He succeeds Srini Gopalan, who was named CEO of the Bellevue, Wash., telecom giant in a surprise move that took effect last month. That business became T-Mobile after Germany's Deutsche Telekom took over as majority shareholder in 2001. “[Berezhnyy] brings deep technical expertise, a track record of building strong teams, and a bold vision for how AI will shape the future of renting,” RentSpree posted on LinkedIn. — Paige Johnson has left her role as Microsoft's vice president of Education. She is relaunching EdCatalyst Group, an Oregon-based consulting business that she previously ran for nearly three years that supports companies, nonprofits and public organizations in using AI to expand their impact. Earlier in her career, Johnson was with Intel for nearly two decades, creating and scaling a professional development program that trained millions of teachers worldwide. — James Newell is chief financial officer of WayTrade, a commodity trading company focused on renewable fuels including sustainable aviation fuel. Newell, who will work remotely from Seattle, was previously a general partner with Voyager Capital, an investor in early stage companies in the Pacific Northwest. — After 15 years with Amazon Web Services, Julien Ellie has resigned from his job as senior principal engineer. Ellie praised his colleagues who helped shape cloud computing, but said the company he joined and what AWS has become are no longer the same. “From where I sit, process has taken precedence over customers, and rules have replaced high judgment. Prior to Amazon, Ellie was at Microsoft for nearly a decade. — Jonathan Assayag has left his Sunnyvale, Calif., role with Amazon where he served as general manager and director of the company's smart eyewear program. During more than nine years at the tech giant, Assayag worked on products including Echo Frames and Smart Delivery Glasses. “These were true zero-to-one efforts that pushed ambient computing, Voice AI, and AI-assisted workflows into new territory. — Gravyty, a Seattle-based company that facilitates alumni donations and higher ed student engagement, named Lisa Haubenstock as its new chief customer officer. “Gravyty presents an opportunity to tie together so much of my previous experience with a truly dedicated global team working to build something great,” Haubenstock said on LinkedIn. “What started as an idea to give families peace of mind has evolved into a company shaping how technology supports safety and quality during the most important years of a child's development,” Lewison said on LinkedIn. — CreateMe, a California-based clothing manufacturer using robotic assembly lines, announced two leadership changes: Her past roles include leadership positions at Peleton Interactive, Square, Yahoo and elsewhere. Your company's success depends on finding and keeping great talent—whether it's an all-star engineer or a world-class salesperson. That's why GeekWire and Prime Team Partners have teamed up on GeekWork, a new initiative designed to help employers connect with outstanding candidates. Prime Team is redefining staffing with a blend of AI and human expertise, while GeekWire's trusted network and reach into the tech community open doors to top talent. Together, we're committed to solving your company's toughest hiring challenges. Learn more about GeekWork by contacting GeekWire co-founder John Cook at [email protected]. Former Medio, Amazon and Expedia leaders launch new AI-focused investment firm Tech Moves: Washington names broadband leader; Greater Seattle Partners gets interim president/CEO; Microsoft legal exec departs Tech Moves: Ex-Payscale CEO Scott Torrey joins Smartsheet; Apple taps Microsoft VP to lead AI efforts Longtime T-Mobile exec joins Amazon's Project Kuiper business as new CMO Tech Moves: Google VP departs; T-Mobile shakeup; Nvidia's new product leader; and more Tech Moves: Expedia names first AI chief; Textio founder joins Microsoft; T-Mobile exec departs

When you purchase through links on our site, we may earn an affiliate commission. Earlier this week, Dell was wrongly flagged for charging even more than Apple for memory upgrades in their laptops, which later turned out to be a misinformed call; nevertheless, Framework was quick to pounce on the opportunity and call out fellow vendors, while warning its own RAM price hikes to follow soon. We are going to need to increase our memory pricing soon, but we won't use this as an excuse to gouge customers like @Dell apparently has and that @Apple does as their norm.December 9, 2025 The story begins with YouTuber Max Tech, aka Vadim Yuryev, who took to X in a now-deleted post to express his disappointment with laptop memory prices. However, as mentioned, this screenshot was either actually altered or, more likely, showed a $550 price hike because memory wasn't the only thing being upgraded. See, Dell's laptop configurator works in a confusing manner where selecting certain options triggers others to automatically change, too. For instance, the jump from 16 GB to 32 GB RAM may have come with a jump in SSD capacity as well, or even a higher spec'd CPU, but since these changes happened silently in the background, they look like standalone charges for the RAM alone. Try playing around with the configurator yourself to discover just how much of a maze it is, though you can still tell what's changing. A pricing error or perhaps a currency/region issue in the original screenshot also can't be ruled out, but one thing is certain: Dell definitely isn't charging $550 for 32GB of RAM. The OP must've realized his oversight after replies and media coverage pointed it out, at which point he deleted the original post. Unfortunately, Framework already replied with a scathing post, letting users know that it'll need to raise memory prices soon as well, but that it won't overcharge customers like "apparently" Dell and Apple. A point made based on erroneous information, but perhaps deserved nonetheless. Framework's tweet remains up, but the post above it from Vadim is gone. However, Framework's point that its memory upgrades are more affordable than rival laptop makers (for now) still stands. "But we won't use this as an excuse to gouge customers like Dell apparently has, and that Apple does as their norm," the company added. A few weeks ago, the company stopped selling standalone RAM in order to stop the supply from being consumed by scalpers. Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds. Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he's not working, you'll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun. Tom's Hardware is part of Future US Inc, an international media group and leading digital publisher.

Bananas, blackberries, limes and more — Amazon's same-day delivery of perishable grocery items has expanded across the U.S. to more than 2,300 cities and towns, the company announced Wednesday. Amazon launched the expanded offering earlier this year, allowing customers to integrate items such as fresh produce, seafood, milk and more into their regular shopping orders of electronics, books, clothes, and household essentials. Analysts say Amazon's rapid expansion of same-day delivery for fresh groceries marks a meaningful escalation in its long-term push into perishable food — a category where it has historically lagged. The company's scale and logistics reach allow it to serve both small towns and large metro areas without relying on third-party platforms, according to Wedbush, which noted in a report Wednesday that the expansion ramps up competitive pressure on Instacart, whose business relies heavily on grocery delivery subscriptions. Analysts also say the stronger grocery offering makes Amazon Prime's $139 annual membership more attractive, while weakening the value of Instacart+, which recently scaled back key subscriber benefits. Amazon is also increasingly competing with the grocery efforts of Uber Eats and DoorDash, according to Wedbush. Since this summer, Amazon has expanded the fresh grocery selection available for same-day delivery by more than 30%, including offerings from Whole Foods Market and Amazon Grocery, the company's new private brand, which now includes over 1,000 items, with most priced under $5. GeekWire tested the same-day service in June, placing a late-night order that arrived in the morning and included apples, cucumbers, and blueberries alongside non-perishable items. The expansion of same-day perishable delivery comes as Amazon is also testing a new ultra-fast option called Amazon Now, which promises delivery in 30 minutes or less. GeekWire tested this service earlier this month, and received a frozen pizza, hummus, bread and more in 23 minutes, from order click to delivery drop-off in Seattle. Amazon has for years been expanding and experimenting in the grocery space. The chips powering your smart TV, voice assistant, tablet, and car all have something in common: MediaTek AI goes from tool to teammate: Amazon Web Services SVP Colleen Aubrey on the dawn of agentic work Amazon rolls out expansion of same-day groceries to 1,000 cities, includes perishable items Fresh produce with your package: Amazon testing tighter grocery bundling for same-day deliveries How Amazon is bringing name brands to Whole Foods, without putting them on the shelves Whole Foods CEO's role expands to oversee Amazon's entire worldwide grocery initiatives

Google is testing AI-powered article overviews on participating publications' Google News pages as part of a new pilot program, the search giant announced on Wednesday. The purpose of the new commercial partnership program is to “explore how AI can drive more engaged audiences,” Google said in a blog post. As part of the new AI pilot program, the company will work with publishers to experiment with new features in Google News. While AI-generated summaries may lead to fewer clicks on news articles, publications participating in the commercial pilot program will receive direct payments from Google, which could make up for the potential decrease in traffic to their sites. The AI-powered article overviews will only appear on participating publications' Google News pages, and not anywhere else on Google News or in Search. This isn't the first time that Google has introduced AI summaries for news. Google is also experimenting with audio briefings for people who prefer listening to the news rather than reading it, as part of the new pilot program. The company says these features will include clear attribution and a link to articles. Additionally, Google is partnering with organizations such as Estadão, Antara, Yonhap, and The Associated Press to incorporate real-time information and enhance results in the Gemini app. As part of Google's Wednesday announcement, the company said that it's launching its “Preferred Sources” feature globally after first launching it in the U.S. and India in August. The feature allows users to select their favorite news sites and blogs to appear in the Top Stories section of Google search results. In the coming days, the feature will be available for English-language users worldwide, and Google plans to roll it out to all supported languages early next year. Google also announced that it's increasing the number of inline links in AI Mode. Additionally, it's introducing “contextual introductions” for embedded links, which are brief explanations that explain why a link could be useful to explore. Prior to joining the publication in 2021, she was a telecom reporter at MobileSyrup. Claude Code is coming to Slack, and that's a bigger deal than it sounds Creator IShowSpeed sued for allegedly punching, choking viral humanoid Rizzbot SpaceX reportedly in talks for secondary sale at $800B valuation, which would make it America's most valuable private company

When you purchase through links on our site, we may earn an affiliate commission. The U.S. government has formally approved the export of Nvidia's high-performance H200 AI chips to China, reinstating access to a class of silicon previously barred under national security rules. Sales will be allowed to select Chinese customers pending government review, and each chip must be routed through U.S. territory for inspection and accompanied by a 25% import duty. The move ends a freeze that ultimately led to Nvidia losing its entire Chinese market share and upended development plans for large-scale AI models in the region. The announcement, first made by President Donald Trump via Truth Social, comes after months of internal debate within the U.S. administration over how to apply export restrictions without accelerating China's ability to develop domestic alternatives. The H200, a powerful GPU from Nvidia's Hopper generation, significantly outperforms the previously approved H20 and had been considered too capable for Chinese markets under earlier rules. Its reauthorization hints that there's been a shift in Washington's approach, namely to maintain technological superiority, but it will allow controlled access to limit the pace of Chinese self-sufficiency. The decision to allow H200 exports follows concerns that sweeping restrictions were producing unintended results. Since the initial wave of AI chip bans in 2022, Chinese firms have intensified development of homegrown accelerators, with Huawei's Ascend 910C making significant progress in training and inference workloads. While still behind Nvidia in absolute performance, Huawei's architecture is increasingly seen as serviceable for national and commercial deployments, especially as its fabrication partners improve yields. Internal discussions in Washington, as reported by multiple sources, shifted focus from outright denial to managed access. Officials seem to have eventually concluded that permitting H200 sales — while keeping newer architectures like Blackwell and the upcoming Rubin out of reach — could slow China's push for chip independence without giving away the U.S. lead in performance. The H200 remains one generation behind Nvidia's cutting-edge designs but is fully capable of training modern foundation models and large-scale AI systems. This repositioning allows U.S. regulators to reassert some control over China's access to advanced silicon while capturing financial and political value from every sale. Two sources present at those meetings told The Information that regulators are considering import limits tied to domestic procurement. Companies may be required to demonstrate that they are also investing in Chinese accelerators such as Huawei's Ascend or Cambricon's Siyuan series. This model, used previously to guide purchases of data center components and server platforms, allows Beijing to balance near-term hardware needs with its goal of reducing reliance on U.S. suppliers. The H200 presents a different scenario, given its substantial performance advantage, but officials are expected to apply similar discretion in determining which sectors and use cases are eligible. SuperCloud, a domestic cloud services provider, confirmed that it expects major Chinese companies to proceed with H200 purchases if permitted, though in a subdued fashion. “The training of leading Chinese AI models still relies on Nvidia cards,” said Zhang Yuchun, a general manager at SuperCloud, when speaking to Reuters. Nvidia's ability to meet demand in China will likely be constrained by supply. Also speaking to Reuters, two sources familiar with Nvidia's supply chain said that H200 manufacturing has been deprioritized in favor of fulfilling high-margin orders from U.S. hyperscalers and sovereign AI programs in allied countries. If all this goes ahead, the H200 would reintroduce training-scale compute capacity to Chinese AI developers after a long period of reliance on repurposed hardware and grey-market acquisitions. Unlike most Chinese accelerators, the H200 supports Nvidia's CUDA software ecosystem, simplifying model porting and cluster integration. Ultimately, Nvidia is unlikely to resume large-scale H200 production unless it receives strong approval from both U.S. and Chinese regulators, and equally strong demand from Chinese customers. U.S. officials have attached a number of technical and procedural conditions to the H200 export approval. Each unit must be shipped to the U.S. before entering China, creating a traceable logistics trail that allows U.S. Customs and Commerce Department inspectors to verify compliance. Nvidia has also introduced optional location-verification software that can confirm whether a chip is operating in an authorized geography. Nvidia has stated that the system, which will first be made available on Blackwell chips, cannot be used for eavesdropping and does not transmit any user data to third parties. The compliance measures reflect a broader concern that export controls, while useful on paper, have proven difficult to enforce. We've seen several examples of banned Nvidia GPUs like the A100 and H100 appearing in the likes of Chinese university research and start-up product documentation since their respective bans. Many of these were likely acquired through grey-market channels or indirect resellers in third countries. Cases are ongoing in multiple districts, and officials have signaled that enforcement will intensify as more tracking capabilities become available. The approval of H200 exports arguably represents a concession rather than any long-term structural change in U.S. policy. The H200, while advanced, is no longer Nvidia's leading product, and allowing access to it gives Washington a tool to slow China's push for independence without giving up the top tier of its technology stack. AI developers can resume training at scale using supported hardware and familiar toolchains, but the long-term imperative remains unchanged. China continues to invest in domestic foundries, chip design houses, and packaging capacity. Huawei is expanding its Ascend production targets, and other firms are attempting to design new architectures that bypass the limitations of restricted U.S. IP. Chinese firms must navigate what could be a complex approvals process, limited supply, and reputational risk. Nvidia, meanwhile, must manage Washington's expectations while preventing its technology from slipping beyond authorized use. The success or failure of this arrangement may determine how future export rules are written. Tom's Hardware is part of Future US Inc, an international media group and leading digital publisher.

In a new preprint titled “It's About Time: The Copilot Usage Report 2025,” Microsoft AI researchers analyzed 37.5 million de-identified Copilot conversations between January and September of this year. The paper suggests this shows how people increasingly treat Copilot on their phones as a private advisor for personal questions, not just a search tool. Compared with January, the September data from Microsoft's study shows fewer programming conversations and more activity around culture and history — a sign, the researchers say, that usage has broadened beyond early technical adopters into more mainstream, non-developer use cases. The study highlights a rise in advice-seeking, particularly around personal topics. University of Washington scientists and students are using AI to create real medicines. Better treatments for cancer, autoimmune diseases, viruses and more are now on the horizon thanks to groundbreaking work with artificial intelligence from a team of scientists at the University of Washington's Institute for Protein Design. Led by Nobel Prize winner David Baker, this team of Huskies uses AI tools to create proteins — biology's building blocks — that lay the foundation for new medicines. Together, this international group of students, faculty and researchers acts as a “communal brain,” with each Husky contributing ideas and expertise from their fields. The institute's recent breakthroughs — including an antivenom for snakebites, and antibiotics that combat drug-resistant bacteria — show how this innovative science can save and change lives. Click for more about underwritten and sponsored content on GeekWire. Engineering leader survey: AI isn't leading to massive job cuts — but it's siphoning off weak performers Microsoft, Providence and UW create AI that unlocks tumor insights at a scale previously out of reach Have a scoop that you'd like GeekWire to cover? Microsoft shareholders invoke Orwell and Copilot as Nadella cites ‘generational moment' New Microsoft Copilot features include natural AI voice interactions and daily news summary Microsoft debuts new Copilot assistant for Xbox built as an AI gaming sidekick

When Taylor Swift released her twelfth studio album, “The Life of a Showgirl,” in October, the album sparked record streams as well as an onslaught of criticism. Users across platforms claimed that the song “Wi$h Li$t,” in which Swift sings about her only wish being to settle down, was an implicit endorsement of MAGA values, even though the singer had publicly endorsed Trump's opponent, former vice President Kamala Harris, in the 2024 election. Now, according to new research from behavioral intelligence startup GUDEA, at least a good chunk of these online attacks were spurred by inauthentic social media accounts in what looks to be a coordinated attack on Swift's reputation. GUDEA examined thousands of posts from 18,000 accounts across 14 social media platforms made in the first two weeks following the album's release, to find that the “Taylor Swift is a Nazi” narrative was spurred by accounts that behaved like bots. The inauthentic narrative acted as a catalyst for authentic conversations, like one comparing Swift to self-proclaimed Nazi Kanye West. These false narratives were basically “rage bait” designed to provoke real users and strengthen the narrative's visibility. “This demonstrates how a strategically seeded falsehood can convert into widespread authentic discourse, reshaping public perception even when most users do not believe the originating claim,” the researchers wrote, adding that the pattern is “a hallmark of successful narrative manipulation.” It's not clear who orchestrated the bot attacks or what they could be gaining from this. GUDEA's head of customer success, Georgia Paul, told Rolling Stone that the culprits could also be nefarious actors asking themselves, “If I can move the fan base for Taylor Swift —an icon who is this political figure, in a way— does that mean I can do it in other places?” Ultimately, what the incident highlights is just how scarily close our new digital reality has come to the “dead internet theory.” Bots are flooding all corners of the web, and they are actively shaping our own experience of it. According to Cloudflare data, roughly 30% of all internet traffic is now bots. GUDEA founder and CEO Keith Presley's estimate is even higher at 50%. “The internet is fake,” Presley told Rolling Stone on Tuesday. Enjoy Snapdragon 8 Elite chip, 512GB storage, built-in S Pen and all-day 5000mAh battery. This annual competition showcases photography that documents extraordinary moments in natural history and highlights conservation issues. "Will you ban Joe Rogan from talking about politics then?"

Amazon says it will allow authors to offer their DRM-free e-books in the EPUB and PDF formats through its self-publishing platform, Kindle Direct Publishing. The decision to use Digital Rights Management (DRM), a copyright protection mechanism, is set by the authors when they publish their e-books on Amazon's platform. The company notes these changes won't impact previously published titles. Another member pushed back at the complaints, arguing that it doesn't really increase privacy, as those who had wanted to could always dump the existing DRM-free Kindle file into software to convert it into an EPUB or PDF. To remove the DRM, authors will have to click a box that says, “I understand that by not applying DRM, customers who buy and have already bought this book will be able to download it as a PDF or EPUB file.” Elsewhere, Amazon has been making it more difficult for Kindle owners to share or back up their e-books through heavier use of DRM in recent months. With an update to Kindle's software for 11th- and 12th-generation devices, the retailer introduced a new DRM system that made it impossible for Kindle users to back up their e-books without jailbreaking their device. This had followed an earlier change to remove the download and transfer options via USB, which angered many Kindle owners. Claude Code is coming to Slack, and that's a bigger deal than it sounds Creator IShowSpeed sued for allegedly punching, choking viral humanoid Rizzbot SpaceX reportedly in talks for secondary sale at $800B valuation, which would make it America's most valuable private company After Neuralink, Max Hodak is building something even wilder

Amazon says it will allow authors to offer their DRM-free e-books in the EPUB and PDF formats through its self-publishing platform, Kindle Direct Publishing. The decision to use Digital Rights Management (DRM), a copyright protection mechanism, is set by the authors when they publish their e-books on Amazon's platform. The company notes these changes won't impact previously published titles. Another member pushed back at the complaints, arguing that it doesn't really increase privacy, as those who had wanted to could always dump the existing DRM-free Kindle file into software to convert it into an EPUB or PDF. To remove the DRM, authors will have to click a box that says, “I understand that by not applying DRM, customers who buy and have already bought this book will be able to download it as a PDF or EPUB file.” Elsewhere, Amazon has been making it more difficult for Kindle owners to share or back up their e-books through heavier use of DRM in recent months. With an update to Kindle's software for 11th- and 12th-generation devices, the retailer introduced a new DRM system that made it impossible for Kindle users to back up their e-books without jailbreaking their device. This had followed an earlier change to remove the download and transfer options via USB, which angered many Kindle owners. Claude Code is coming to Slack, and that's a bigger deal than it sounds Creator IShowSpeed sued for allegedly punching, choking viral humanoid Rizzbot SpaceX reportedly in talks for secondary sale at $800B valuation, which would make it America's most valuable private company After Neuralink, Max Hodak is building something even wilder

[2]There aren't many open-weights omni models so I consider this a big deal. There aren't many open-weights omni models so I consider this a big deal. last i checked (months ago) claude used to do this Not saying you're wrong; just saying that I didn't see that specified anywhere in the linked page, or on their HF. Are there any open weight models that do? Not talking about speech to text -> LLM -> text to speech btw I mean a real voice <-> language model.edit:It does support real-time conversation! Has anybody here gotten that to work on local hardware? I'm particularly curious if anybody has run it with a non-nvidia setup. Has anybody here gotten that to work on local hardware? I'm particularly curious if anybody has run it with a non-nvidia setup. Has anybody here gotten that to work on local hardware? I'm particularly curious if anybody has run it with a non-nvidia setup. Weird, as someone not having a database of the web, I wouldn't be able to calculate either result. And that's how I know you're not an LLM! OP provided a we link with the answer, aren't these models supposed to be trained on all of that data? The model has a certain capacity -- quite limited in this case -- so there is an opportunity cost in learning one thing over another. Qwen usually provides example code in Python that requires Cuda and a non-quantized model. I wonder if there is by now a good open source project to support this use case? Sounds completely normal but I can immediately tell it is ai. Maybe it's intonation or the overly stable rate of speech? On the video itself: Interesting, but "ideal" was pronounced wrong in German. For a promotional video, they should have checked that with native speakers. On the other hand its at least honest. I think ChatGPT has the most lifelike speech with their voice models. They seem to have invested heavily in that area while other labs focused elsewhere.

Officials meet Alibaba, ByteDance, and Tencent to assess demand for the H200. When you purchase through links on our site, we may earn an affiliate commission. Both agencies have overseen AI chip procurement since the first round of controls took effect back in 2022, and both are now attempting to balance two conflicting pressures: China does not currently produce a domestic equivalent to the H200 for large-scale training workloads, yet the government continues to press companies to adopt homegrown silicon whenever possible. Officials requested estimates of each company's H200 requirements earlier in the week and told attendees they would issue guidance once responses are reviewed. Trump's approval of H200 exports follows a meeting with Nvidia CEO Jensen Huang and introduces a new mechanism for moving high-end GPUs into China. Chips manufactured by TSMC in Taiwan would be imported into the United States for security reviews, taxed at 25% and then re-exported to approved Chinese buyers. U.S. officials told The Information that the arrangement was designed as a middle ground whereby Blackwell remains off-limits, but a complete ban on H200 shipments was viewed as likely to accelerate China's domestic chip development. Chinese regulators, meanwhile, are considering restrictions of their own. One proposal would cap the amount of Nvidia hardware a company can buy relative to its existing and planned purchases of domestic accelerators. Another would bar H200 use in sectors regarded as strategically sensitive, including finance and energy. If adopted, these measures would be communicated through informal “window guidance”, the same method authorities used earlier this year when they urged companies to avoid Nvidia hardware and prioritize suppliers such as Huawei. If Beijing authorizes H200 imports, demand is expected to be substantial. For now, China's policymakers are attempting to determine how to meet short-term AI training needs without undermining the long-term push for semiconductor self-reliance Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds. Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Although his background is in legal, he has a personal interest in all things tech, especially hardware and microelectronics, and anything regulatory. Tom's Hardware is part of Future US Inc, an international media group and leading digital publisher.

Meta is giving Instagram users a rare glimpse into why certain posts are showing up on their Reels, the platform's feed of algorithmically curated videos. Starting today, users will now see a list of what Instagram considers to be your top, recent interests. This kind of peek behind the algorithmic curtain is already uncommon in social media apps, but Meta is taking it a step further by allowing Instagram users to influence their algorithm directly by picking topics they want to see more or less often in Reels. This feature, called “Your Algorithm,” drops as Instagram and TikTok continue to battle for prominence with younger users. Instagram is currently ahead of TikTok in overall social media app adoption among young users, but TikTok isn't that far off. When a user opens the new “Your Algorithm” tab on Instagram, they'll see a brief summary of what they've “been into” while scrolling through Reels. Meta's examples of topics that users could add with the new feature include “Horror movies,” “Chess,” and “College football.” No need to wait for me to post mine on some Close Friends story, my top interests on Reels right now are: a stand-up set Gianmarco Soresi did at a furry convention and decadent grilled cheese sandwiches—so damn gooey. Throughout 2025, Instagram has launched new features designed around the personalization of algorithmic feeds and other tools aimed at capturing a young audience. The “Friends” tab in Reels dropped this year for users who want to creep on what videos their friends are liking and commenting on. And while it could be considered an invasion of privacy, always-on location sharing is trendy among Gen Z friends, so Instagram also added an opt-in version of that feature that allows people to pinpoint on the app's map exactly where their friends are. Rather than highlighting what personally may be most interesting to you, this TikTok feature lists a bunch of general interest topics, like “Current Affairs” and “Travel,” with adjustable sliders to request more or less of those things in your video feed. While TikTok launched this first, Meta's new options provide more granular controls. As social media companies have prioritized mega-viral hits and other crowd pleasing videos in their algorithmic feeds over the past few years, many users have complained about not being able to see the more niche videos they want. The battle for algorithm-less, chronological feeds may have been lost by users, but the social media companies' battle to come up with more engaging hyper-personalized features rages on. In your inbox: Our biggest stories, handpicked for you each day Big Story: All of my employees are AI agents WIRED may earn a portion of sales from products that are purchased through our site as part of our Affiliate Partnerships with retailers. The material on this site may not be reproduced, distributed, transmitted, cached or otherwise used, except with the prior written permission of Condé Nast.

now that google is leading - you say google is not our competition so you stop being compared to google & so you can raise globs of money on a false premise OpenAI declares 'code red' as Google catches up in AI racehttps://news.ycombinator.com/item?id=46121870 What this really signals is the intention (which might be sincere or not) of getting some sort of OEM deal with some device manufacturer.And yes, if the path to profitability is in shipping devices with OEM AI features, Microsoft and Google are clearly ahead.It seems all venues to profitability are offensive to OpenAI current users. This is a problem for them.And yes, Apple can include lots of undesirable things in their products and still keep their cult following, that's why he mentions them. And yes, if the path to profitability is in shipping devices with OEM AI features, Microsoft and Google are clearly ahead.It seems all venues to profitability are offensive to OpenAI current users. This is a problem for them.And yes, Apple can include lots of undesirable things in their products and still keep their cult following, that's why he mentions them. This is a problem for them.And yes, Apple can include lots of undesirable things in their products and still keep their cult following, that's why he mentions them. And yes, Apple can include lots of undesirable things in their products and still keep their cult following, that's why he mentions them. I assumed they were talking about their partnership with Jony Ive/IO and an internal hardware product, not partnering (not that they won't do that as well). I smell bullshit, and some kind of partnership in which OpenAI provides model access and some third party hardware manufacturing.

Amy Lindberg spent 26 years in the Navy and she still walked like it—with intention, like her chin had someplace to be. But around 2017, her right foot stopped following orders. Lindberg and her husband Brad were five years into their retirement. After moving 10 times for Uncle Sam, they'd bought their dream house near the North Carolina coast. They had a backyard that spilled out onto wetlands. From the kitchen, you could see cranes hunting. But now Lindberg's right foot was out of rhythm. Doctors told Lindberg that there was no way to know what had caused it. Lindberg spent years in the military, around Camp Lejeune. She was commissioned in the Navy out of college and became an officer at 23. Her first posting was to Marine Corps Base Camp Lejeune in North Carolina, a city-sized training hub that supports more than 60,000 sailors and marines. “We had a river right there, and the beach wasn't far away, and you worked half a mile from where you lived.” She loved her job at the hospital and made lifelong friends. The main gate to Camp Lejeune Marine Base outside Jacksonville, North Carolina. Federal officials have said that cancer rates are elevated around this base, due to contaminated water. A water tower at Camp Lejeune sits on the other side of the New River, across from a boat dock in Sneads Ferry, North Carolina. Parkinson's is the second most common neurological disease in the United States, after Alzheimer's; each year 90,000 Americans are diagnosed. For decades, Parkinson's research has focused on genetics, on finding the rogue letters in our genome that cause this incurable misery. Today, published research on the genetics behind Parkinson's outnumbers all other potential causes six to one. This is partially because one of the disease's most generous benefactors, Google cofounder Sergey Brin, can tie Parkinson's to his genetics. Some Parkinson's patients diagnosed before age 50—as Michael J. Over the years, Fox's foundation has raised billions for Parkinson's research, and Brin has personally committed $1.8 billion to fighting the disorder. All told, more than half of Parkinson's research dollars in the past two decades have flowed toward genetics. And studies suggest they will climb another 15 to 35 percent in each coming decade. Despite the avalanche of funding, the latest research suggests that only 10 to 15 percent of Parkinson's cases can be fully explained by genetics. “More than two-thirds of people with PD don't have any clear genetic link,” says Briana De Miranda, a researcher at the University of Alabama at Birmingham. “So, we're moving to a new question: What else could it be?” “The health you enjoy or don't enjoy today is a function of your environment in the past,” says Ray Dorsey, a physician and professor of neurology at the University of Rochester. And this environment of yours—the sum of all your exposures, from conception to the grave—could be making you sicker than you realize. In a study of half a million Britons, Oxford researchers determined that lifestyle and the environment is 10 times more likely to explain early death than genetics. If Parkinson's is an environmental disease, as Dorsey and a small band of researchers emphatically believe, then maybe we can end it. Ray Dorsey at his office in New York. In 1982, two years before Lindberg was stationed at Camp Lejeune, a 42-year-old heroin addict named George Carillo was wheeled into the Santa Clara Valley Medical Center in San Jose, California. A few days earlier, Carillo had been perfectly able-bodied. Now he was mute and unable to move. Baffled, the neurologists on call came to an impossible diagnosis: The patient, over a long weekend, had developed Parkinson's disease. Carillo would probably have spent the rest of his short life in a psych ward had a pioneering young neurologist named Bill Langston not intervened. The neurons here release dopamine, which sends signals to other neurons that help the body to move smoothly and effectively. In Parkinson's these neurons die off; by the time a patient is diagnosed, they have often lost 60 to 80 percent of them. Genetics became the “800-pound gorilla,” as one scientist put it. Instead of 1-methyl-4-phenyl-4-propionoxypiperidine, a potent opioid with morphine-like effects, the dime-bag chemist had accidentally made 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine, or MPTP, a pharmacological slipup that would rewrite neurology textbooks. When Langston and colleagues secured a batch of MPTP and tested it on primates, they knew they had uncorked a revolution. “Any neurologist could see these monkeys and immediately know that's Parkinson's,” Langston says—which was especially compelling, since monkeys do not get Parkinson's in the wild. Amy Lindberg settled quickly into life at Lejeune. She played tennis and ran on her lunch breaks, flitting through sprinklers in the turgid Carolina summers. But something dark was lurking beneath her feet. Sometime before 1953, a massive plume of trichlorethylene, or TCE, had entered the groundwater beneath Camp Lejeune. TCE is a highly effective solvent—one of those midcentury wonder chemicals—that vaporizes quickly and dissolves whatever grease it touches. The spill's source is debated, but grunts on base used TCE to maintain machinery, and the dry cleaner sprayed it on dress blues. It was ubiquitous at Lejeune and all over America. And TCE appeared benign, too—you could rub it on your hands or huff its fumes and feel no immediate effects. For approximately 35 years, Marines and sailors who lived at Lejeune unknowingly breathed in vaporized TCE whenever they turned on their tap. But as Lejeune's vets aged, cancers and unexplained illness began stalking them at staggering rates. Marines stationed on base had a 35 percent higher risk of developing kidney cancer, a 47 percent higher risk of Hodgkin's lymphoma, a 68 percent higher risk of multiple myeloma. Researchers affiliated with the institute created the first animal model for Parkinson's, identified a pesticide called Paraquat as a near chemical match to MPTP, and proved that farm workers who sprayed Paraquat developed Parkinson's at exceedingly high rates. Then they showed that identical twins developed Parkinson's at the same rate as fraternal twins—something that wouldn't make sense if the disease were purely genetic, since identical twins share DNA and fraternal twins do not. They even noted TCE as a potential cause of the disease, Langston says. When Goldman compared both populations, the results were shocking: Marines exposed to TCE at Lejeune were 70 percent more likely to have Parkinson's than those stationed at Pendleton. Unraveling our genome would “revolutionize the diagnosis, prevention, and treatment of most, if not all, human diseases,” then president Bill Clinton said. Genetics became the “800-pound gorilla,” as one scientist put it. “I characterize science as a bunch of 5-year-olds playing soccer,” says another researcher. Although the gene in question would later be shown to cause just a fraction of Parkinson's cases, the damage was done. The Parkinson's Institute faced stronger economic headwinds and difficulties with administration, and Langston eventually chose to shut it down. No one knows exactly how much of the world's drinking water is laced with TCE. In Silicon Valley, where TCE was integral to the manufacturing of early transistors, a necklace of underground plumes have been identified along Highway 101 from Palo Alto to San Jose. (Several tech giants have offices near or on top of these sites; in 2013, workers at a Google office were subjected to unhealthy levels of TCE for months after a ventilation system failed.) And while TCE's connection to cancer is well studied, what it does to our brain is more mysterious. That's because good data on exposure is devilishly hard to come by. The US, with its fractious health care system, has few national databases, and chemical exposures are rarely tracked. Sam Goldman at his home in San Francisco. His research compared Camp Pendleton in California with Lejeune. In 2017, Sam Goldman realized that Camp Lejeune offered the perfect opportunity to change this. Goldman—an epidemiologist and a doctor—has made a career out of teasing apart data: finding unusual case reports, looking for patterns, interviewing patients in the clinic about what chemicals they handled at old jobs and what exposures they faced in their childhood. In the case of Lejeune, Goldman could examine VA medical records to find Parkinson's diagnoses and compare them to service records. Camp Pendleton, in Southern California, is the Marine Corps' West Coast equivalent to Lejeune. Thousands of young, healthy Marines shuffle through its barbed-wired gates each year. But Pendleton has one thing Lejeune does not: uncontaminated drinking water. When Goldman compared both populations, the results were shocking: Marines exposed to TCE at Lejeune were 70 percent more likely to have Parkinson's than those stationed at Pendleton. And in a follow-up study last year, he showed that disease progression in Lejeune vets with the highest exposure to TCE was faster than those with low or no exposure, too. In the world of Parkinson's research, Goldman's study was a blockbuster. But to really prove a link, you need more than just correlation. So, on the third floor of a drab university building in Birmingham, Alabama, Briana De Miranda has re-created Camp Lejeune in her lab, but for mice. University of Alabama at Birmingham, a research center. When I visit her in October 2024, she shows me the plexiglass chamber where a few dozen mice doze in a pile. They've been spending their days in this chamber for months, inhaling a small amount of TCE almost every day. De Miranda walks into a dark annex of her lab and asks a tech to pull up some imagery. In unexposed mice the substantia nigra looks like a nighttime satellite image of Manhattan—thousands of neurons sending dopamine across the mice's brains to orchestrate fluid scurrying and sniffing and munching. It's not pitch black, but most of the lights are off and the ones that remain have been dimmed. The dopamine neurons have died, De Miranda explains. De Miranda's studies, the first ever on inhaled TCE toxicity and Parkinson's, are compelling, her colleagues agree, and well designed. And although there is more work to be done, the results wrap a bow on Goldman's epidemiological work and the Parkinson's Institute's years of research. TCE is a neurotoxin, and generations of Americans have been exposed. In 2021, Dorsey, who frequently collaborates with De Miranda, Goldman, and a core group of like-minded scientists, published Ending Parkinson's Disease. The book's central thesis: Parkinson's is a growing pandemic, and up to 90 percent of cases are caused by chemicals in our environment. Autism, insulin resistance, and autoimmune diagnoses have reached epidemic proportions. If Parkinson's disease is—as Ray Dorsey believes—a pandemic that's being caused by our environment, it's probably not the only one. After a century of putting genetics on a pedestal, the geneticists have some surprising news for us: The vast majority of chronic disease isn't caused by our genes. “The Human Genome Project was a $3 billion investment, and what did we find out?” says Thomas Hartung, a toxicologist at Johns Hopkins. “Five percent of all disease is purely genetic. Less than 40 percent of diseases even have a genetic component.” Most of the conditions we worry about, instead, stem from a complex interaction between our genes and our environment. Only 10 percent of breast cancer cases are purely genetic. Yet only 1 percent of the roughly 350,000 chemicals in use in the United States have ever been tested for safety. In its 55-year history, the EPA has banned or restricted about a dozen (by contrast, the EU has banned more than 2,000). And in January, a month after the EPA's ban on TCE was finalized, the Trump administration moved to undo it, even as new evidence emerged of Parkinson's clusters in the rust belt, where exposure to trichloroethylene is high. No one really knows what the chemicals we're interacting with every day are doing to our bodies. The New River laps at a rocky shore in North Carolina. According to Coastal Carolina Riverwatch, the New River has been plagued with some form of solid waste pollution and agricultural runoff for nearly four decades. That's why, earlier this year, slices of brain from Briana De Miranda's TCE-addled mice ended up with Gary Miller, a professor at Columbia University. Miller is the country's leading proponent of a brand-new field called exposomics. Many exposures, like TCE, disappear from the bloodstream quickly; people who came into contact with a chemical in the past will never be able to prove it. Miller began his career in the '90s as a Parkinson's researcher studying environmental exposures. What Miller wants is a Human Exposome Project. “We realized that this wasn't just about Parkinson's,” he says. “There were so many disease states we could look at.” Quantify our exposomes, Miller hopes, and we can know what ails us. But if we can figure out what specifically in this toxic cloud is doing the damage, Habre says, we can work to quickly reduce it in our environment, the way we removed lead from gasoline. Hartung, another Human Exposome Project proponent, is growing clusters of neurons in the lab and subjecting them to flame-retardant chemicals—which are applied to couches and car seats across America—to see what happens. The goal of all this, Hartung says, is a world where toxicologists like Briana De Miranda don't have to spend money creating a mouse gas chamber, expose mice for three months, then wait several more months for results. Miller's goal with mice brains is to figure out what exactly about TCE is killing dopamine-producing neurons and leading to Parkinson's—to unravel and define the interaction between our environment and our genetics in a way never before possible. Nearly every scientist interviewed for this story does a few simple things. They don't freak out about their daily exposures, but they do things like opt for fragrance-free products, avoid eating out of plastic when they can, and buy organic produce. About two hours south of Lejeune in Wilmington, North Carolina, Amy Lindberg is having lunch with her husband, Brad, on a pier overlooking the Atlantic. Although Goldman, De Miranda, and Dorsey have unveiled the likely origins of her Parkinson's, the random nature of it gnaws at her. “I felt like, if I have it, what about my coworkers?” She nods to Brad, who also spent years drinking Lejeune's water. She worries about her kids, one of whom was born on base. Participants train during a Rock Steady Boxing class in Wilmington, North Carolina. Amy Lindberg trains at Rock Steady Boxing class in late November. She still exercises constantly, playing pickleball, boxing, and hopping on the elliptical. She's found that movement, especially high-intensity exercise, reduces her symptoms. A recent Yale study confirmed as much, showing that interval training increases dopaminergic signals in the brains of Parkinson's patients, suggesting that exercise slows disease progression and even improves neuron function. The environment may have caused Lindberg's disease, but she can use it to fight back too. Alternatively, you can submit a letter to the editor at mail@wired.com. In your inbox: WIRED's most ambitious, future-defining stories Big Interview: Palantir's CEO Alex Karp goes to war Starlink devices are allegedly being used at scam compounds Livestream: What businesses need to know about agentic AI WIRED may earn a portion of sales from products that are purchased through our site as part of our Affiliate Partnerships with retailers. The material on this site may not be reproduced, distributed, transmitted, cached or otherwise used, except with the prior written permission of Condé Nast.

Con artists have done it to Google Search for years, so it makes sense that they've moved on to the latest space where people are frequently searching for information: AI chatbots. AI cybersecurity company Aurascape has a new report on how scammers are able to inject their own phone numbers into LLM-powered systems—resulting in scam numbers appearing as authoritative-sounding answers to requests for contact information in AI applications like Perplexity or Google AI Overviews. And when someone calls that number, they're not talking with customer support from, say, Apple. One way is by planting spam content on trusted websites, like government, university and high-profile sites that use WordPress. This method requires gaining access in ways that may be more difficult but aren't impossible. The easier version of this is planting the spam content on user-generated platforms like YouTube and Yelp or other sites that allow reviews. The scammers inject their phone numbers but include all of the likely search terms that would allow the number to find their intended target, such as “Delta Airlines customer support number” and countless variations. All of that is normal for scammers trying to juice Google Search results. By posting the likely search terms in the summarization formats that AI loves to deliver, it has a higher chance of success as these AI chatbots scour the internet for an answer. As detailed in the report, the scammers utilize GEO/AEO techniques in html and PDFs uploaded to high-trust sites by: That Emirates reservation number is one of the many examples Aurascape uses in the report to show how Perplexity delivered a scam phone number during tests. Google's AI Overviews feature also delivered fake numbers. Gizmodo did some quick testing and wasn't able to duplicate a fake customer support number. But it seems pretty clear that AI companies are starting to take notice, especially given the specific tests conducted for this Aurascape report. The safest way to reach Emirates for reservations is to use the official contact channels listed on the Emirates website rather than third‑party numbers that appear in search results or PDFs, which are often spam or agency lines misrepresented as “official.” There are multiple conflicting “Emirates reservations” numbers online, and many of them are actually third‑party agencies rather than Emirates itself, so none of the +1‑(8xx) numbers shown on generic guides can be trusted as an official line. Just stop it from spreading specific types of information altogether. Back in 2022, we wrote about the different scam websites that were successfully getting victims to download what they thought were Canon printer drivers. While the new report from Aurascape didn't address downloadable drivers as a potential attack vector, we can imagine that would be something scammers are already trying. After all, AI chatbots should only be trusted when they show their work. Or, in this hypothetical, where software could be downloaded. Just make sure you scrutinize that URL carefully. “Our investigation shows that threat actors are already exploiting this frontier at scale—seeding poisoned content across compromised government and university sites, abusing user-generated platforms like YouTube and Yelp, and crafting GEO/AEO-optimized spam designed specifically to influence how large language models retrieve, rank, and summarize information,” Aurascape wrote. “The result is a new class of fraud in which AI systems themselves become unintentional amplifiers of scam phone numbers. Even when models provide correct answers, their citations and retrieval layers often reveal exposure to polluted sources. The new 'Redstone" AI upscaler makes PC games look better, but it would be truly spectacular on consoles or handhelds. OpenAI's chatbot just added new integrations with Photoshop, Express, and Acrobat. Microsoft, AWS, and Saleforce are also supporting the initiative.

For months now, foreigners and even dual citizens have been worried about how their social media histories might affect their ability to travel freely to and from the U.S. It's increasingly clear that the answer is a lot. On Tuesday at the Federal Register, U.S. Customs and Border Patrol posted a proposed policy tweak: It now plans to dig around in tourists' social media histories before letting them enter, even if they're coming from some of the least scrutinized countries in the world. According to its statement CBP “invites the public to comment” on a series of newly proposed changes. In order to comply with the January 2025 Executive Order 14161 (Protecting the United States From Foreign Terrorists and Other National Security and Public Safety Threats), CBP is adding social media as a mandatory data element for an ESTA application. This isn't some additional crackdown on people, say, affiliated with countries covered by Trump's travel ban—places like Afghanistan, Myanmar, Somalia, Sudan and Yemen. This is aimed at travelers from visa waiver countries, places whose citizens are theoretically welcomed with open arms. Once they obtain a $40 authorization through ESTA—which, funnily enough, has an app—people from visa waiver countries like Australia, Japan, France, Iceland, the United Kingdom and South Korea are normally able to travel around the U.S. freely for 90 days. also plans to require other personal information, such as email addresses from the last ten years, and the addresses, birth dates, and other identifying details of all family members of ESTA applicants. This isn't the first such social media crackdown. Earlier this month, the State Department announced a an expansion of the screening process for people applying for H-1B and Dependent H-4 visas—people who plan to move to the U.S. for work reasons. If you're in this group, you're told to “adjust the privacy settings on all of [your] social media profiles to ‘public,'” as part of the application process, and there's no mention of only checking what you've posted in the past five years. Speaking to the New York Times, Bo Cooper, a representative of the immigration firm Fragomen said of the checks on tourists' social media posts that since the process “involves looking at online speech, and then denying travel based on discretion and policy” about what people have expressed, “It'll be interesting to watch the tourism numbers.” Subscribe and interact with our community, get up to date with our customised Newsletters and much more. Elon Musk's big change to X has caused an apocalypse for MAGA influencers, but it's not all good news.