iQOO launches its flagship performance phone, the 15 Ultra
xAI releases Grok Imagine 1.0
Alibaba Cloud Qwen team releases Qwen3-Coder-Next
Apple reminds users to upgrade the Home app
Microsoft to discontinue standalone OneDrive business subscriptions
Adobe cancels plans to discontinue Animate
iQOO launches its flagship performance phone, the 15 Ultra
On February 4, the iQOO brand officially unveiled its performance-focused Ultra flagship, the iQOO 15 Ultra. Aimed at gamers and creators, the device is enhanced across four key areas—performance, controls, live streaming, and design—and comes in two color options: 2077 (Black) and 2049 (Silver).
The iQOO 15 Ultra is powered by the fifth-generation Snapdragon 8 Elite SoC, paired with LPDDR5X Ultra Pro memory and UFS 4.1 storage, offering configurations up to 24GB + 1TB. For cooling, it features a custom Ice Dome air-cooling system that combines airflow channels with an extra-large VC, along with the in-house esports chip Q3, enabling full-scene ray tracing, 2K native-resolution upscaling, and 120FPS frame interpolation. The display is a 2K Samsung “Zhufeng” panel supporting a dynamic refresh rate from 1 to 144Hz, with peak brightness reaching 8000 nits and Dolby Vision certification.
In terms of controls, the iQOO 15 Ultra adopts a system-level ultra-sensory control framework that integrates pressure-sensitive shoulder buttons, a gyroscope, and touch input. It also debuts the Warhammer MAX dual-axis haptic motors, paired with dual speakers and Dolby Atmos. The phone packs a 7400mAh Blue Ocean battery, supports IP68 and IP69 dust and water resistance, and includes Wi-Fi 7, an ultrasonic 3D fingerprint sensor, and support for 31 5G bands, including n79.
For imaging, the device features a rear triple-camera setup with three 50MP sensors and a 32MP front-facing camera. Both the main camera and the periscope telephoto support CIPA 4.5-level stabilization, and the system is powered by the NICE 3.0 imaging algorithm and the Magic 2.0 image restoration engine. Pricing starts at RMB 5,499 for the 16GB + 256GB version during the initial sales period (February 4 to February 12), with the 16GB + 512GB variant priced at RMB 5,999. Higher-end options include 16GB + 1TB at RMB 6,999 and 24GB + 1TB at RMB 7,699. Source
xAI releases Grok Imagine 1.0
On February 2, the xAI team announced the release of its video generation tool Grok Imagine 1.0, describing it as the company’s “largest upgrade to date.” According to xAI, the new version has begun rolling out more broadly, and users can try it via the Imagine page on grok.com.
Grok Imagine 1.0 is xAI’s audio-and-video generative foundation model. It currently supports generating short videos of up to 10 seconds in length at a maximum resolution of 720p, and it “significantly improves” audio capabilities, with an emphasis on producing richer speech and soundtrack effects. Source
Alibaba Cloud Qwen team releases Qwen3-Coder-Next
On February 4, Alibaba Qwen announced the release of Qwen3-Coder-Next, an open-weight language model purpose-built for coding agents and local development. The model is built on Qwen3-Next-80B-A3B-Base and adopts a new architecture that combines hybrid attention with Mixture-of-Experts (MoE). Through large-scale executable task synthesis, environment interaction, and reinforcement learning for agent training, the model delivers strong programming and agent capabilities while significantly reducing inference costs.
Recently, Apple sent a notice to Apple Home users, reminding them to complete an upgrade of the Home app by February 10, 2026. Users who do not complete the upgrade may encounter issues with smart accessories and automation scenes failing to run properly, and may even be unable to access their smart home devices within the app.
In addition, users who do not upgrade will be unable to use new features such as robot vacuum support, and will miss out on important security patches and performance optimizations.
To upgrade, open the Home app, tap the three-dot icon in the top-right corner to enter Home Settings, then select Software Update → Update Now, and follow the on-screen instructions. Source
Microsoft to discontinue standalone OneDrive business subscriptions
On February 3, Microsoft confirmed in updated support documentation that it plans to remove the standalone low-priced subscription offerings for OneDrive for Business and SharePoint Online (Plan 1 and Plan 2). These plans do not include core productivity apps such as Word, Excel, and PowerPoint, but have been attractive to users with limited budgets who only need specific features.
Microsoft plans to stop selling these standalone products to new customers in May 2026, and will disable renewals for existing users in June. The official end-of-life (EOL) date for current subscriptions is January 2027, with Microsoft planning to fully end all technical support for these products in December 2029, at which point the related services will be completely taken offline. Source
Adobe Cancels Plans to Discontinue Animate
Following opposition from multiple parties, Adobe stated in a FAQ published on February 4 that Animate, which was previously slated for full discontinuation, will instead enter maintenance mode. Animate will continue to receive ongoing security updates and vulnerability fixes, and both existing and new users will be able to keep using the software. However, no new features will be introduced going forward. Source
Mozilla announced on February 3 that it will offer Firefox users an option to disable all AI features. Starting with Firefox version 148, scheduled for release on February 24, users will be able to enable a “Block AI enhancements” option in settings. Once enabled, this will block AI-related pop-ups and prompts in the current version as well as in future updates. In addition, users can use the new AI control switches to manage individual AI features separately. Source
OpenAI launches the Codex desktop app
OpenAI announced on February 3 the release of a desktop Codex app for macOS, designed to help developers manage and coordinate multiple AI agents. The app supports parallel task execution, long-running projects, and intelligent automated workflows. Positioned as a developer’s “central command hub,” it connects workflows and tools through Skills, allows users to define scheduled automated tasks, and adapts to individual developers’ usage habits.
Codex is included with existing Plus, Pro, Business, Enterprise, and Edu subscriptions, while Free and Go users are subject to certain rate limits. OpenAI also plans to release a Windows version of the Codex app in the future. Source
Xbox PC client update integrates NetEase UU Accelerator
Microsoft announced on February 3 that the Xbox PC client will now directly integrate the NetEase UU Accelerator. According to the official Xbox statement, this update is “based on a deep understanding of Chinese players’ gaming habits.” After launching the Xbox PC client, users can find the UU Accelerator under “My Apps.” It should be noted that using the NetEase UU Accelerator may require a UU membership. The service is provided by NetEase UU Accelerator and must be purchased separately. Microsoft does not sell, manage, or operate UU membership services, and bears no responsibility for any transactions, content, or services related to UU membership. Source
Nintendo Switch family sales surpass 155 million units
According to Nintendo’s latest financial report released on February 3, cumulative sales of the Switch family reached 155.37 million units as of December 31, 2025. This officially surpasses the 154.02 million units sold by the Nintendo DS, making the Switch the best-selling game console in Nintendo’s history. The second-generation console, Switch 2, delivered a strong performance during the holiday season, selling 7.01 million units in the quarter and 17.37 million units over the first three quarters of the fiscal year, and was described by Nintendo as “the fastest-selling dedicated gaming platform in the company’s history.”
Driven by strong hardware and software sales, Nintendo reported revenue of ¥803.32 billion (approximately USD 5.2 billion) for the third quarter of fiscal year 2026, up 86% year over year, while profit rose 20% to ¥159.93 billion (approximately USD 1.03 billion). Source
On February 2, Sony-owned anime streaming platform Crunchyroll announced a price increase for all membership tiers in the United States and select international markets. Each plan will see a USD 2 monthly increase. Following the adjustment, the Fan plan will cost USD 9.99 per month, the Mega Fan plan will rise to USD 13.99 (supporting streaming on up to four devices simultaneously and offline viewing), and the Ultimate Fan plan will increase to USD 17.99 (supporting up to six devices and including access to the manga library).
As compensation, the Fan plan now adds offline downloads on a single device. The platform has also recently rolled out several feature updates, including multiple profile switching, parental control locks, and “skip intro/ending” options. The new prices take effect immediately for new users, while existing subscribers will be charged at the new rates starting with their first billing date after March 4, 2026. Source
Rumors You Can Just Glance At
The Verge reports, citing court documents it obtained, that Google is developing a desktop operating system called Aluminium OS based on Android. The OS is expected to open testing to commercial partners in 2026, while the full official release aimed at enterprise and education users has been delayed until 2028. Affected by existing Chromebook hardware compatibility limitations and Google’s promised ten-year support period, the current ChromeOS will enter a gradual phase-out, with plans to be officially discontinued in 2034. Source
Detailed specifications and official renders of the Sony WF-1000XM6 were accidentally leaked on the website of Thai retailer Power Buy. According to the leaked information, the WF-1000XM6 features a fully matte finish, with significant changes to the earbud design, while the charging case adopts a more angular, faceted shape. Based on information from Dealabs and retailers, the earbuds are scheduled to open for pre-orders on February 12, 2026, with official shipments beginning on February 23. The expected U.S. price is USD 329.99. Source
Hypergryph responds to overseas payment issues in Arknights: Endfield
Malicious skills found in the OpenClaw skill library
ROG Kithara premium gaming headset announced and released
Rumors You Can Just Glance At
SpaceX acquires xAI
On February 2, SpaceX announced its acquisition of Elon Musk’s artificial intelligence company xAI. The deal makes SpaceX the most valuable private company in the world, with a valuation reaching USD 1.25 trillion. In the acquisition memo, Elon Musk stated that the primary goal of this acquisition is to build space-based data centers, aiming to avoid the power consumption and cooling challenges associated with running AI in large terrestrial data centers. Source
Adobe discontinues Animate
On February 2, Adobe announced that it will discontinue Adobe Animate effective March 1, 2026, and remove the application from its official website, after which users will no longer be able to purchase it. Enterprise customers will be able to continue using the app, download content, and receive technical support until March 1, 2029. Other users will be able to keep using the app, download content, and access technical support until March 1, 2027. Adobe Animate was previously known as Adobe Flash Professional / Macromedia Flash and has been used to create vector graphics and animations. In its announcement, Adobe stated that due to technological advancements and the emergence of new platforms and paradigms, better applications are now available to support users’ creative needs. The company recommends After Effects or Adobe Express as alternatives. Source
Hypergryph responds to overseas payment issues in Arknights: Endfield
On February 1, Hypergryph responded to Southern Metropolis Daily regarding a serious PayPal payment interface error that occurred on January 22, the launch day of Arknights: Endfield, affecting overseas servers. According to Hypergryph, when the payment issue occurred, most overseas regions were outside peak gaming hours. Upon receiving user feedback, the company promptly disabled the PayPal payment option, and the overall response was considered timely. In total, 3,429 orders were affected, with the total amount involved not exceeding USD 80,000. Fewer than 1,800 users were impacted, and the highest abnormal charge for a single user reached USD 5,289. As of that day, Hypergryph had completed full refunds for all affected users. Source
Malicious skills found in the OpenClaw skill library
According to OpenSourceMalware, between January 27 and February 1, two waves totaling more than 230 malicious skills were published to OpenClaw (formerly ClawdBot/MoltBot) skill repositories, including ClawHub and GitHub. These malicious skills primarily target cryptocurrency traders, disguising themselves as legitimate crypto trading automation tools while stealing and exfiltrating sensitive information on Windows and macOS. All of the malicious skills share the same underlying strategy and rely on sophisticated social engineering to persuade users to execute harmful commands. Once run, these commands steal trading API tokens, wallet private keys, SSH credentials, browser passwords, and other sensitive data. Source
ROG Kithara premium gaming headset announced and released
On February 2, the ROG Kithara premium gaming headset was officially released and made available for purchase. Developed in collaboration with audio manufacturer HIFIMAN, the headset features a 100 mm HIFIMAN planar magnetic driver tuned by ROG, along with an open-back design. It comes with a balanced, interchangeable headphone cable, making it compatible with DACs, amplifiers, game consoles, and PCs. The headset is also equipped with a full-bandwidth inline MEMS microphone. The launch price is CNY 1,999. Source
Rumors You Can Just Glance At
Tencent announced on February 2 that QQ Classic Farm will officially return on February 6, launching simultaneously on QQ and WeChat. Based on current official teasers, this new release will include a mobile version and largely retain the gameplay of QQ Farm, which first launched in 2009. Players take on the role of a farmer—planting crops, raising animals, managing agricultural products—and can visit friends’ farms to “steal crops” or “plant weeds” as mischievous interactions. Source
Baidu’s autonomous ride-hailing service Apollo Go (Luobo Kuaipao) officially announced it will launch in Qingdao at 9:00 a.m. on February 3, with operations limited to the Hongdao area of Chengyang District. Available vehicle options include Arcfox models and the Luobo 06, featuring a “business-class ultra-spacious cabin,” 135-degree intelligent ten-point massage heated seats, and support for VIP music streaming. Users can hail rides by logging into their Baidu account via the Apollo Go app or mini program, and will receive an SMS notification when the vehicle arrives. Source
According to The Wall Street Journal, Apple is considering shifting the production of some lower-end processors away from TSMC to another supplier. The report does not name the alternative manufacturer, though prior rumors have suggested that Intel may begin supplying chips for Apple in 2027 or 2028, potentially including processors for some Mac and iPad models. Source
According to DigiTimes, Apple’s rumored plan to enter the smart glasses market later this year has already begun to impact the AR optics supply chain. The report says multiple Taiwanese AR optics suppliers are investing to expand capacity and shift R&D focus. Kinko Optical has reportedly invested around USD 5.6 million to develop a new AR/VR/MR research center, while companies such as Asia Optical, MetaOptics, Chung Yang Optics, and Zhihui Innovation are also accelerating related product development. Source
Recently, media outlets discovered that a Google app called Desktop Camera was listed on the Google Play Store on January 28. The app description suggests a minimalist UI, closely resembling ChromeOS or the Pixel Camera interface. It is currently speculated that this app may be a system component of the previously rumored “desktop-style Android interface,” Aluminium OS. Source
@Vanilla: Over the past week, the locally self-deployed AI tool OpenClaw (originally named Clawdbot, later renamed Moltbot) has truly taken over timelines, even sparking a small wave of Mac mini purchases. Unlike traditional chatbots, OpenClaw’s biggest strengths lie in three areas: controllable privacy, freedom of model choice, and convenient invocation.
In this installment of PReview, I don’t want to dwell on dry parameters or technical principles. Instead, I’d rather talk about what OpenClaw actually is—and what it can do for your digital life.
OpenClaw is neither a simple large language model nor a plain coding CLI. It is, in essence, a local “digital life hub.” It integrates a modular architecture including Model, Agent, Skill, and MCP, and achieves cross-platform access and asynchronous scheduling through a Gateway that resides on a Mac mini or a server. No matter where you are, you can interact with it in real time through various instant messaging tools.
At the same time, all memories, file indexes, and personal habits generated within OpenClaw are stored in your own local workspace (such as Memory.md, index databases, and Skills scripts). According to the official claims, you retain full control over your data sovereignty, with no need to worry about privacy leaks.
If you happen to own a Mac mini, that’s naturally the best option—it effectively gives you a low-power personal server running 24/7, with seamless access to Apple-ecosystem-related Skills. If you don’t have a Mac mini, there’s no need to buy one specifically: cloud providers like Tencent Cloud, Alibaba Cloud, Cloudflare, and DigitalOcean all offer OpenClaw-specific server images, making one-click deployment extremely convenient.
Next, I’d like to share two real-world use cases from my recent experience with OpenClaw.
In the past, doing Vibe Coding usually meant sitting in front of desktop clients like Claude Code, Codex, Gemini CLI, or Antigravity. Now, I simply send instructions to a bot via a chat app on my phone, and OpenClaw automatically handles all the logic writing and debugging on the server side, returning the results in real time.
For example, I recently switched my primary iPhone input method to “Cang Input,” but wasn’t satisfied with its default skin. I took an existing hskin theme file and dropped it directly to the bot, asking it to precisely tweak the skin code based on my aesthetic preferences and typing habits. Once the adjustments were done, I imported the generated skin file directly on my phone—the entire process without touching a computer keyboard.
I’ve also created a personal knowledge base locally on my computer, complete with source files, indexes, and management scripts. Through OpenClaw, I can issue commands to this knowledge base remotely—whether it’s syncing newly added file indexes or invoking Python for deep full-text search, analysis, or even assisted writing, all of it can be done instantly from a mobile device. This way, I get the comprehension power of large models without uploading my private database to cloud services, truly achieving “data stays local, compute goes with you.”
For now, I’ll stop here with the introduction to OpenClaw. I plan to write a more in-depth article later to explore it in greater detail. In my view, OpenClaw’s greatest success isn’t some revolutionary technological breakthrough, but rather how precisely it captures real user needs—vastly expanding the boundaries of AI usage through its Gateway and Channels.
While there is still ongoing discussion around its local control permissions and privacy boundaries, it’s hard to deny that the world depicted in the film Her feels closer than ever. And OpenClaw might just be the first truly handy tool to help you build your own version of “Samantha.”
Celestia: A Universe Database You Can Freely Roam
Platform: iOS / iPadOS / macOS / Android / Windows
Keywords: Universe, Science Education
@大大大K: When it comes to “star map” tools, most of us probably think of apps like Stellarium. That said, Stellarium focuses more on observing the night sky from Earth’s perspective, and the sense of scale at a truly cosmic level can feel a bit limited. If what you want instead is the experience of freely roaming through space and landing on planets, then Celestia is well worth a try.
In fact, Celestia is far from a newcomer—its initial release even predates Stellarium, with mobile apps arriving around 2020. As an open-source project, it never achieved the same level of mainstream recognition as Stellarium. I only discovered Celestia recently, after being inundated with space-related short videos and wanting a more “immersive” experience.
What sets Celestia apart is that it doesn’t just let you observe the universe—it lets you move through it freely. By default, Celestia locks the view onto a celestial body itself, and swiping the screen allows you to view it from different angles. Switch from the “Object” perspective to “Camera” mode in the toolbar, and you can observe the universe in a full 360 degrees centered on your current position (though most of the time, all you’ll see is darkness). If you spot a celestial object that catches your interest, simply long-press on it to bring up a menu and select “Go To,” and Celestia will whisk you at light speed to hover above it.
Even better, Celestia offers a wealth of deeply immersive features. Take the time control function in the menu, for example: it allows you to observe changes in celestial bodies across different time scales at any speed you choose. Using the “Lambda Aurigae binary system” that I stumbled upon by chance as an example—when I cranked up the time flow to extreme speeds, I could see the two bodies performing a kind of cosmic “waltz” in space. That, in turn, sparked my curiosity about this special type of system, leading me to learn that “Castor” and the famous “Sirius” are also similar binary systems.
Beyond that, thanks to the power of the open-source community, Celestia supports a highly extensible plugin system. Whether it’s real-world spacecraft, higher-resolution nebulae and galaxy packs, or even fictional celestial bodies and spacecraft from works like 2001: A Space Odyssey and Star Trek, all of these can be loaded into Celestia. Hardcore sci-fi fans can even install script plugins and treat Celestia as a sandbox game set in space.
Overall, while Celestia may not boast a flashy UI, it offers extremely robust functionality and impressive depth. Whether you use it as a pocket science education tool or as a way to spark children’s interest in space exploration, it’s an excellent choice.
Celestia also supports a remarkably wide range of devices and platforms. On desktop, it runs on Windows, macOS, and Linux and can be downloaded from the official website, though the source code itself hasn’t seen updates for many years. On mobile, it supports Android, iOS, and even visionOS, and is still actively updated as of 2026. You can download it from the official website, Google Play, the App Store, and F-Droid.
Halide Mark III: A New Approach to Photographic Style
Platform: iOS / iPadOS
Keywords: Camera, Style Photography
@ElijahLee: The renowned mobile photography app Halide has released a public preview of Mark III, introducing a new image style system called Looks. Looks are not simple color filters; instead, they are stylistic processing methods embedded directly into the act of shooting. Unlike traditional preset filters, they capture and process the image algorithmically at the moment of exposure.
After downloading the latest version of Halide from the App Store, open the app and tap the new “③” button on the screen to switch to Mark III. The app currently offers two Looks: Process Zero II and the black-and-white film style Chroma Noir.
Process Zero II is an upgraded version of the original Process Zero. During capture, it disables all computational photography features, resulting in images with stronger contrast, finer detail, and a more natural overall look. The upgrade adds HDR support, improving performance in high-contrast scenes such as sunsets and vivid color environments, and it can also be used in conjunction with ProRAW. Of course, if you ultimately feel that HDR isn’t right for you, it can be turned off in Settings → Looks.
Mark III can also apply Process Zero II to ProRAW files, preserving both the flexibility of Apple ProRAW and the natural rendering of Process Zero. If you don’t want to completely eliminate the influence of computational photography, Mark III includes a Tone Fusion slider for this exact need. It allows you to adjust highlights and shadows, restoring detail in bright and dark areas.
Chroma Noir is a moderately high-contrast black-and-white film look designed to more faithfully simulate film characteristics. The team emphasizes that it’s not just about tonal adjustment—it also incorporates physical effects such as enhanced halation, using real-world light values to build its imaging engine.
Finally, Halide is experimenting with bringing HDR support to film simulation. Traditional film, despite its centuries-long history, has been constrained by the low dynamic range of print media, making it incapable of truly representing high dynamic range imagery. If film aesthetics don’t appeal to you but you still enjoy black-and-white photography, you can disable film simulation in the app’s settings.
Halide Mark III is expected to officially launch in the summer. What’s currently available is a public preview that can be accessed via an in-app toggle, with incomplete features and a limited selection of Looks. More carefully designed Looks will be added in the future. You can download Halide for free from the App Store, but experiencing Mark III requires a paid subscription: ¥68 per month, ¥148 per year, or ¥398 for a lifetime license.
Escape Launcher: A Minimalist Home Screen to Help You Escape Your Phone
Platform: Android
Keywords: Home Screen, Launcher
@Peggy_: Many years ago, the Android ecosystem was a vibrant landscape of diverse apps. Even within the niche of third-party launchers, there was no shortage of well-known options like Nova Launcher and Lawnchair. As Android has gradually matured and refined its own system features, however, many of these third-party launchers have faded away. In contrast, a new wave of minimalist third-party launchers—built around design philosophies entirely different from stock Android—has begun to flourish.
Escape Launcher is one such app. Its developer has been quietly updating it for over a year now. Through a series of deliberate design choices, Escape Launcher maintains minimalism in both form and function. Visually, its main screen contains nothing but a text-based list of apps. Swiping right brings up phone-wide and per-app usage statistics, while swiping left reveals the full list of installed apps.
The apps shown on Escape Launcher’s home screen are entirely user-defined. On first launch, you’re asked to select a set of essential apps to appear on the home screen, and you can add or remove apps later via settings. Other customizable options are limited to whether the time and date are displayed, and the size and style of the time. As for widgets, testing shows that Escape Launcher only supports displaying a single widget above the app list on the home screen—minimalism taken to an almost frugal extreme. Adding a widget also tends to hurt the visual balance, so it’s probably best to skip widgets altogether.
On the usage statistics screen, Escape Launcher shows your total phone usage for the day, what percentage of the day that usage represents, and whether it exceeds the app’s recommended limit. The main panel then breaks things down by individual app, using red and green arrows to clearly indicate whether usage has increased or decreased compared to the previous day. To further reduce screen time, Escape Launcher also includes an app countdown feature: when enabled, a selected app will display a countdown every time you open it, and only launch once the timer finishes.
In the app list view, while I understand the good intentions behind minimalist launchers avoiding icons altogether, searching for apps in a wall of text undeniably adds friction for users. Fortunately, Escape Launcher puts real thought into lowering that barrier. A prominent search bar makes manual searching easy, and if even tapping that feels like too much, you can enable “auto-start search” in settings. With this turned on, every swipe to the app list immediately brings up the keyboard, allowing you to type and launch apps almost instantly.
Overall, Escape Launcher’s design is clean and restrained, with built-in color themes that are easy on the eyes. Customization options are limited, but the details are thoughtfully handled—for example, long-pressing anywhere opens settings, tapping the clock jumps straight to the system clock app, and tapping the date opens your default calendar. If you’re looking for both a third-party launcher and app usage tracking in one place, Escape Launcher is well worth considering.
Escape Launcher is open source and completely free. You can download and try it via F-Droid.
Posturr: A Posture-Reminder Utility for Mac
Platform: macOS
Keywords: Posture Reminder
@化学心情下2: Perhaps because I’m fairly tall, I often find myself unconsciously hunching forward when using a computer. After years of this habit, it’s taken a toll on my neck. So the question becomes: is there a way to constantly nudge myself into maintaining proper posture while working at my desk?
Posturr is a small utility I recently discovered on GitHub that does exactly that—reminding you to sit properly while using your computer. The idea behind it is fairly simple: using Apple’s Vision framework, Posturr accesses your device’s camera to detect your posture. When you start slouching or bending forward, it immediately blurs the screen so you can’t see anything clearly. Only when you straighten up and sit properly does the screen return to normal.
On first launch, Posturr walks you through a visual calibration process for your posture. You just need to follow the on-screen instructions: when a focus point appears in one corner of the full-screen black calibration page, look at the screen and press the spacebar to complete the setup.
After that, Posturr minimizes itself to the menu bar and begins monitoring via the camera. When your posture starts to slip or you hunch forward, not only does the screen content instantly blur, but the Posturr menu bar icon also changes—from a “standing figure” to a “slipping figure”—as a clear reminder that your posture is off. If you don’t sit up straight, you won’t be able to see the screen properly. In practice, this pushed me to sit more upright just so I could see what I was doing. While it’s admittedly less comfortable than slouching forward with my neck stretched out, it feels well worth it when I think about my long-term health.
In the settings, you can also fine-tune a number of options. For the blur effect, in addition to the default Blur, you can choose Vignette (where the screen gradually darkens) or Border (where the screen edges change color). You can also enable options like automatically pausing the blur when you step away, or when your laptop display becomes the only active screen—adding a bit more flexibility and personalization.
There’s also a statistics section that shows roughly what percentage of your total computer usage time was spent with incorrect posture while detection was enabled.
Overall, Posturr has been extremely useful for someone like me who already struggles with poor posture. This kind of strong, enforced feedback helps me gradually break the habit of slouching and craning my neck while using a computer. If you’re like me and tend to sit poorly when working on a Mac, this is definitely a tool worth trying. You can download and install Posturr for free from its GitHub page.
Sidebar Calendar: A Sidebar Calendar Widget
Platform: macOS
Keywords: Calendar, Widget
@Snow: Sidebar Calendar is a desktop widget app that brings your schedule to the edge of the screen for constant visibility. Whether you’re overwhelmed by a packed workload or hopping from one online meeting to another, simply turning on your screen ensures you won’t miss important events.
Unlike typical apps that live in the menu bar or appear as desktop widgets, Sidebar Calendar places your schedule along the left or right edge of the screen. No matter what’s currently on display, clicking the corresponding screen edge reveals a vertically arranged, hour-by-hour agenda for the day, sliding out from the sidebar. From here, you can view or edit events directly, or quickly create new ones by click-dragging to select a time range. To provide clearer reminders, the app also shows explicit text prompts in the menu bar while an event is in progress.
Right-click within the agenda to open the settings panel, and enable “Settings – Display Reminders” to manage your to-dos for the day directly in the schedule view. The app automatically pulls relevant items from the system Reminders. Double-clicking an empty spot on the timeline lets you quickly add a to-do with the selected time set as its deadline.
In the app’s settings, beyond toggling whether to display to-dos, you can customize theme colors, time formats, hide/show behavior, interface width, keyboard shortcuts, and more. For better schedule awareness, I recommend enabling “Settings – Collapsed Calendar” and “Time Indicator.” The former condenses your events into semi-transparent colored bars pinned to the screen edge, while the latter adds a red time line to clearly show the current moment. Used together, a quick glance at your screen is enough to understand your schedule’s progress and what’s already been completed—helping you avoid missing important tasks.
Google has recently and unexpectedly leaked a preview of the interface for Aluminium OS, a new desktop operating system based on Android and ChromeOS. The preview showcases a multi-window environment and a taskbar-style interface. The system is reportedly expected to see an official release in 2026. Source
Tesla plans to discontinue the Model S and Model X
Rumors You Can Just Glance At
Xiaomi launches the REDMI Turbo 5 series smartphones
On January 29, Xiaomi officially unveiled the REDMI Turbo 5 series smartphones.
Among them, the REDMI Turbo 5 Max is powered by the Dimensity 9500s chip built on a 3nm process. According to the company, the Dimensity 9500s adopts an all–big-core architecture and features a large cache. The CPU reaches a peak clock speed of up to 3.73GHz, with cache capacity increased to 29MB. In terms of display and design, the new phone comes with a 6.83-inch 1.5K-resolution screen using new M10 emissive materials, offering a peak brightness of up to 3500 nits and supporting both 3840Hz PWM dimming and DC dimming.
For battery life and charging, the REDMI Turbo 5 Max packs a 9000mAh battery. Xiaomi claims its endurance rivals that of some 10,000mAh devices. It supports 100W wired fast charging, the 100W PPS protocol, and 27W wired reverse charging. On the imaging front, the REDMI Turbo 5 Max features a 50MP primary camera with an ƒ/1.5 aperture and 1.6µm pixels, and supports high dynamic range video recording.
The REDMI Turbo 5 Max is available in four colors: black, blue, white, and orange. The base configuration includes 12GB of RAM and 256GB of storage, with a starting price of 2,499 yuan.
The standard REDMI Turbo 5 uses the Dimensity 8500-Ultra chip built on a 4nm process. It features a 6.59-inch display, while most other specifications remain consistent with the REDMI Turbo 5 Max. The REDMI Turbo 5 is equipped with a 7560mAh battery, supports 100W wired fast charging, 27W wired reverse charging, and is compatible with 100W-class PPS fast charging. The standard REDMI Turbo 5 comes in three colors—black, green, and white. The base model offers 12GB of RAM and 256GB of storage, starting at 1,999 yuan. Source
WeChat issues a special enforcement notice targeting third-party violations
On January 29, WeChat released a special enforcement announcement targeting third-party violations, with a focus on cracking down on false marketing, excessive marketing, and activities that threaten data security.
WeChat stated that, based on recent user complaints and reports, it has launched concentrated governance actions against third-party violations that seriously disrupt the platform ecosystem and infringe on user rights. These mainly include false marketing and related fraudulent activities, excessive marketing and induced sharing behaviors, as well as actions that endanger privacy and security, such as illegally obtaining user data and using plug-ins or external tools. In accordance with relevant laws, regulations, and multiple WeChat platform agreements, the platform will impose tiered penalties on violating links, mini programs, and third-party apps, including access restrictions, functional bans, removal from the platform, and account suspensions.
WeChat further noted that false marketing often involves fabricated cashback or lottery schemes and impersonation of official identities for fraud, frequently resulting in financial losses and illegal collection of personal information. Excessive marketing harms user experience through high-frequency push notifications and forced redirects. Meanwhile, plug-ins and data theft activities manipulate WeChat functions through automated scripts and technical means, posing direct threats to platform security. WeChat said it will continue to clear such violations through a combination of technical inspections and user reports. Source
MiniMax (Xiyu Technology) releases the MiniMax Music 2.5 model
On January 29, MiniMax (Xiyu Technology) officially released its music generation model, MiniMax Music 2.5. The model supports full-section tag control and precisely handles 14 musical structure variants, including Intro, Bridge, Interlude, Build-up, and Hook, enabling expressive creation of highly complex musical works.
For Mandarin pop music scenarios, MiniMax Music 2.5 has been deeply optimized to cover a wide range of styles, including ballads, rap, fully Chinese lyrics, and mixed Chinese–English compositions. Vocal synthesis has also been enhanced, delivering smoother and more nuanced pitch transitions, more natural vibrato dynamics, and flexible switching between chest and head resonance. In male–female duet scenarios, MiniMax Music 2.5 presents more coordinated vocal interplay, supporting alternating vocals and multi-layered harmony arrangements.
In terms of timbre, the model supports more than 100 instruments and includes optimized mixing to keep vocals and accompaniment clearly separated. At the same time, MiniMax Music 2.5 is deeply integrated with professional creative workflows. According to the company, it can be applied to professional scenarios such as film and television scoring, dynamic game audio effects, studio-grade pop music production, and branded sound design. Source
Microsoft commits to not using Copilot interactions to train AI
On January 28, Microsoft issued a statement responding to long-standing public concerns over its data collection and privacy practices. The company reiterated that users retain full control over their personal data and can access, transfer, or delete relevant information at any time. Data will only be used for purposes such as personalized advertising with explicit user consent.
Addressing privacy concerns from enterprise and individual users regarding Microsoft 365 Copilot, Microsoft further pledged to strictly isolate user prompts, generated content, and business data, and clearly stated that none of this information will be used to train any AI systems, including foundational large language models (LLMs). Microsoft said that Copilot fully inherits Microsoft 365’s existing identity management, permission controls, and compliance framework, ensuring that organizational data remains within the company’s own tenant environment and is not accessed externally or exposed. Source
On January 29, Google announced the introduction of new AI capabilities to the Chrome browser, deeply integrating Gemini 3 technology into the browser’s sidebar to deliver a more intelligent interactive experience.
At the same time, Chrome is rolling out a new feature called “Automated Browsing,” available to users subscribed to Google AI Pro or Ultra. This feature allows AI to carry out multi-step tasks through natural language commands, including opening pages across websites, filling out forms, comparing prices, and managing appointments and subscriptions, thereby reducing repetitive operations. For sensitive actions such as payments or content publishing, the system still requires manual user confirmation.
In addition, the newly added Gemini panel on the right side of the browser can work in conjunction with services such as Gmail, Calendar, Maps, Flights, and Shopping. Based on cross-app data, the AI can provide intelligent suggestions—for example, extracting itinerary details from emails, matching them with flights, and automatically generating calendar schedules. Chrome also includes a built-in generative image tool called Nano Banana, allowing users to generate or edit images directly within the browser using text prompts. Source
Tesla plans to discontinue the Model S and Model X
Elon Musk revealed during the latest earnings call that Tesla plans to cease production of the Model S and Model X in the second quarter of 2026, in order to free up manufacturing capacity at the Fremont factory for the Optimus humanoid robot project. He said the move reflects the company’s strategic shift toward autonomous driving and robotics, and advised consumers interested in these models to place orders as early as possible.
The Model S was launched in 2012, followed by the Model X in 2015, and both long served as key parts of Tesla’s flagship lineup. As the company has increasingly focused its resources on the Model 3 and Model Y, sales of the two high-end models have continued to decline. Data shows that sales classified under the “other models” category fell by more than 40% year over year in 2025, while Tesla’s profits in the fourth quarter also saw a notable drop. Source
Rumors You Can Just Glance At
According to reports, Logitech G is set to officially launch the G325 LIGHTSPEED wireless over-ear headset. The headset features 32 mm drivers with an impedance of 32 Ω, and is equipped with an omnidirectional beamforming microphone that supports AI noise reduction and 24-bit audio. In terms of design, it uses seamless breathable knitted fabric, a soft headband, and dual-layer ear cushions, with a total weight of 212 g. For connectivity, it supports Bluetooth 5.2 as well as Logitech’s 2.4 GHz LIGHTSPEED wireless mode, offering a battery life of over 24 hours. The headset is priced at €79.99. Source
The Raspberry Pi Foundation is reportedly preparing an intelligent display module for the Compute Module 5. Essentially an adapter board, this module integrates the CM5’s computing power and energy efficiency directly into compatible displays, while also providing an additional HDMI output to drive a second independent video signal, catering to multi-display setups or complex information presentation needs. The module board also reserves an M.2 expansion slot, allowing users to add AI acceleration modules and leaving headroom for local inference and intelligent applications. Designed in accordance with the Intel SDM specification, the intelligent display module is positioned for use cases such as flight information systems, retail and enterprise digital signage, and industrial-grade display terminals. The module is expected to launch later this year and will make its first public appearance next week at ISE 2026 in Spain. Source
Moonshot AI releases the Kimi K2.5 model and open-sources it
NVIDIA officially introduces RTX Remix Logic
Intel officially rolls out XeSS 3 multi-frame generation feature
Google upgrades anti-theft mechanisms in Android 16
Apple Creator Studio officially launches
On January 28, Apple officially launched Apple Creator Studio, the creator tool suite it announced not long ago. The product is a one-stop bundled subscription service aimed at professional creators, covering nearly all of Apple’s professional creative software. With a single subscription, users gain access to the Pro App suite on Mac and iPad, the iWork suite, and Pixelmator Pro—the professional image editing and design tool Apple previously acquired—along with additional AI-powered features.
Moonshot AI releases the Kimi K2.5 model and open-sources it
On January 27, the Moonshot AI team announced the release and open-sourcing of its latest model, Kimi K2.5. At the same time, version K2.5 of the Kimi AI assistant also went live, and the original K2 model in the chat interface has been automatically upgraded to K2.5.
The model is currently Moonshot AI’s most intelligent model, achieving open-source state-of-the-art performance across agents, coding, images, video, and a wide range of general intelligence tasks. It is also Kimi’s most versatile model to date, featuring a native multimodal architecture that supports both visual and text inputs, thinking and non-thinking modes, as well as conversational and agent-based tasks.
According to examples provided by Moonshot AI, the model can generate complete front-end page code from natural language instructions and handle interactive logic such as dynamic layouts and scroll-triggered behaviors. Combined with its visual capabilities, Kimi K2.5 can also break down user-provided screen recordings, analyze the underlying interaction structures, and generate the corresponding implementation code.
Beyond single-agent capabilities, Kimi K2.5 introduces a new agent cluster mechanism. This allows the model, when faced with complex tasks, to dynamically spawn multiple sub-agents to work in parallel on different subtasks instead of operating as a single agent.
Alongside the Kimi K2.5 model, Moonshot AI also released Kimi Code, a programming tool designed for developers. It can run in a command-line environment and supports integration with mainstream editors and IDEs such as VS Code, Cursor, the JetBrains suite, and Zed.
Kimi K2.5 is now available on the Kimi website, mobile app, and its API platform. General users can access its features through different modes, while developers and enterprises can integrate it via the API. Source
NVIDIA officially introduces RTX Remix Logic
On January 27, NVIDIA updated its NVIDIA App to add a new feature called RTX Remix Logic, which allows mod creators to dynamically trigger visual effects based on real-time in-game events (such as player position or button inputs) without accessing the game’s source code. For example, mod authors can set up simple “if… then…” rules (such as “if the player walks here, start raining”) to make the visuals of classic games change in real time according to player actions—without requiring knowledge of complex programming.
To lower the technical barrier, NVIDIA introduced a highly visual, no-code, node-based interface. Creators can build complex interaction logic simply by dragging and connecting “trigger” nodes with “action” nodes. The interface also provides dedicated sliders for fine-tuning parameters and supports real-time preview directly within the Remix editor.
For advanced developers, the framework also supports plugin extensions, allowing the creation of custom event triggers. In NVIDIA’s demo, opening a door in the RTX version of Half-Life 2 instantly triggered a dramatically different “Ravenholm multiverse” scene. Source
Intel officially rolls out XeSS 3 multi-frame generation feature
On January 27, Intel began pushing the latest graphics driver update to Arc GPUs, officially bringing the XeSS 3 multi-frame generation feature.
At its core, XeSS 3 follows a Multi Frame Generation (MFG) approach: after each traditionally rendered frame, up to three AI-generated “interpolated frames” are inserted. This significantly increases frame rates and improves animation smoothness without adding extra load to the game’s native rendering pipeline. Intel emphasized that XeSS 3 relies on an optical flow network, using motion vectors and depth buffers from the game to predict and generate these additional frames.
Unlike some competing solutions, XeSS 3 performs optical flow calculations only once per batch of AI-generated frames. This design makes algorithm development more complex and time-consuming, but helps strike a balance between performance and visual quality.
In addition to introducing XeSS 3, the driver update also fixes several known issues. These include a crash bug in the Pragmata Sketchbook demo under certain conditions on Arc B-series discrete GPUs and on Core Ultra Series 2 processors with integrated Arc graphics. Intel also corrected an error in its graphics software where the variable refresh rate (VRR) status was inaccurately reported on the display settings page. Source
Google upgrades anti-theft mechanisms in Android 16
On January 27, Google’s Android security team published a post announcing the deployment of multiple “theft protection” mechanisms, elevating phone theft protection from simple device recovery to broader data and financial security.
For devices running Android 16 and later, Google has significantly strengthened the “Identity Check” feature. Previously limited to untrusted locations, this update expands its coverage to all apps that invoke Android’s Biometric Prompt. Key tools such as third-party banking apps and Android password managers can now automatically receive system-level mandatory biometric verification protection, meaning that even if a thief knows the lock-screen passcode, they still cannot easily access sensitive data.
Google has also adjusted its anti-guessing mechanism for screen unlocking. Users can now find a separate toggle called “Authentication Failure Lockout” in settings. When the system detects too many failed login attempts, it will automatically lock the device. The new mechanism not only extends the lockout time after repeated failures, but also introduces intelligent detection: if the system finds that the same incorrect passcode is entered repeatedly (for example, when a child accidentally taps the same spot multiple times), those attempts will no longer be counted.
Finally, in terms of post-loss remediation, Google has optimized the “Remote Lock” tool available on Android 10 and later devices. When users lock their phone remotely via the web-based Find My Device service, they can choose to add an extra “security question or challenge” to verify that the operator is indeed the device owner. Source
Sennheiser releases CX 80U and HD 400U wired earphones
Sonos releases multi-channel power amplifier Amp Multi
Bigme unveils B10 color E Ink office notebook
Three updates related to Google Gemini
Yahoo launches AI Q&A search tool Yahoo Scout
Alibaba releases Qwen3-Max-Thinking
Anthropic launches interactive Claude app
OpenAI releases scientific research AI collaboration space Prism
Rumors You Can Just Glance At
Sennheiser releases CX 80U and HD 400U wired earphones
On January 26, Sennheiser officially unveiled the CX 80U wired in-ear earphones and the HD 400U wired over-ear headphones. As upgraded versions of the CX 80S and HD 400S, both new models replace the traditional 3.5 mm jack with a USB-C connector and feature a built-in digital converter that supports 24-bit/96 kHz lossless audio playback, along with an in-line remote equipped with an integrated MEMS microphone.
The CX 80U uses a dynamic driver with an impedance of 28 Ω and a frequency response of 17 Hz–20 kHz, and comes with three sets of silicone ear tips in different sizes. The HD 400U features a 32 mm dynamic driver with an impedance of 18 Ω and a frequency response of 18 Hz–20 kHz, a foldable closed-back design, and includes a carrying pouch. Both products went on sale on January 26, 2026, with the CX 80U priced at USD 40 and the HD 400U at USD 100. Source
Sonos releases multi-channel power amplifier Amp Multi
On January 27, Sonos officially introduced its first professional-grade multi-channel streaming power amplifier, Amp Multi, designed specifically for large residential properties and complex custom installation scenarios. Amp Multi features a gallium nitride (GaN) power architecture and Class D post-filter feedback (PFFB) technology, delivering eight channels of 125 W (8 Ω) each in a compact 1.5U chassis, and can be configured as four independently controlled zones. Its hardware configuration includes a quad-core 1.9 GHz A55 processor, Wi-Fi 6, and dual Ethernet ports. It supports Apple AirPlay 2 and Spotify Connect, and each channel can drive up to three Sonos Architectural series speakers. Pricing has not yet been announced, and the product is expected to go on sale in the coming months through authorized installers. Source
Bigme unveils B10 color E Ink office notebook
On January 27, Bigme launched the 10.3-inch B10 4G color E-Ink office tablet on overseas platforms. The device is powered by a Dimensity 1080 processor and comes with 8 GB of RAM, 256 GB of storage, and Android 14. It supports 4G connectivity across major bands and is equipped with a 6,900 mAh battery, an eight-microphone array, a 4,096-level pressure-sensitive stylus, and fingerprint recognition. The B10 features a Kaleido 3 color E-Ink display with up to 30 FPS refresh rate and 36 levels of adjustable dual-tone (warm/cool) color temperature. It also includes both front and rear cameras, supporting document scanning, OCR text recognition, and video conferencing.
The Bigme B10 adopts a design with eight customizable touch keys and comes with four built-in AI models, supporting features such as meeting speech-to-text transcription. An optional keyboard accessory is available. The B10 is priced at USD 699, with a pre-sale price of USD 594. Source
Three updates related to Google Gemini
On January 27, Google announced the introduction of Agentic Vision capabilities to the Gemini 3 Flash model. This feature aims to shift image understanding from static processing to a dynamic process based on a “Think, Act, Observe” loop. By combining visual reasoning with Python code execution tools, the model can autonomously perform actions such as zooming in to inspect details, rotating and cropping images, drawing bounding-box annotations, conducting visual math calculations, and generating charts. Technical tests show that with agentic vision enabled, Gemini 3 Flash achieves a 5–10% quality improvement across most visual benchmarks and partially mitigates hallucination issues in long-form visual tasks. The feature is now officially available via the Gemini API (including Google AI Studio and Vertex AI), where developers can enable Code Execution in the toolbar to use it. Source
At the same time, Google also announced that Gemini 3 has been officially set as the default underlying model for its global “AI Overviews” feature, and that it has introduced a seamless “AI Mode” conversational experience. This allows users to ask follow-up questions directly below AI Overviews, turning search from one-way information display into contextual, streaming conversations. Google said that the model upgrade also improves reasoning quality for complex questions. The new AI Overviews and AI Mode conversation features are now available globally on both mobile and web search, with the basic service offered free to all users. Source
Finally, Google announced that its entry-level AI subscription plan, Google AI Plus, has officially launched in 34 additional countries and regions worldwide, including the United States. Priced at USD 8 per month, the plan is aimed at everyday users and provides access to the Gemini 3 Pro model, Deep Research, and the Nano Banana Pro model. Subscribers also receive 200 GB of Google One storage and unlock AI filmmaking tool Flow, image-to-video tool Whisk, and advanced research features in NotebookLM (including support for up to 100 input sources and audio overviews). Notably, existing Google One Premium 2 TB members will automatically receive all AI Plus features, and new subscribers can enjoy a promotional price of USD 4 per month for the first two months. Source
Yahoo launches AI Q&A search tool Yahoo Scout
On January 27, Yahoo announced the launch of its AI-powered Q&A search feature, Yahoo Scout, which is supported by Anthropic’s Claude. The feature deeply integrates 50 billion user profiles and 18 trillion consumer behavior signals accumulated by Yahoo over the past three decades. It supports synthesizing information from across the web—including Yahoo’s own channels such as News, Finance, and Sports—using natural language, and presents the results in the form of structured tables, rich multimedia cards, and prominent blue source links. Each answer contains an average of up to nine cited sources. At the same time, the feature has been embedded across Yahoo’s entire product lineup, offering enhanced experiences such as email summaries, real-time financial market analysis, news comment trend extraction, and in-depth breakdowns of sports events.
Yahoo Scout is currently available in beta in the United States, with access via a standalone web version as well as iOS and Android apps. The service remains free to use and is planned to be monetized through bottom-page advertising and affiliate commissions. Source
Alibaba releases Qwen3-Max-Thinking
On January 26, Alibaba released its flagship reasoning model Qwen3-Max-Thinking, with total parameters exceeding one trillion. The model underwent large-scale reinforcement learning post-training and has enhanced its native agent capabilities for autonomously calling tools across multiple key benchmarks, enabling on-demand use of search engines and code interpreters. It also supports test-time scaling techniques, significantly improving reasoning performance. On major reasoning benchmarks, it surpasses Gemini 3 Pro, while also substantially reducing model hallucinations.
Qwen3-Max-Thinking is now available on Qwen Chat, and its corresponding API (model name: qwen3-max-2026-01-23) has been released via Alibaba Cloud. Source
Anthropic launches interactive Claude app
On January 26, Anthropic announced the launch of an interactive app for its Claude chatbot, allowing users to directly connect everyday workplace tools such as Slack, Canva, Figma, Box, and Clay within the chat interface. Salesforce integration will be added later. Claude can send Slack messages on users’ behalf, generate charts, or access cloud files within these services, depending on which app permissions are enabled.
In a blog post, Anthropic said that tasks such as data analysis, content design, and project management are inherently better suited to dedicated visual interfaces. When combined with Claude’s intelligence, users can significantly improve efficiency in ideation, revision, and iteration. The new feature is currently available to Claude Pro, Max, Team, and Enterprise subscribers, and can be enabled and managed at claude.ai/directory. Source
OpenAI releases scientific research AI collaboration space Prism
On January 27, OpenAI officially unveiled Prism, an AI-native collaborative workspace designed specifically for scientific research, and simultaneously announced the acquisition of the cloud-based LaTeX platform Crixet as its underlying infrastructure. Prism deeply integrates the GPT-5.2 Thinking model, enabling AI to directly understand paper structures, formulas, and references within project environments. It supports automatically generating TikZ diagrams, retrieving and citing papers from arXiv, auto-formatting bibliographies, and converting handwritten formulas or sketches on whiteboards into LaTeX code. In addition, Prism offers real-time multi-user collaboration and can generate graduate-level course plans and accompanying exercises based on specific research topics.
Prism is currently available for free to all users with a personal ChatGPT account, supporting the creation of unlimited projects and collaboration seats. It will also be rolled out soon to ChatGPT Business, Team, Enterprise, and Education plans. Source
Rumors You Can Just Glance At
On January 28, Samsung began teasing a new privacy protection feature that will be introduced on Galaxy devices. The feature is designed to address the risk of “shoulder surfing” in public places through deep integration of hardware and software. It is said to adjust display visibility at the pixel level to selectively obscure password entry, specific app interfaces, and notification pop-ups, allowing only the person directly in front of the screen to clearly view certain content. Source
On January 26, Meta announced plans to begin testing premium subscription services for its three major apps—Instagram, Facebook, and WhatsApp—in the coming months. The subscription system aims to unlock more powerful productivity, creativity, and AI-enhanced features through paid plans, while keeping the core basic experience free. Source
According to reports from several media outlets and a settlement agreement recently submitted to a federal court in San Jose, Google has agreed to pay USD 68 million to settle a class-action lawsuit over illegal recordings made by its Google Assistant voice assistant. The case centers on Google Assistant’s “false accept” mechanism, in which devices are activated and capture sensitive user audio for human review or targeted advertising even without hearing the “OK Google” wake word. This is similar in nature to Apple’s previous settlement of USD 95 million over comparable issues with Siri. The settlement covers U.S. users who, since May 18, 2016, have had accounts associated with devices preinstalled with Google Assistant (including Pixel phones, Google Home smart speakers, and Nest Hub displays). Under the agreement, eligible device purchasers are expected to receive compensation of USD 18 to USD 56, while affected individuals living in environments with such devices may claim USD 2 to USD 10. Source
According to internal documents obtained by the New Mexico Attorney General’s Office, Meta CEO Mark Zuckerberg personally led and vetoed a decision to add parental control features to the company’s AI chatbots, instead pushing for looser safety restrictions. The allegations stem from a lawsuit filed by the state of New Mexico accusing Meta of failing to protect children from sexual harassment. A previous investigation by The Wall Street Journal had also pointed out risks that Meta’s AI bots could engage in sexually explicit conversations with minors or impersonate them. As a result, Meta suspended access to AI character features for teen accounts last week while developing relevant parental control tools. Source
On January 27, Google accidentally leaked screenshots of a desktop interface codenamed Aluminum OS (ALOS) on the Chromium bug tracker. The images came from an HP Elite Dragonfly Chromebook powered by a 12th-generation Intel Core processor, running system version ZL1A.260119.001.A1 based on Android 16. The screenshots show a desktop environment optimized for large-screen multitasking, featuring a taskbar, second-level time display, an M3E battery icon, a Gemini AI entry point, and a screen recording control panel. The bundled Chrome browser also supports desktop extensions. This further confirms earlier rumors that Google is testing a deeply integrated version of Android and ChromeOS on Chromebooks. Source
X faces EU investigation over pornographic deepfake content spread by Grok
China’s National Press and Publication Administration releases game approval list for January 2026
State Post Bureau deploys measures to ensure parcel delivery services during the 2026 Spring Festival travel season
Apple releases new AirTag and other new products
HMD unveils multiple new products including the HMD Watch X1 and P1
Microsoft releases the Maia 200 AI accelerator chip
Nintendo releases “Chatty Huahua”
Rumors You Can Just Glance At
Adobe apps gain native support for Windows on ARM
Adobe has recently released version 26.0 updates for Premiere Pro, After Effects, Audition, and Media Encoder, officially adding native support for Windows on ARM. These apps can now run directly on Windows devices powered by Qualcomm’s Snapdragon X series ARM processors. The software requires at least Windows 11 (Build 26100.2033), a minimum of 16 GB of RAM, and Qualcomm Adreno GPU driver version 31.0.121.1 or later. Adobe also made it clear that older versions (25.x and earlier) were not designed or tested for Windows on ARM devices, and normal operation cannot be guaranteed.
However, the native ARM versions still come with certain limitations. For example, Premiere Pro currently does not support Loudness Radar, Wraptor DCP export, GoPro CineForm import and export, or P2 Movie format export. After Effects lacks the Cinema 4D renderer and only supports third-party plugins that have been specifically compiled for Windows on ARM. Source
X faces EU investigation over pornographic deepfake content spread by Grok
In recent weeks, users on X in multiple countries have reported that Grok has generated pornographic deepfake content and that it has been circulated on the platform. Recently, the UK communications regulator has formally launched an investigation into whether X has violated the UK Online Safety Act. France and India have also followed by issuing formal accusations against X. As global criticism continues to escalate, the European Union has officially issued an investigation order. This probe covers all 27 EU member states and focuses on whether X has adequately assessed and mitigated the relevant risks in accordance with the Digital Services Act.
The act sets strict regulatory red lines for harmful and illegal online content. If online platforms fail to properly address illegal content and disinformation, or otherwise breach the relevant rules, the EU may impose fines of up to 6% of their global annual revenue. Source
China’s National Press and Publication Administration releases game approval list for January 2026
On January 26, China’s National Press and Publication Administration officially released the Approval Information for Domestic Online Games for January 2026 and the Approval Information for Imported Online Games for January 2026. A total of 177 domestic games were approved, including titles such as Soul Land: Journey and I Am MT 5. Five imported games were approved, including Rainbow Six: Siege, Angry Birds: Classic Returns, and Gold Rush in the Secret Realm. Source 1, Source 2
State Post Bureau deploys measures to ensure parcel delivery services during the 2026 Spring Festival travel season
China’s State Post Bureau recently issued the Work Plan for Ensuring Delivery Services During the 2026 Spring Festival Travel Season (hereinafter referred to as the “Work Plan”), stressing the need to balance meeting public demand for parcel delivery with protecting the legitimate rights and interests of couriers, and to effectively ensure delivery services during the 2026 Spring Festival travel period. Postal and courier companies are required to focus on pre-holiday consumer demand for New Year goods, collect parcels in a timely manner, and provide delivery services as agreed. China Post Group is required to ensure the continuous provision of universal postal services. All courier companies are also instructed to scientifically forecast changes in business volume, reasonably formulate their holiday operating plans, announce their service hours and service areas to the public in advance, and strictly provide services in accordance with the published information to meet basic delivery needs during the Spring Festival holiday.
At the same time, multiple courier companies have released service notices for the 2026 Spring Festival period. Announcements from companies including SF Express, JD Logistics, and Deppon indicate that they will continue to guarantee delivery services during the holiday, while charging a peak-season resource adjustment fee to cope with increased demand for resource allocation. Source
Apple releases new AirTag and other new products
On January 26, Apple introduced a new AirTag, with pricing unchanged at RMB 249 for a single unit and RMB 849 for a four-pack. The new AirTag is equipped with the same second-generation ultra-wideband chip found in devices such as the iPhone 17, making it easier to locate and extending the tracking range by up to 50% compared with the previous generation. The upgraded Bluetooth chip also allows users, for the first time, to use Precision Finding to locate AirTag via Apple Watch Series 9 and later models, as well as Apple Watch Ultra 2 and later models. The built-in speaker is also 50% louder than before. The new AirTag seamlessly integrates the item location sharing feature, allowing users to temporarily and securely share the location of lost items with trusted third parties. Apple has partnered directly with more than 50 airlines so that, through this feature, users can share the location of their misplaced items with participating airlines’ customer service teams to help recover delayed or lost luggage. Source
On January 26, Apple also released a new Apple Watch Unity Connection braided solo loop band to commemorate Black History Month and celebrate the power of connection. The band will be available in Apple Store retail locations later this week, priced at RMB 779. Source
HMD unveils multiple new products including the HMD Watch X1 and P1
HMD has unveiled two new smartwatches, the Watch X1 and Watch P1. The Watch X1 features a 1.43-inch AMOLED display with a peak brightness of 600 nits and a resolution of 466 × 466. It supports always-on display (AOD) and has an IP68 rating for dust and water resistance, with battery life of up to five days with AOD enabled. The Watch P1 comes with a 1.83-inch LCD screen with a peak brightness of 550 nits and a resolution of 240 × 284, an IP67 dust and water resistance rating, and battery life of up to four days. Both run on HMD’s proprietary RTOS and can be managed via the HMD Watch app. In addition, HMD also released six new TWS earbuds: the HMD Dub X50 Pro, Dub X50, Dub S60, Dub P70, Dub P60, and Dub P50. Source
HMD Watch X1
Microsoft releases the Maia 200 AI accelerator chip
On January 26, Microsoft unveiled the Maia 200 chip. Built on TSMC’s 3nm process, the Maia 200 features native FP8/FP4 tensor cores, a 216 GB HBM3e memory system with 7 TB/s of bandwidth, and 272 MB of on-chip SRAM. Microsoft described the Maia 200 as a breakthrough inference accelerator that is 30% faster than its predecessor, the Maia 100. The chip has already been deployed in Microsoft’s U.S. Central data center and its U.S. West 3 data center. Source
Nintendo releases “Chatty Huahua”
Last week, Nintendo announced its new accessory, the “Talking Flower,” on its official website and YouTube channel. The accessory is designed after a classic NPC from the Super Mario series. Compact in size, the Talking Flower is suitable as a decorative item for the home or desktop. When left alone, it will occasionally “talk to itself,” roughly twice per hour. Users can also interact with it by pressing the button in front of the flower, with each press delivering a different surprise. The Talking Flower is currently known to support 10 languages, including Chinese, English, Japanese, and Korean, with the option to switch between them. It is scheduled to go on sale on March 12 this year, priced at 4,500 yen. Source
Rumors You Can Just Glance At
Yesterday, Jia Yueting posted on Weibo claiming that FF’s EAI robot has officially completed U.S. regulatory certification for its first product and is about to begin sales. He also said that regulatory certification for other first-batch products is progressing rapidly and is expected to be completed soon, aiming to achieve a strategy of “launch and sell immediately, sell and deliver immediately,” with deliveries starting in the same month as release. Source
Recently, Tencent held its 2025 annual employee meeting. At the meeting, Pony Ma said that Tencent is cautiously evaluating AI integration within the WeChat ecosystem, as an “all-in-one AI suite” may not be favored by everyone. In the future, Tencent will continue to adhere to a decentralized approach, planning WeChat’s intelligent ecosystem in a way that balances user needs and privacy protection. Source
According to overseas reports, Google will redesign the UI and interactions of Android 17 based on Material 3 Expressive. The most noticeable change will be a significant increase in blur effects. While this differs somewhat from Apple’s “Liquid Glass” style, Google’s blur treatment appears more restrained and lightweight. Whether this redesign will be limited to system settings or rolled out more broadly across apps remains to be seen. Source
According to a report by Jiemian News on January 26, Tencent’s AI assistant “Yuanbao” has begun internal testing of a new feature called “Yuanbao Party.” Positioned as a multi-user social space, the feature allows users to create or join “parties” for AI-assisted chatting, screen sharing, collaborative movie watching, music listening, and other interactive activities. It is built on Tencent Meeting’s audio and video capabilities and integrates sharing links with WeChat and QQ, supporting one-tap invitations for friends to join. Unlike traditional single-user chatbots, Tencent is attempting to increase the stickiness of its AI tools through social connections. In “Yuanbao Party,” AI can act as a social facilitator or productivity tool, for example by livening up group chats or managing schedules and tasks. Source
On January 26, Blizzard Entertainment released a 35th anniversary video and announced that it will make “major game announcements” for four titles on January 29, February 4, February 9, and February 11, respectively: World of Warcraft, Overwatch, Hearthstone, and Diablo. Source
Microsoft’s Windows 11 update released in January 2026 triggered a series of bugs, including issues such as inability to shut down, corrupted apps, and apps freezing. Microsoft subsequently released multiple emergency patches, including KB5078127 and KB50781432. However, the latest reports indicate that these patches have caused some users to encounter the “UNMOUNTABLE_BOOT_VOLUME” error, preventing their systems from booting. Microsoft is aware of the issue and has advised affected users to perform manual recovery. Source
@ElijahLee: TabTab is a window switcher designed specifically for macOS. It is similar to macOS’s native ⌘ Command + Tab feature, but more powerful. In addition to switching between application windows, TabTab can also switch browser tabs and tabs within certain apps, helping you quickly locate your target among multiple windows and tabs and improve multitasking efficiency.
First is the window-switching feature. To distinguish itself from the system’s default shortcut, TabTab uses ⌥ Option + Tab to trigger window switching. Its switcher interface is displayed in a vertical layout, and the app offers three positioning options: docked to the left edge of the screen, the right edge, or centered, which users can choose freely. Holding down the shortcut cycles through app icons, and releasing it switches to the selected window, consistent with the native behavior. For users accustomed to the default shortcut, the app can also be configured to use ⌘ Command + Tab to replace the system’s built-in function.
TabTab supports search. Using the shortcut Shift + Command + L brings up the switcher window and activates search. By typing keywords, you can search across app names, window contents, and more. It supports searching by app name, window title, tab title, and even URL.
Based on this, TabTab also supports tab switching. For applications that support tabs—such as browsers (Safari, Google Chrome, Arc, etc.), development tools (VS Code, Xcode, Cursor, etc.), system apps (Finder, Terminal, etc.), as well as apps like Notion and Figma—it allows you to search directly for content, locate the corresponding tab, and switch to it. This is faster and more precise than macOS’s native switching. However, based on my brief testing, the search function is basically unusable on systems set to Chinese.
TabTab emphasizes an experience close to native macOS usage. Its interface is clean, performance is lightweight, memory usage is extremely low, and everything runs locally. In the settings, the app also provides various customization options. The mouse-trigger feature is disabled by default; when enabled, moving the cursor to the edge of the screen will bring up the switcher. The tab-group feature is enabled by default; disabling it will list all tabs individually. You can also set a blacklist for specific apps to prevent them from being included in window or tab switching.
You can download TabTab for free from its official website. The free version supports switching between only five windows and tabs, and search is also limited to five windows and tabs. The paid version unlocks all features, including unlimited windows and tabs and full search functionality. Pricing is based on the number of devices activated: USD 15.19 for one device, USD 19.99 for two devices, and USD 26.39 for three devices. It is a one-time purchase with lifetime access.
KashCal: Syncing Your Calendar with iCloud on Android
Platform: Android
Keywords: calendar sync
@Peggy_: For many users who use both iOS and Android devices, keeping data in sync has always been a headache. As a syncing solution deeply rooted in Apple’s ecosystem, iCloud can be said to be almost unbeatable within that ecosystem.
However, for various reasons, I first used a Google account as my primary account, then later switched to Outlook because of syncing issues. After all that back and forth, calendar syncing—something that was never a problem before—has now become one. At present, I mainly use Apple products, and while iCloud’s syncing is far from perfect, it is indeed the most convenient option. So I started thinking about migrating my data again. Around that time, I came across a calendar app that can sync iCloud account data on Android: KashCal.
As a calendar app, KashCal claims to prioritize privacy, so it can be used completely offline. However, doing so means giving up syncing, making the data visible on only a single device—something most of us probably don’t need from a fully offline local calendar app.
When it comes to data syncing, KashCal says it has implemented multiple security measures. It does not analyze user data, and its codebase is open source, which means anyone can audit the code to help ensure security. As its killer feature, KashCal allows you to use your iCloud account as the syncing account.
Open the app’s settings, and you can choose to sign in to iCloud. On this screen, you need to enter your Apple ID and generate an app-specific password online exclusively for this application. The app provides detailed instructions on how to generate this password on the login page. This way, you don’t have to worry about your main Apple ID password being exposed. Once logged in, you can use your iCloud account to sync calendar events across devices.
Of course, KashCal also includes the basic features found in other calendar apps. For example, you can subscribe to calendars to bring in calendars from other accounts or even shared calendars created by other people. You can also import local calendar files in .ics format, or export the calendar data from within the app.
If you want to use your iCloud account to sync schedules across both iOS and Android devices, there are currently not many options available. KashCal may be one of the more hassle-free solutions you can find right now. You can download and try KashCal via F-Droid. The app is open source and completely free to use.
aShell You: An ADB Command Tool with Material You Design
@大大大K: The fun of tinkering with Android is not just about installing countless apps. A deeper level of play lies in using ADB commands to query system outputs and parameters. In the past, using ADB usually meant connecting to a computer. Later, a number of apps emerged that made it possible to run ADB commands locally through various methods—one of which is the aShell You app introduced here.
Unlike the “clunky and utilitarian” command-line tools people might be used to, aShell You adopts a Material You design. Its interface is not only modern and visually appealing, but also supports dynamic theme colors and linear motor haptic feedback. Even the gear icon on the settings page has an animation, giving the whole app a very polished feel.
In terms of functionality, aShell You supports three ADB debugging modes. If you want to run higher-privilege commands directly on the local device, you must execute them via Root or Shizuku. Given that rooting phones has become increasingly impractical in recent years, Shizuku is probably the simplest and most straightforward option. Alternatively, you can connect to the current device via wireless ADB debugging, similar to how Shizuku is activated. aShell You also supports connecting to other Android devices via wireless debugging and OTG mode, making its usage highly flexible.
In practice, however, aside from professional developers, even Android enthusiasts with some technical background are unlikely to remember so many ADB commands. To address this, aShell You has many of them built in. Tap the button to the right of the input box to view and search preset commands. For each command, the app thoughtfully adds attribute tags, such as whether it accesses a file or a directory, or whether it is a system command.
For your own frequently used custom commands, in addition to adding them to the preset list, you can also bookmark them for easier access. If you use many custom commands, it’s recommended to save them as bookmarks to avoid mixing them up with the built-in presets.
ADB command functionality is already included in many all-in-one system utility apps, but few do it with this level of refinement and attention to detail. If you’d like to try this convenient tool, aShell You is available for free on GitHub and F-Droid.
Lexibird: An All-Around AI-Powered Translation Tool
AI-based applications have been booming over the past year or two, and translation tools are one of the most important categories among them. However, few products truly integrate language translation with language learning. The tool introduced today, called Lexibird, is more like a comprehensive AI toolkit designed to help users learn languages more effectively.
In addition to the text translation features commonly found in translation apps, Lexibird also bundles a series of tools useful for language learning, including a built-in dictionary, text-to-speech, writing tools, and transcription. Take text and image translation as examples: compared with web-based translators that we often use, I personally encounter far more situations where I need to translate local documents.
When I want to check the meaning of a paragraph of English or another language in a document, I used to open Google Translate and copy and paste the text. Lexibird solves this pain point: I only need to copy the text into the app’s translation dialog box to get the result instantly. What’s more, I can choose different types of large language models to make the translation sound more “human.”
Beyond text translation, Lexibird also supports image OCR translation and voice translation. For images, you simply upload the file, and Lexibird will first perform OCR recognition and then translate it into your target language. As for audio translation, you can upload an audio clip and have it converted and translated into the desired language, which is especially useful for work scenarios that involve speech transcription.
Another highly practical feature of Lexibird is its writing tools, which are particularly useful for people working in publishing, journalism, or self-media. For example, if I need to adjust the style of a piece of English news text, I can open the “Rewrite” option, paste the text directly, and wait for the model to analyze and rewrite it. For more fragmented content such as meeting minutes or communication records, the “Summarize” function can be used to extract key points. And for content you’ve already written, you can use “Proofread” to check and correct it.
The final feature, in my view, is even more helpful for language learning: the dictionary. If you encounter a word you don’t understand at all during your studies, you can look it up to see its meaning, example sentences, and translations, helping you quickly grasp how the word is used.
Overall, Lexibird covers most of the scenarios involving language translation and learning that we encounter in both work and daily life. With AI, tasks that used to be time-consuming—such as transcription, organization, and proofreading—become much easier. In terms of pricing, the free basic version provides 100,000 AI credits per month along with full access to features like text translation and transcription. You can also purchase a Pro subscription to get more credits or use more advanced AI models. You can download the desktop client from the official website or use the web version directly, and you can also pair it with a browser extension for better translation while browsing the web.
Halo: Can AI Really Help You Build Better Habits?
Platform: iOS
Keywords: habit building
As one of the evergreen categories on the App Store, habit-tracking apps should be familiar to almost everyone. SSPAI has also published detailed reviews and comparisons of various popular habit apps in previous articles. With the arrival of the AI era, habit-building apps have naturally entered a phase of “upgrading and transformation.” Halo is one such app. It comes from the developer of the well-known read-it-later tool Matter and adds an AI assistant on top of traditional habit-tracking features, aiming to help users cultivate habits more effectively.
When you open Halo for the first time, the app guides you to create a habit you want to build. Most habits can automatically pull data from system APIs such as HealthKit and Screen Time to complete check-ins automatically. This is nothing new—veteran habit apps like Streaks have been doing this for years. Tap the plus button in the top-right corner to add new habits. Halo provides many built-in templates you can use directly, and of course you can also create custom habits based on your needs. For each habit, you can freely adjust the check-in frequency, reminder method, and detailed goals.
After completing a habit for three consecutive days via the app or home screen widgets, you unlock Halo’s Progress and Toolkit sections. As the names suggest, the Progress page shows each habit’s development score and duration, while the Toolkit page offers a collection of video tutorials about habit building. Halo categorizes and organizes these YouTube videos to help you acquire relevant knowledge and build habits more effectively.
Naturally, the most eye-catching feature is Halo’s AI Coach. However, Halo’s AI functionality is not built directly into the app. Instead, it provides an entry point to the Messages app, where all interactions take place. Before starting a conversation with the Coach, you need to add Halo as a contact, verify your phone number, and then communicate with the Coach via SMS. The AI Coach provides personalized advice based on your habit-building progress.
If you have a strong need for a habit-tracking app, Halo is worth trying. Its basic features are free to use, and its overall design and interaction are quite polished. As for the AI Coach, however, it feels more like a gimmick—the annual subscription price of RMB 398 is hard to justify.
WidgetWall: A “Wall” of Desktop Widgets for macOS
Platform: macOS
Keywords: widgets
@Snow: Since macOS Sonoma introduced support for desktop widgets, the macOS desktop—long overlooked—has returned to users’ attention. Unlike the “Swiss Army knife” widget apps on iOS, which let standalone widgets of different sizes blend into the native home screen layout and truly shine, similar apps on macOS—such as DeskWidgets or Widgetify—mostly adopt the approach of creating a new “desktop overlay.” It’s like a transparent glass wall attached to your desktop, on which you can freely arrange built-in widgets. WidgetWall, which we introduced previously, has recently added multiple style options and new widget choices in its latest update, so it’s worth revisiting.
The current version of WidgetWall includes 29 functional widgets. Common tools such as clocks, timers, photos, calculators, and music controls are all there, along with some widgets designed to boost your work (or “slacking-off”) efficiency. The app provides a web browser widget that supports layouts up to 2×4 and 4×2, adapting well to both landscape and portrait screens. You can “pin” any webpage to your desktop for quick viewing. On top of that, it comes pre-optimized for TikTok and Pinterest—after logging in, you can start scrolling short videos and image feeds right away.
WidgetWall is also one of the few apps that supports video widgets. You can play specified video files or videos from the system photo library, and it even supports playing Live Photos as videos on the desktop. The video widget offers basic playback controls such as a progress bar and play/pause, as well as adjustable size and volume, making it a very practical video preview tool.
Pomodoro timers and white noise have both been proven to help improve focus. If you don’t yet have tools you’re satisfied with in these areas, you can make do with the built-in widgets here.
Beyond productivity, WidgetWall also includes some widgets with a strong sense of ritual. One example is the nostalgia-filled iPod widget. Functionally it’s just a simple music player, but its four size options mimic several generations of classic products like the iPod shuffle and iPod nano. Long-buried memories start flooding back.
If you’ve just finished writing a piece of code or completing a proposal, you can enable the built-in confetti widget and treat yourself to an explosion of colorful paper as a burst of positive energy. And if confetti alone isn’t enough to brighten your dull workday, you can also use the built-in sticker tool to further decorate your desktop and add a bit more fun.
WidgetWall’s desktop overlay is adapted to the grid size of macOS native widgets, so you can either use it on its own with freely arranged widgets or combine its grid with native widgets. Unfortunately, WidgetWall does not currently support automatically changing colors based on desktop activity. Even though it supports light and dark modes and offers four levels of adjustable widget materials, there is still some visual mismatch when used alongside native widgets.
The app includes an in-app “Pro” upgrade that unlocks all widgets and stickers. You can subscribe for RMB 15 per month or RMB 148 per year, or make a one-time purchase of RMB 298 for lifetime access. If you are a Setapp subscriber, you can use the Pro version for free.
Dozens of iOS apps have been found by security agencies to contain serious vulnerabilities that could lead to the leakage of sensitive information such as chat records, user IDs, and contact details. Most of these apps are related to AI technologies or categories. Source
A cooperation agreement worth USD 800 million between Epic Games and Google has come under court review. The deal involves licensing of Unreal Engine technology and joint promotion efforts. The court’s focus is on whether the transaction could affect Epic’s stance on the Android app ecosystem and its antitrust claims. Source
High-speed train customer service and experts respond to “excessive CO₂ levels on high-speed trains”
cURL to terminate its bug bounty program after being overwhelmed by AI-generated reports
China Unicom launches “Apple Fan” mobile plan
Microsoft confirms handing over BitLocker encryption keys to the FBI
Meta layoffs spark concerns over a “VR winter”
First-generation Pokémon becomes a popular method for testing AI models
Rumors You Can Just Glance At
High-speed rail customer service and experts respond to “excessive CO₂ levels on high-speed trains”
According to reports by The Beijing News and other outlets, a video showing carbon dioxide concentration measurements taken inside a second-class carriage on a high-speed train has recently sparked heated discussion online. The footage shows that before passengers boarded, the CO₂ concentration inside the carriage fluctuated at around 880 ppm (parts per million, a unit of concentration). As passengers were boarding and taking their seats, the CO₂ level had already begun to rise. After the train had been running for some time, the CO₂ concentration inside the carriage exceeded 2,000 ppm, at which point most of the passengers shown in the video had already fallen asleep. The report noted that under China’s current Indoor Air Quality Standard, the standard limit for indoor carbon dioxide (CO₂) concentration is ≤ 1,000 ppm (i.e., ≤ 0.1%). The CO₂ levels shown in the video were therefore far above this limit.
In response, different 12306 customer service representatives gave varying explanations. One said that if passengers feel unwell during the journey, they can “use oxygen on their own.” Another claimed that “there is currently no such standard.” Yet another said that air is replaced in real time during operation, and system checks are carried out both before departure and after arrival, so “there is definitely no problem with the air,” adding that the higher CO₂ concentration might be due to increased passenger density.
Later, Xinhua News Agency interviewed Tao Guidong, an expert on high-speed trainsets from CRRC Qingdao Sifang Co., Ltd., a manufacturer of multiple-unit trains. Tao explained that China’s standard for controlling indoor CO₂ concentration in passenger trains follows TB/T 3493-2017, Railway Vehicle Air Conditioning – Pressure Protection Device, under which the standard limit under normal operating conditions is no more than 2,500 ppm. European standards, he added, allow up to 5,000 ppm. The civil building standards used by some netizens for comparison do not apply to passenger trains.
Tao also said that air exchange inside high-speed train carriages is entirely handled by the air-conditioning and ventilation system. When operating in non-tunnel sections, continuous air exchange can be achieved, and CO₂ concentrations generally do not exceed 1,500 ppm. When passing through continuous groups of tunnels, pressure protection measures are taken to close off channels to outside air, during which CO₂ levels inside the carriage may rise for a short period of time.
cURL to terminate its bug bounty program after being overwhelmed by AI-generated reports
On January 22, the globally well-known open-source data transfer tool cURL announced that it will officially terminate its bug bounty program at the end of this month. The updated security.txtfile on cURL’s website (a file commonly used in the industry to specify channels for submitting security reports) now states: “We do not offer any reward or other form of compensation for reported issues, but we will clearly express our thanks and recognition in the documentation confirming the issue. If you waste our time with garbage reports, we will ban you and ridicule you in public.”
In response, project founder and lead developer Daniel Stenberg said that the team has recently been inundated with large numbers of low-quality vulnerability reports generated by artificial intelligence, leaving the maintainers overwhelmed. To safeguard the “mental health” of team members and ensure the project can continue to operate normally, the official team had no choice but to make this decision.
For a long time, the cURL team, like many software developers, has used cash rewards to encourage external researchers to submit high-severity security vulnerabilities. However, since May last year, Stenberg has been warning that AI-generated “slop” has been surging. Large numbers of speculators blindly use large language models to generate reports, which are filled with hallucinations, including fabricated CVE IDs, nonexistent function signatures, and even code snippets that cannot be compiled at all. This invalid information forces maintainers to spend huge amounts of time on verification and investigation.
Despite its massive user base, cURL is in fact a small open-source project run by only a handful of core maintainers. Stenberg said the team has no ability to change the reality of people abusing AI tools, and can only respond by cutting off the source of incentives. It is worth noting that the project is not completely opposed to AI-assisted security research. In September last year, Stenberg publicly praised a researcher who used an AI-based code analysis tool (ZeroPath) to discover and help fix 22 real vulnerabilities. He stressed that the core problem lies not with the tool itself, but with speculative behavior by people who neither understand code nor perform even basic checks on AI outputs before submitting them directly.
China Unicom launches “Apple Fan” mobile plan
On January 23, China Unicom announced the launch of the “Apple Fan · King Card” plan. This product is a derivative version of the “King Card 3.0” plan, with its core selling point being the integration of AppleCare Services benefits into the mobile tariff. The discounted monthly fee for the first year is RMB 49. After activating the SIM card and making an initial top-up of RMB 50, users can receive up to 24 months of benefits, including unlimited accidental front-screen repairs, free battery replacement (when capacity drops below 80%), and Apple officially certified technical support.
In terms of basic mobile services, the plan includes 30 GB of dedicated data, 20 GB of general data, and an additional 30 GB of general data provided as a promotional bonus each month. China Unicom also highlights features such as eSIM support, “one number for two devices” for smartwatches and phones, and international roaming packages priced at RMB 9.9 per day.
According to the terms of the benefits, users must claim AppleCare Services within 90 days after activating the SIM card. The bound iPhone must be a new mainland China model activated within the past 30 days, and its IMEI must be verified. The system checks eligibility on a monthly basis; if the actual monthly fee paid for the plan is below RMB 59, or if the account is not in normal active status, the AppleCare benefits for that month will automatically become invalid. If users choose not to claim AppleCare Services, they may instead opt for internet subscription benefits such as Tencent Video, iQIYI, or QQ Music.
It is worth noting that, unlike AppleCare+ sold through Apple’s official channels, AppleCare Services is a protection plan provided by Apple via third-party suppliers, and the specific scope of coverage depends on the supplier. For example, the AppleCare Services offered through the “Apple Fan · King Card” plan does not include repairs for damage other than screen breakage and battery issues. In the past, Apple has also sold AppleCare Services through authorized retailers that covered only screen repair.
Microsoft confirms handing over BitLocker encryption keys to the FBI
According to Forbes, Microsoft recently confirmed that it had handed over BitLocker recovery keys to the U.S. Federal Bureau of Investigation (FBI) in compliance with a legal order, helping unlock three laptops involved in a suspected fraud case. This is the first known case in which Microsoft has directly provided such encryption keys to law enforcement. BitLocker is a built-in disk encryption tool in Windows designed to protect hard-drive data from unauthorized access.
In response, a Microsoft spokesperson said that the company receives about 20 such key-access requests each year, and stressed that as long as there is a valid legal search warrant and Microsoft has access to the keys, it will cooperate. In this case, because the users had chosen to back up their keys to Microsoft’s servers, Microsoft technically had the ability to assist law enforcement. The spokesperson also pointed out that while cloud-based recovery provides convenience, it does carry the risk of external access. If users choose to store their keys locally on hardware such as a USB drive rather than in the cloud, Microsoft would be unable to provide assistance.
The incident highlights differences between Microsoft’s privacy architecture and that of other tech giants. Companies such as Apple and Meta adopt stricter “end-to-end” encryption strategies, under which even the service provider itself cannot access users’ decryption keys. For example, Apple refused to help the FBI unlock the iPhone of a suspect in the 2016 San Bernardino shooting, and has implemented technical designs in FileVault and its cloud services that prevent law enforcement from requesting such keys. By contrast, Microsoft’s current default settings leave cloud-backed-up keys in a state that the company itself can read.
Meta layoffs spark concerns over a “VR winter”
According to CNBC, Meta has recently carried out large-scale layoffs in its Reality Labs division, which is responsible for the company’s metaverse business. About 10% of employees were affected, involving roughly 1,000 people. The layoffs mainly targeted VR teams working on the Quest headsets and the virtual social platform Horizon Worlds, and some internal studios were also shut down. The move is widely seen as a sign that Meta’s strategic focus is shifting from virtual reality (VR) to AI and wearable devices such as its Ray-Ban co-branded smart glasses, triggering industry concerns that a “VR winter” may be approaching.
Market data supports this shift. According to the latest IDC report, while overall shipments of extended reality (XR) devices are expected to grow by 41.6% in 2025, the growth will be driven mainly by AI smart glasses rather than traditional VR headsets. IDC forecasts that shipments of VR and mixed reality headsets will plunge by 42.8% in 2025 to 3.9 million units. By contrast, shipments of AI glasses are expected to more than double. Analysts note that market feedback shows VR headsets remain largely confined to a niche group of gamers, and that mainstream consumers have little interest in wearing bulky headsets for long periods.
In response to the criticism, Meta’s chief technology officer said the company is not abandoning VR but adjusting the scale of its investment. He acknowledged that the VR market is growing more slowly than expected, making it necessary to rebalance spending. Oculus co-founder Palmer Luckey also said that while the layoffs are regrettable, they may benefit the industry’s long-term health. Since the end of 2020, Reality Labs has accumulated losses of more than USD 70 billion.
The strategic shift has dealt a blow to third-party developers that rely on Meta’s ecosystem. The head of Owlchemy Labs, a VR studio under Google, likened the current VR market to the Atari era before the U.S. video game crash of 1983, arguing that VR is undergoing a necessary market correction. According to IDC, Apple’s China-based manufacturing partner Luxshare Precision has stopped producing the Vision Pro headset, indicating that even high-end VR devices are facing weak demand.
First-generation Pokémon becomes a popular method for testing AI models
According to The Wall Street Journal, AI labs in Silicon Valley are widely adopting an unconventional benchmarking method: using Nintendo’s classic 1990s game Pokémon to evaluate models’ reasoning and decision-making abilities. Companies including Anthropic, OpenAI, and Google have been having their AI models play Pokémon Red/Blue to visually demonstrate progress in long-term planning and complex task handling.
The trend was initiated by David Hershey, head of applied AI at Anthropic, who launched a “Claude Plays Pokémon” Twitch livestream in February last year, which was soon followed by similar efforts from OpenAI and Google. At present, developer-built projects such as “GPT Plays Pokémon” and “Gemini Plays Pokémon,” with the help of specialized auxiliary frameworks, have successfully completed the original games and are moving on to their sequels; meanwhile, Anthropic’s latest model, Claude Opus 4.5, is currently attempting a full playthrough live on stream.
Researchers point out that, unlike traditional question-and-answer benchmarks, Pokémon requires sustained reasoning and decision-making over long periods. Models must navigate mazes, assemble teams, and devise battle strategies based on opponents’ attributes. This nonlinear and complex environment is less constrained than traditional board games and more accurately simulates the challenges AI systems face when handling long-horizon, multi-step tasks in the real world, making it an ideal testbed for evaluating AI agent capabilities.
Beyond evaluating the models themselves, the project has also become a hands-on exercise in developing AI support software, known in the industry as a “harness.” To enable smooth gameplay, developers need to build external tools such as memory systems to help models retain key information. David Hershey said that he has already applied the experience gained from building Claude’s in-game memory module directly to advising enterprise clients on deploying complex commercial AI systems.
Rumors You Can Just Glance At
Recently, JD.com and Tmall announced that they will launch a special promotion for the iPhone Air at 20:00 on January 25. After stacking a RMB 2,000 platform coupon and a RMB 500 national subsidy available in some regions, the 256 GB model will drop to RMB 5,499 on both platforms; JD.com will also offer an additional RMB 400 subsidy for trade-ins. The offer is limited by promotional stock. Previously, supply chain sources and media reports suggested that the iPhone Air has underperformed expectations. For example, blogger @数码闲聊站 recently claimed that cumulative activations of the iPhone Air currently stand at fewer than 200,000 units. By comparison, all models in the iPhone 17 lineup have each surpassed 4 million activations over the same period, with the iPhone 17 Pro Max exceeding 8.26 million units. In addition, Apple launched its Chinese New Year limited-time promotion on January 24, mainly covering the previous-generation iPhone 16 series, iPad, and Mac product lines. Among them, MacBook Pro models with M4/M5 chips receive discounts of up to RMB 1,000, while the iPhone 16 series is discounted by up to RMB 300.
According to Bloomberg, citing people familiar with the matter, the owner of German camera maker Leica Camera AG is considering selling a controlling stake in the company, in a potential deal that could value it at around EUR 1 billion. Potential bidders include HSG (formerly Sequoia China) and private equity firm Altor Equity Partners. Leica’s current major shareholders are the family of Austrian billionaire Andreas Kaufmann and Blackstone Inc. Sources said the Kaufmann family may choose to reinvest after a deal is completed, retaining part of its stake. Discussions are still at an early stage, and there is no guarantee a transaction will be finalized. Blackstone, HSG, and Leica all declined to comment. Although Leica faced severe financial difficulties in 2005, its performance has been steady in recent years after being taken private by the Kaufmann family in 2012. In the fiscal year ending March 2025, Leica’s revenue rose 7.6% to about EUR 596 million, mainly driven by growth in its core photography equipment and mobile imaging businesses.
Mark Gurman claims that the first batch of features in the new version of Siri is expected to officially launch with iOS 26.4, scheduled for release in March or April. Although internally codenamed Apple Foundation Models v10, its core computing power is in fact provided by Gemini and will initially run on Apple’s private cloud servers. Later, at the WWDC developer conference in the summer, Apple will unveil a brand-new Siri architecture codenamed “Campos,” to debut with iOS 27. This version will feature deep multi-turn conversations and contextual understanding similar to ChatGPT, and plans to directly use Google’s cloud TPU infrastructure to improve response speeds.
Gurman also claims that Apple plans to intensively refresh the MacBook Pro, MacBook Air, and Mac Studio product lines in the first half of the year, and may release a long-unupdated Studio Display. A low-cost MacBook powered by an A-series chip is under development, targeting the education and entry-level markets. In addition, development of the next-generation M6 chip is reportedly ahead of schedule and may debut sooner than expected.
According to Android Authority, text strings such as “unverified direct install,” “unable to verify app developer,” and “online verification required” have appeared in the latest version of Google Play’s code. This suggests that when users attempt to install APK files, installation will still be possible, but they will have to go through a cumbersome confirmation process, with mandatory pop-up warnings about risks. Matthew Forsyth, director of developer experience and product management for Google Play, confirmed that a “high-friction” sideloading process will be introduced into Android, with the goal of “layered responsibility.” Previously, Google announced a controversial policy last summer that initially planned, starting in 2026, to require all Android developers—regardless of whether they distribute via the Play Store—to register and verify their identities with Google. After strong backlash, Google compromised, promising to retain manual installation options for “advanced users.” Based on the latest confirmation, this feature has entered the preparation stage, and the barrier for ordinary users to obtain third-party apps will effectively be raised. The aforementioned verification plan is expected to begin pilot implementation this September in Brazil, Indonesia, Singapore, and Thailand, before gradually rolling out globally.
According to Sherwood, as of early January 2026, the number of registered .ai domain names worldwide has officially surpassed one million, bringing substantial fiscal windfalls to Anguilla, a British Overseas Territory in the Caribbean. It is estimated that the region generated about USD 70 million from domain-related business alone over the past year. Anguilla has a population of only around 15,000. In the mid-1990s, the Internet Corporation for Assigned Names and Numbers (ICANN) assigned .ai as the country-code top-level domain (ccTLD) for the territory. Currently, the standard fee to register a .ai domain is typically USD 140 for two years, with a renewal rate as high as 90%, providing the local government with stable cash flow. In addition, expired-domain auctions held by registrar Namecheap have been highly profitable. Data shows that the domain you.ai sold for USD 700,000 last September; in just the past week, 31 expired domains have been auctioned for a total of about USD 1.2 million.
On January 24, Tencent QQ announced via its official Weibo account that it will bring back the QQ Show feature. The new version of QQ Show can be generated by AI based on uploaded images, replacing avatars on personal profile pages and in chat interfaces, and can generate animations together with other users’ avatars in group chats. The company also teased the return of more classic features. QQ Show first appeared in QQ2000C as a virtual avatar design system. Virtual clothing, scenes, and characters in the QQ Show store could be used to decorate users’ avatars displayed in the PC version of QQ. Tencent once launched a “Red Diamond” subscription service priced at RMB 10 per month, allowing paying users to use all QQ Show items for free. In 2020, the Red Diamond service stopped accepting renewals and top-ups. In November 2021, most QQ Show display sections were removed from the PC version of QQ. At the time, “QQ Show discontinued” trended on Weibo.
