack surface for hackers, requiring stronger security at both the software and silicon levels, including cryptographic safeguards and transforming the trust model of compute environments. More than 70% of respondents to a 2024 DarkTrace survey reported that AI-powered cyber threats significantly impact their organizations, while 60% say their organizations are not adequately prepared to defend against AI-powered attacks. Collaboration is essential to forging common frameworks. Universities contribute foundational research, companies apply findings to develop practical solutions, and governments establish policies for ethical and responsible deployment. Organizations like Anthropic are setting industry standards by introducing frameworks, such as the Model Context Protocol, to unify the way developers connect AI systems with data. Arm is another leader in driving standards-based and open source initiatives, including ecosystem development to accelerate and harmonize the chiplet market, where chips are stacked together through common frameworks and standards. Arm also helps optimize open source AI frameworks and models for inference on the Arm compute platform, without needing customized tuning.  How far AI goes to becoming a general-purpose technology, like electricity or semiconductors, is being shaped by technical decisions taken today. Hardware-agnostic platforms, standards-based approaches, and continued incremental improvements to critical workhorses like CPUs, all help deliver the promise of AI as a seamless and silent capability for individuals and businesses alike. Open source contributions are also helpful in allowing a broader range of stakeholders to participate in AI advances. By sharing tools and knowledge, the community can cultivate innovation and help ensure that the benefits of AI are accessible to everyone, everywhere. Learn more about Arm’s approach to enabling AI everywhere. This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. This content was researched, designed, and written entirely by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.

close A majority of small businesses are using artificial intelligence A majority of small businesses are using artificial intelligence and finding out it can save time and money. So, you want to start making money using AI but you’re not trying to build Skynet or learn 15 coding languages first? Good, because neither am I. You don’t need to become the next Sam Altman or have a Ph.D. in machine learning to turn artificial intelligence into real income. What you do need is curiosity, a dash of creativity, and the right prompts.💸 Enter to win $500 for you and $500 for your favorite person or charity in our Pay It Forward Sweepstakes. Hurry, ends soon!I’ve pulled together five powerful, practical prompts you can throw into ChatGPT (or your AI tool of choice) to help you start earning extra cash this week. These aren’t pie-in-the-sky dreams or $10K-a-month YouTube ad schemes. They’re doable, even if your calendar is already packed.5-MINUTE CLEANUP FOR YOUR PHONE AND COMPUTERLet’s get to it.1. Fast-Track Your Freelance LifePrompt to use:"Act as a freelance business coach. Suggest 3 services I can offer on Fiverr or Upwork using AI tools like ChatGPT, Midjourney or Canva. I have [insert skill: writing/design/admin/accounting/managerial] experience."Why this works:Freelance work is exploding right now. Platforms like Upwork and Fiverr are filled with small businesses and entrepreneurs who need help—but don’t have the budget to hire full-time staff. If you’ve got any kind of professional background, you can use AI tools to turbocharge your services. Writing blog posts? ChatGPT can give you a draft. Creating logos or social media templates? Midjourney and Canva are your new best friends.You don’t need a team. You don’t need fancy software. You just need a good prompt and the confidence to say, "Yes, I can do that." AI helps you scale what you already know how to do. A man is pictured with a smartphone and laptop computer on January 31, 2019.  (Neil Godwin/Future via Getty Images)2. Make Product Descriptions Sexy AgainPrompt to use:"Rewrite this Etsy or Shopify product description to make it more compelling and SEO-friendly. Target audience: [insert group]. Here’s the original: [paste description]."Why this works:Let’s face it—most product descriptions online are a snooze. But good copy sells. Whether you’re running your own shop or helping someone else with theirs, compelling product descriptions convert clicks into customers. Use ChatGPT to punch up the language, fine-tune for SEO, and speak directly to your ideal buyer.DON’T SCAM YOURSELF WITH THE TRICKS HACKERS DON’T WANT ME TO SHARERemember: people don’t just want to buy a weird mug. They want to buy what it says about them. That’s where a smart rewrite can turn browsers into buyers.3. Social Posts That SellPrompt to use:"Create 5 attention-grabbing Instagram captions to promote this [product/service]. Keep the tone [fun, confident, expert] and include a strong call to action."Why this works:We live in a scroll-happy world. Your social captions need to grab attention in less than three seconds. But not everyone’s a copywriter—and not everyone has time to be. AI can help you crank out engaging content in the tone and style that fits your brand. Add a great photo, post consistently, and you’re suddenly a one-person content agency without the overhead (or endless Zoom meetings). A photo taken on October 4, 2023 in Manta, near Turin, shows a smartphone and a laptop displaying the logos of the artificial intelligence OpenAI research company and ChatGPT chatbot. (MARCO BERTORELLO/AFP via Getty Images)If you’re managing social for clients or your own biz, this prompt is gold. Use it to build content calendars, write reels scripts, or even draft ad copy.4. Polite Emails That Save You MoneyPrompt to use:"Write a short, polite email to ask for a lower rate or discount on [tool/service/platform]. Mention that I’m a loyal customer comparing alternatives."Why this works:Negotiating discounts doesn’t always feel comfortable but it absolutely works. Companies often have unpublished deals, especially for longtime users or small businesses. And customer service reps? They're human beings. A kind, well-written email might be all it takes to get a discount on that software you’re using every month.20 TECH TRICKS TO MAKE LIFE BETTER, SAFER OR EASIERI’ve personally saved hundreds of dollars just by sending quick, respectful emails like this. AI can help you strike the perfect tone confident but kind, assertive but not pushy.5. Your Passive Income KitPrompt to use:"Give me 3 high-demand, low-competition ideas for a short e-book or low-content book I can sell on Amazon. I have experience in [insert topic]."Why this works:You have knowledge people want. Package it. Sell it. Repeat. Whether it’s a short guide on starting a backyard garden or a workbook for productivity hacks, e-books and low-content books (like journals or planners) sell surprisingly well. And AI can help you brainstorm ideas, outline chapters, even draft content to polish up. In this photo illustration the logo of Apple Mail Programme Mail can be seen on a smartphone next to a finger on March 27, 2024 in Berlin, Germany. (Photo Illustration by Thomas Trutschel/Photothek via Getty Images)Upload it to Amazon KDP or Gumroad, and now you’ve got a digital product that can earn money in your sleep. People pay for convenience, and you have life experience worth sharing.Final ThoughtYou don’t need to master AI to start earning with it. You just need to start using it. These five prompts are a low-risk, high-potential way to get your feet wet. And if you need a hand turning these sparks into something bigger, I’m here.I built my multimillion-dollar business with no investors and no debt. I’ve done this without a big team or expensive consultants. And I’d love to help you do the same.Drop me a note. I read every one.Get tech-smarter on your scheduleAward-winning host Kim Komando is your secret weapon for navigating tech.National radio: Airing on 500+ stations across the US - Find yours or get the free podcast.Daily newsletter: Join 650,000 people who read the Current (free!)Watch: On Kim’s YouTube channelCopyright 2025, WestStar Multimedia Entertainment. All rights reserved. 

From large language models (LLMs) to reasoning agents, today’s AI tools bring unprecedented computational demands. Trillion-parameter models, workloads running on-device, and swarms of agents collaborating to complete tasks all require a new paradigm of computing to become truly seamless and ubiquitous. Silicon’s mid-life crisis AI has evolved from classical ML to deep learning to generative AI. The most recent chapter, which took AI mainstream, hinges on two phases—training and inference—that are data and energy-intensive in terms of computation, data movement, and cooling. At the same time, Moore’s Law, which determines that the number of transistors on a chip doubles every two years, is reaching a physical and economic plateau. For the last 40 years, silicon chips and digital technology have nudged each other forward—every step ahead in processing capability frees the imagination of innovators to envision new products, which require yet more power to run. That is happening at light speed in the AI age. As models become more readily available, deployment at scale puts the spotlight on inference and the application of trained models for everyday use cases. This transition requires the appropriate hardware to handle inference tasks efficiently. Central processing units (CPUs) have managed general computing tasks for decades, but the broad adoption of ML introduced computational demands that stretched the capabilities of traditional CPUs. This has led to the adoption of graphics processing units (GPUs) and other accelerator chips for training complex neural networks, due to their parallel execution capabilities and high memory bandwidth that allow large-scale mathematical operations to be processed efficiently. But CPUs are already the most widely deployed and can be companions to processors like GPUs and tensor processing units (TPUs). AI developers are also hesitant to adapt software to fit specialized or bespoke hardware, and they favor the consistency and ubiquity of CPUs. Chip designers are unlocking performance gains through optimized software tooling, adding novel processing features and data types specifically to serve ML workloads, integrating specialized units and accelerators, and advancing silicon chip innovations, including custom silicon. AI itself is a helpful aid for chip design, creating a positive feedback loop in which AI helps optimize the chips that it needs to run. These enhancements and strong software support mean modern CPUs are a good choice to handle a range of inference tasks. Beyond silicon-based processors, disruptive technologies are emerging to address growing AI compute and data demands. The unicorn start-up Lightmatter, for instance, introduced photonic computing solutions that use light for data transmission to generate significant improvements in speed and energy efficiency. Quantum computing represents another promising area in AI hardware. While still years or even decades away, the integration of quantum computing with AI could further transform fields like drug discovery and genomics. The developments in ML theories and network architectures have significantly enhanced the efficiency and capabilities of AI models. Today, the industry is moving from monolithic models to agent-based systems characterized by smaller, specialized models that work together to complete tasks more efficiently at the edge—on devices like smartphones or modern vehicles. This allows them to extract increased performance gains, like faster model response times, from the same or even less compute. Researchers have developed techniques, including few-shot learning, to train AI models using smaller datasets and fewer training iterations. AI systems can learn new tasks from a limited number of examples to reduce dependency on large datasets and lower energy demands. Optimization techniques like quantization, which lower the memory requirements by selectively reducing precision, are helping reduce model sizes without sacrificing performance.  New system architectures, like retrieval-augmented generation (RAG), have streamlined data access during both training and inference to reduce computational costs and overhead. The DeepSeek R1, an open source LLM, is a compelling example of how more output can be extracted using the same hardware. By applying reinforcement learning techniques in novel ways, R1 has achieved advanced reasoning capabilities while using far fewer computational resources in some contexts. The integration of heterogeneous computing architectures, which combine various processing units like CPUs, GPUs, and specialized accelerators, has further optimized AI model performance. This approach allows for the efficient distribution of workloads across different hardware components to optimize computational throughput and energy efficiency based on the use case. As AI becomes an ambient capability humming in the background of many tasks and workflows, agents are taking charge and making decisions in real-world scenarios. These range from customer support to edge use cases, where multiple agents coordinate and handle localized tasks across devices. With AI increasingly used in daily life, the role of user experiences becomes critical for mass adoption. Features like predictive text in touch keyboards, and adaptive gearboxes in vehicles, offer glimpses of AI as a vital enabler to improve technology interactions for users. Edge processing is also accelerating the diffusion of AI into everyday applications, bringing computational capabilities closer to the source of data generation. Smart cameras, autonomous vehicles, and wearable technology now process information locally to reduce latency and improve efficiency. Advances in CPU design and energy-efficient chips have made it feasible to perform complex AI tasks on devices with limited power resources. This shift toward heterogeneous compute enhances the development of ambient intelligence, where interconnected devices create responsive environments that adapt to user needs. Seamless AI naturally requires common standards, frameworks, and platforms to bring the industry together. Contemporary AI brings new risks. For instance, by adding more complex software and personalized experiences to consumer devices, it expands the attack surface for hackers, requiring stronger security at both the software and silicon levels, including cryptographic safeguards and transforming the trust model of compute environments. More than 70% of respondents to a 2024 DarkTrace survey reported that AI-powered cyber threats significantly impact their organizations, while 60% say their organizations are not adequately prepared to defend against AI-powered attacks. Collaboration is essential to forging common frameworks. Universities contribute foundational research, companies apply findings to develop practical solutions, and governments establish policies for ethical and responsible deployment. Organizations like Anthropic are setting industry standards by introducing frameworks, such as the Model Context Protocol, to unify the way developers connect AI systems with data. Arm is another leader in driving standards-based and open source initiatives, including ecosystem development to accelerate and harmonize the chiplet market, where chips are stacked together through common frameworks and standards. Arm also helps optimize open source AI frameworks and models for inference on the Arm compute platform, without needing customized tuning.  How far AI goes to becoming a general-purpose technology, like electricity or semiconductors, is being shaped by technical decisions taken today. Hardware-agnostic platforms, standards-based approaches, and continued incremental improvements to critical workhorses like CPUs, all help deliver the promise of AI as a seamless and silent capability for individuals and businesses alike. Open source contributions are also helpful in allowing a broader range of stakeholders to participate in AI advances. By sharing tools and knowledge, the community can cultivate innovation and help ensure that the benefits of AI are accessible to everyone, everywhere. Learn more about Arm’s approach to enabling AI everywhere. This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. This content was researched, designed, and written entirely by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.

College graduates were left speechlessA video that has since gone viral on social media shows students having a QR on their smartphone scanned, only to have an AI read their name out loud — an irritating use of the tech that turned what should've otherwise been a proud moment in their academic career into what commenters likened to using the self-checkout counter at the supermarket.It's especially galling for an institution that charges north of $50,000 a year in tuition alone."So even public event speakers are getting replaced," one Reddit user wrote. As the New York Post points out, Pace University ostensibly performed the stunt for the sake of accuracy. Well ahead of the event, officials directed students to visit a website where they could phonetically spell their names and confirm how they were pronounced."To ensure your name is pronounced correctly, you must register to attend Commencement through Tassel by Wednesday, April 23," reads an FAQ on the university's website.Tassel is a private company that claims to have helped more than 600 schools "deliver the perfect moment for millions of students over 20+ years."The company offers "live ceremony stage-crossing software" designed to streamline convocation events and ensure "accurate pronunciation" of graduate names."With AI and our proprietary multilingual models — powered by a growing database of over 2 million names, phonemes, and linguistic patterns across multiple languages — our platform allows students to hear exactly how their names will be pronounced as they cross the stage," Tassel boasts on its website.If the AI butchers their name even after it was coached by the student, graduates can "submit an audio recording of their name" instead.The software raises some thorny questions about when it's truly appropriate to deploy an AI, particularly concerning jobs that could've easily been taken care of by a human."When I graduated, the person reading out the names had done a run through, just to check they were getting it right," one user wrote in a post on Threads. "It didn't take that long, and was less weird than showing your phone for them to scan."Worst of all, despite all those technical bells and whistles, Tassel reportedly still managed to butcher some names."They told us to write our names phonetically so it's said correctly, and they still said my name wrong, which is forever documented in videos," one user commented on Instagram.The use of AI to read student names has been met with significant blowback from college grads across the country.Student journalists at Northeastern University, which also deployed Tassel, published an op-ed arguing that the institution should "read our names at graduation, it’s the least you can do.""To many students like myself, a small switch from an AI voice bot to a physical person would mean the world," reads the op-ed, penned by student Henry Bova and published in The Huntington News. "A human can reciprocate our sense of joy and can understand the gravity of the moment and the academic rigor that we endured to get to this point.""All the AI voice does is reduce the recognition of our accomplishments as a menial task to push through with brute computations," Bova wrote.An online petition by students at the University of North Georgia called on the university to stop "using an AI speaker for graduation," garnering over 2,000 signatures.The backlash was so fierce in some parts of the country that West Chester University in Pennsylvania ditched the AI altogether, as Axios reported earlier this month.Other onlookers pointed out the irony of AI coming for the jobs of recent college graduates in particular."A whole lifetime of studies and 100k in debt just to have your name announced by the very entity that's going to make your studies useless," one Reddit user joked.However, not everybody was opposed to the use of the tech."I appreciate that everyone had the chance to hear their name pronounced correctly," one Reddit user argued. "Imagine being an international student and worried that someone will butcher your name. I'm ok with the solution the university gave here."Share This Article

Heads up if you or someone you know is still rocking an older iPhone: WhatsApp is officially dropping support for several legacy smartphones starting this Sunday, June 1st. As part of its regular update cycle, the Meta-owned messaging app is raising its minimum requirements. Originally, the change had been planned for May 5th. However, it seems the new cutoff date was pushed by a few weeks and landed on June 1st. So starting this Sunday, WhatsApp will only work on iPhones running iOS 15.1 or later. Devices stuck on Android 5.0 or earlier are also being dropped. In practice, this means WhatsApp will no longer work on the following iPhone models: iPhone 5s iPhone 6 iPhone 6 Plus Some reports have included these models in the list: iPhone 6s iPhone 6s Plus iPhone SE (1st gen) However, these devices support up to iOS 15.8.4. So if they’re up to date with the latest supported iOS version, they should be good until the next WhatsApp cutoff update, which likely won’t happen for another year or two. While these phones are long past their prime, some are still kicking around as secondary or hand-me-down devices for less tech-inclined users. That may not be your case, but with approximately 3 billion users around the world, the change is bound to catch a few people off guard. Add 9to5Mac to your Google News feed.  FTC: We use income earning auto affiliate links. More.You’re reading 9to5Mac — experts who break news about Apple and its surrounding ecosystem, day after day. Be sure to check out our homepage for all the latest news, and follow 9to5Mac on Twitter, Facebook, and LinkedIn to stay in the loop. Don’t know where to start? Check out our exclusive stories, reviews, how-tos, and subscribe to our YouTube channel

From large language models (LLMs) to reasoning agents, today’s AI tools bring unprecedented computational demands. Trillion-parameter models, workloads running on-device, and swarms of agents collaborating to complete tasks all require a new paradigm of computing to become truly seamless and ubiquitous. First, technical progress in hardware and silicon design is critical to pushing the boundaries of compute. Second, advances in machine learning (ML) allow AI systems to achieve increased efficiency with smaller computational demands. Finally, the integration, orchestration, and adoption of AI into applications, devices, and systems is crucial to delivering tangible impact and value. Silicon’s mid-life crisis AI has evolved from classical ML to deep learning to generative AI. The most recent chapter, which took AI mainstream, hinges on two phases—training and inference—that are data and energy-intensive in terms of computation, data movement, and cooling. At the same time, Moore’s Law, which determines that the number of transistors on a chip doubles every two years, is reaching a physical and economic plateau. For the last 40 years, silicon chips and digital technology have nudged each other forward—every step ahead in processing capability frees the imagination of innovators to envision new products, which require yet more power to run. That is happening at light speed in the AI age. As models become more readily available, deployment at scale puts the spotlight on inference and the application of trained models for everyday use cases. This transition requires the appropriate hardware to handle inference tasks efficiently. Central processing units (CPUs) have managed general computing tasks for decades, but the broad adoption of ML introduced computational demands that stretched the capabilities of traditional CPUs. This has led to the adoption of graphics processing units (GPUs) and other accelerator chips for training complex neural networks, due to their parallel execution capabilities and high memory bandwidth that allow large-scale mathematical operations to be processed efficiently. But CPUs are already the most widely deployed and can be companions to processors like GPUs and tensor processing units (TPUs). AI developers are also hesitant to adapt software to fit specialized or bespoke hardware, and they favor the consistency and ubiquity of CPUs. Chip designers are unlocking performance gains through optimized software tooling, adding novel processing features and data types specifically to serve ML workloads, integrating specialized units and accelerators, and advancing silicon chip innovations, including custom silicon. AI itself is a helpful aid for chip design, creating a positive feedback loop in which AI helps optimize the chips that it needs to run. These enhancements and strong software support mean modern CPUs are a good choice to handle a range of inference tasks. Beyond silicon-based processors, disruptive technologies are emerging to address growing AI compute and data demands. The unicorn start-up Lightmatter, for instance, introduced photonic computing solutions that use light for data transmission to generate significant improvements in speed and energy efficiency. Quantum computing represents another promising area in AI hardware. While still years or even decades away, the integration of quantum computing with AI could further transform fields like drug discovery and genomics. Understanding models and paradigms The developments in ML theories and network architectures have significantly enhanced the efficiency and capabilities of AI models. Today, the industry is moving from monolithic models to agent-based systems characterized by smaller, specialized models that work together to complete tasks more efficiently at the edge—on devices like smartphones or modern vehicles. This allows them to extract increased performance gains, like faster model response times, from the same or even less compute. Researchers have developed techniques, including few-shot learning, to train AI models using smaller datasets and fewer training iterations. AI systems can learn new tasks from a limited number of examples to reduce dependency on large datasets and lower energy demands. Optimization techniques like quantization, which lower the memory requirements by selectively reducing precision, are helping reduce model sizes without sacrificing performance.  New system architectures, like retrieval-augmented generation (RAG), have streamlined data access during both training and inference to reduce computational costs and overhead. The DeepSeek R1, an open source LLM, is a compelling example of how more output can be extracted using the same hardware. By applying reinforcement learning techniques in novel ways, R1 has achieved advanced reasoning capabilities while using far fewer computational resources in some contexts. The integration of heterogeneous computing architectures, which combine various processing units like CPUs, GPUs, and specialized accelerators, has further optimized AI model performance. This approach allows for the efficient distribution of workloads across different hardware components to optimize computational throughput and energy efficiency based on the use case. Orchestrating AI As AI becomes an ambient capability humming in the background of many tasks and workflows, agents are taking charge and making decisions in real-world scenarios. These range from customer support to edge use cases, where multiple agents coordinate and handle localized tasks across devices. With AI increasingly used in daily life, the role of user experiences becomes critical for mass adoption. Features like predictive text in touch keyboards, and adaptive gearboxes in vehicles, offer glimpses of AI as a vital enabler to improve technology interactions for users. Edge processing is also accelerating the diffusion of AI into everyday applications, bringing computational capabilities closer to the source of data generation. Smart cameras, autonomous vehicles, and wearable technology now process information locally to reduce latency and improve efficiency. Advances in CPU design and energy-efficient chips have made it feasible to perform complex AI tasks on devices with limited power resources. This shift toward heterogeneous compute enhances the development of ambient intelligence, where interconnected devices create responsive environments that adapt to user needs. Seamless AI naturally requires common standards, frameworks, and platforms to bring the industry together. Contemporary AI brings new risks. For instance, by adding more complex software and personalized experiences to consumer devices, it expands the attack surface for hackers, requiring stronger security at both the software and silicon levels, including cryptographic safeguards and transforming the trust model of compute environments. More than 70% of respondents to a 2024 DarkTrace survey reported that AI-powered cyber threats significantly impact their organizations, while 60% say their organizations are not adequately prepared to defend against AI-powered attacks. Collaboration is essential to forging common frameworks. Universities contribute foundational research, companies apply findings to develop practical solutions, and governments establish policies for ethical and responsible deployment. Organizations like Anthropic are setting industry standards by introducing frameworks, such as the Model Context Protocol, to unify the way developers connect AI systems with data. Arm is another leader in driving standards-based and open source initiatives, including ecosystem development to accelerate and harmonize the chiplet market, where chips are stacked together through common frameworks and standards. Arm also helps optimize open source AI frameworks and models for inference on the Arm compute platform, without needing customized tuning.  How far AI goes to becoming a general-purpose technology, like electricity or semiconductors, is being shaped by technical decisions taken today. Hardware-agnostic platforms, standards-based approaches, and continued incremental improvements to critical workhorses like CPUs, all help deliver the promise of AI as a seamless and silent capability for individuals and businesses alike. Open source contributions are also helpful in allowing a broader range of stakeholders to participate in AI advances. By sharing tools and knowledge, the community can cultivate innovation and help ensure that the benefits of AI are accessible to everyone, everywhere. Learn more about Arm’s approach to enabling AI everywhere. This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. This content was researched, designed, and written entirely by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.

If you're entrenched in tech news, you'd think Apple was on the brink of collapse. The company undoubtedly is having a rough go of all things AI—while companies like ChatGPT, Google, and Microsoft have hit the AI ground running, Apple's AI department is in disarray. Some features, like Clean Up and Writing Tools, have made their way to products like the iPhone, but, others (notably Siri's AI overhaul) are still nowhere to be seen. The situation is, objectively, not great. Apple advertised these features alongside the iPhone 16 line, even casting The Last of Us' Bella Ramsey in a commercial showing off said AI-powered Siri. (The commercial has since been deleted.) While the rest of the tech industry seems to be entirely focused on AI, Apple is, uncharacteristically, struggling to keep up. Things must be dire for the company, right?The iPhone continues to sell like hot cakesWhile I'm not here to read the company's entire pulse, it does seem like the iPhone department is still crushing it. On Wednesday, market research firm Counterpoint released its list of the top-selling smartphones in Q1 of 2025. Perhaps unsurprisingly, the iPhone topped the list: Apple's iPhone 16 was the best-selling smartphone in the world in the first fiscal quarter of this year, followed by the 16 Pro Max, 16 Pro, and iPhone 15. Apple also had the top four spots in the first quarter of 2024—back then, it was the 15 Pro Max in first place, followed by the 15, 15 Pro, and 14.Samsung took the next three spots, as it did in Q1 of 2024 as well. This year, it was the Galaxy A16 5G in fifth place, followed by the Galaxy A06 and the Galaxy S25 Ultra. The Redmi 14C 4G came in eighth—impressive for a smartphone that isn't even sold in the U.S.—followed by the Galaxy A55 5G, and, finally, the iPhone 16 Plus. There's a lot you can take away from the data here. The first impression is that the iPhone continues to be a global force to be reckoned with. The iPhone had five of the top 10 spots in both Q1 2024 and 2025—the only difference between them was the iPhone 15 Plus came in eighth place, while the 16 Plus came in 10th. Samsung, too, is clearly still a reigning champ in the global smartphone race, though it went from five phones in the top 10 to four between those two years—good for Redmi for stealing that eighth place spot.Ecosystems are powerful thingsIt's particularly interesting to see the iPhone continue perform like this in 2025. After all, it's been apparent for months now that Apple did not follow through on its advertised AI promises for the iPhone 16 line. To wit, Counterpoint says that the iPhone 16e, the company's "more affordable" device, ranked sixth in the top selling smartphones of March. People are continuing to buy iPhones in droves. Is it possible these customers are buying iPhones based on Apple's past advertisements? Sure. The company still advertises Apple Intelligence with each iPhone on its site, so AI could still be driving people's desires to buy iPhones. I'm not convinced, though. If AI were a priority, I think most customers would be buying from the companies that have been rolling out AI features at a steady clip. Samsung and Google immediately come to mind: Google's latest I/O event was all about AI, and you can experience a number of AI features on Android devices made by both companies. Again, maybe Samsung's four "top 10" smartphones are a result of its AI efforts. It's entirely possible, but I continue to be unconvinced.I see this list of best-selling iPhones and Galaxies, and I see one thing: established market trends. I think the truth is, a lot of people like Galaxies, and even more people like iPhones. People switch phones all the time, especially in the Android ecosystem, but based on the data, it seems like when it's time to buy a new phone, most iPhone users buy a new iPhone, and most Galaxy users buy a new Galaxy. Ecosystems are powerful things, and when you've poured your entire digital life into one platform—including all the messaging, purchases, and cloud storage—it's rare you want to mix it up. That's me to a T: As much as I respect Android, I'm stuck in the Apple ecosystem, and, as such, really only consider a new iPhone when it comes time to upgrade. Almost every single person in my immediate circle is the same way. The Samsung fans I know also stick to the pattern, just with the newest Galaxy. The decision for me is never whether to buy an iPhone or a Galaxy: it's whether to buy the Pro or the Pro Max. AI enthusiasm isn't strong enough to drive smartphone salesAI is without a doubt the trend in tech right now, and people are using it. But I don't think many are considering it when buying their devices—especially smartphones. I think people buy the phone they like, and then configure it after the fact to access their AI tools. Hell, Apple integrated ChatGPT into my iPhone, and I still have the ChatGPT app. AI features can be useful—it's great that Apple has its own version of Magic Eraser now—but AI features alone aren't enough to sway customers en masse. If OpenAI made a smartphone, would you buy it? I'm guessing probably not. If the AI train continues on, maybe people will start buying the phones and devices that best integrate AI tools out of the box. Android is way ahead of Apple on this front—just look at Google replacing its assistant with Gemini—so perhaps we'll see Galaxy phones take more of a lead in global sales in future quarters, or even an appearance from a Pixel or two. Or, maybe people are fine downloading the apps they need to get their AI fix, and leaving other factors in play when choosing a phone to buy.I can't predict the future; I can only note what I see in the present. And, right now, I'm seeing two things at once—I'm seeing a lot of people talking about ChatGPT, and I'm seeing a lot of people buying and using iPhones. Outside of my tech news circles, I've heard not a peep about Apple's struggles in the AI race.