Meta has sued the people and groups behind three scam operations that used images and deepfakes of celebrities to lure users to scam websites. According to the company, the three entities were based in China and Brazil and targeted people in the US, Japan and other countries. The ads promoted fraudulent investment schemes and fake health products.
Meta said that it had filed lawsuits against several people in Brazil who promoted fake or unapproved healthcare products and online courses promoting them. The company also sued a China-based entity it says used ads featuring celebrities "as part of a larger fraud scheme that lured people into joining so-called investment groups." The company didn't provide details on how many ads these groups had run on Facebook, how many social media users had seen or interacted with the ads or how long the scammers had been operating on the platform.
So-called "celeb bait" ads have been a long-running issue for the company. Engadget has previously documented celeb bait scams on Facebook, including ones that frequently use Elon Musk and Fox News personalities to hawk fake cures for diabetes. The Oversight Board has also criticized the company for not doing enough to combat such scams. In its update, Meta says that "because scam ads are designed to look real, they’re not always easy to detect." The company also noted that it has now enrolled "more than 500,000" celebrities and public figures into its facial recognition system that's meant to automatically detect scam ads using the faces of famous people.
Meta's handling of scammy advertisers has come under increased scrutiny in recent months after Reutersreported that researchers at the company at one point estimated that as much as 10 percent of its ad revenue could be coming from scams. The fact that Meta has made billions of dollars from problematic advertisers has also caused the company to be slow to take action against repeat offenders.
In addition to the groups behind the celeb bait ads, Meta says that it's upgraded its ability to detect scam ads that use cloaking, which has at times hindered its internal review systems. The company also sued a Vietnam-based advertiser it says used scam ads to hawk "deeply discounted items from well-known brands," including Longchamp.
Meta also took legal action against eight former "Meta Business Partners," who promoted services that would "un-ban" or other "account restoration services." The company says it will "consider taking additional legal action, including litigation, if they don’t comply" with cease and desist orders.
This article originally appeared on Engadget at https://www.engadget.com/social-media/meta-sues-advertisers-in-brazil-and-china-over-celeb-bait-scams-190000268.html?src=rss
How much would you pay for a good night's sleep? This is a question I've asked myself repeatedly over the last few weeks as I've been testing the Dreamie, a $250 alarm clock and "bedside companion" that I couldn't stop thinking about after I first encountered it at CES.
Ambient's Dreamie offers many of the conveniences of a smartphone-connected device — highly customizable alarm schedules, a library of soundscapes and noise masks, Bluetooth so you can connect earbuds and podcasts (soon). But it is phone-free every step of the way, with all controls and features built-in so you don't end up getting sucked into a doomscroll while you're trying to wind down. It also has a light ring for ambient lighting modes and sunrise wakeups. This spring, it's expected to start providing sleep insights as well for users who opt-in, using its microphone and motion sensors to get a reading on their nightly habits.
All of that's meant to work together to, according to the website, "help you sleep better and break free from your phone," a goal I was eager to explore. This may be one of the least unique problems to have as an adult in today's world, but sleep has become a really complicated thing for me.
Falling asleep is hard because my brain is always racing, my quality of the sleep is trash and waking up every day feels like an act of torture. It's gotten so bad that at some point in the last couple of years, I started using three alarms to make sure I get out of bed in time for work: a dedicated sunrise alarm clock, my smartwatch and my phone as the final, 11th hour save in case the other two methods don't do the trick. As you might imagine, my partner, who is forced to also endure this horrid morning ritual, hates it.
So if there's a device that can help fix this mess, I'm open to it. And after some time with the Dreamie, I think I've found a promising contender.
Getting into a sleep routine
There's no companion app with the Dreamie and no subscription service you need to sign up for, which feels like a breath of fresh air in 2026. (I'm so tired of subscriptions, free us from this hell!) Your one-time purchase gives you access to everything it offers now and the updates that are in the pipeline.
After taking it out of the box and plugging it in, you'll have to connect to your home Wi-Fi. Then, the Dreamie presents you with a tutorial to walk you through navigating its menus and physical controls. There's a touch strip on the top of the device to turn on the lamp and adjust its brightness, as well as the brightness of any ambient color "scene" that's active. By dragging the dot at the center of the lamp screen, you can throw the light in any particular direction. Volume is adjusted by turning the dial that's around the clockface. To access the menu for alarms and other settings, swipe up. To cycle through the different content modes — ambient, wind down and noise mask — just swipe down from the top of the screen. Easy peasy.
Setting up your actual Sleep Routine takes a little more time and intention. A Dreamie Sleep Routine consists of multiple steps, which you can use all, some or none of for your custom routine. Those include the Bedtime Cue, which lets you know it's the time to start getting ready for bed (you designate this time); the Wind Down, or the sounds you'll fall asleep to; and the Noise Mask, the sounds that keep you asleep. If you wake up in the middle of the night, there's a Back To Sleep option too.
You can choose different sounds from Dreamie's library for each category. Some options come with ambient lighting effects, too. There's a decent selection of soundscapes, from the dramatic Aurora Borealis and the sounds of storms and rivers to different "colors" of noise.
Some noise masks, like Green Noise, coming with lighting effects.
Cheyenne MacDonald for Engadget
The quality of the Dreamie's sound is what initially sold me during my demo at CES, and it holds up in daily use. The Dreamie has a 50 millimeter speaker inside, and the 360-degree grille on the bottom of the device makes it so the sound seems to come from everywhere. (My cats were extremely confused when I first turned it on). It really fills a room, and you don't have to crank it up to achieve that. When Bedtime Cue comes on, I typically turn it down to about 25, and then raise it back up to 45 when I flip it to Wind Down mode. I've never once set it higher than 50, and the alarm in the morning has still been loud enough to wake me up.
After taking a few days to tweak my choices and figure out what I like best, I've settled into a really nice routine: Aurora Borealis as the Bedtime Cue, an hour of Forest Wind as my Wind Down and a Noise Mask of Brown Noise to play throughout the night. I love how easy it is to set the nighttime routine in motion once it's established. When I hear the Aurora Borealis come on, I start making my preparations for bed. Brush teeth, take meds, lights out and, crucially (I'm trying really hard to be disciplined, here), my phone goes face-down on the nightstand until morning. If I want to stay up late that night and ignore the Bedtime Cue, I can just hit the little stop button on the display. But once I'm ready to actually try to fall asleep, all I need to do is swipe down on the display to initiate the Wind Down, and Forest Wind will start playing.
I have my Wind Down set for one hour, after which the Noise Mask begins. And man, that Forest Wind knocks me out. So far, I haven't found myself still up and staring at the ceiling by the time Brown Noise comes on. I've only been able to confirm that it is indeed working and switching to the Noise Mask because my cats regularly wake me up in the middle of the night, and it's been on each time that's happened. But aside from those instances where my head is being used as a springboard by the creatures that share my home, I've been sleeping pretty well through the night.
To minimize distractions when you're trying to sleep, the Dreamie's display will dim in response to the surrounding darkness. There's also a Redshift toggle to make the nighttime display easier on the eyes, a Dark Mode with a simplified appearance and the option to have the display turn off completely when you've been inactive for a while. I set the Dreamie on my nightstand close to where my face is at night, and I haven't had any problems with light from the display keeping me up.
Waking up with Dreamie
In the morning, the light begins to come on 20 minutes before I want to be awake, followed by the gradually increasing sound of the alarm. There are only a handful of alarm sounds at the moment, but the options are all fine. There are no jarring, grating alarms here — even the bird calls option sounds rich and natural, rather than the too-shrill, piercing recordings I've grown used to avoiding on other alarm clocks and sound machines.
You can set multiple alarms with different bedtimes and wakeup times, which is really handy if your schedule is all over the place or you want to allow yourself to sleep in more on certain days. My only real complaint so far is that the sunrise feature isn't quite as strong as I want it to be. The Dreamie's sunrise goes from a warm glow to a bright blue-white, but it never gets big enough to wash over me in the way I expect a sunrise alarm to. Having the light on is helpful for orienting yourself when you're groggy and half-asleep, but it doesn't feel like it's having much effect on my actual wakeup process.
Dreamie next to a Philips Wake-Up Light.
Cheyenne MacDonald for Engadget
Part of the problem may be that none of the light is really directed forward and at the sleeper's face. Even the Dreamie's lamp mode at maximum brightness seems to have more reach than the sunrise feature. (And a note on the lamp, while it's decently bright, it's still a bit too dim for reading in bed unless I'm huddled up to it.)
Still, I've been sleeping well enough that I've been waking up alright most days even without being bathed in artificial sunlight. Don't get me wrong, I'm still hitting snooze a few times before dragging myself out of bed, but there's been a noticeable improvement in both the quality of my sleep and how miserable I feel come morning. I'm even down to using just two alarms: the Dreamie as my primary alarm, which is getting me up on its own for the most part, and my watch as a backup. At this point, I'm kind of attached to this thing.
The Dreamie is refreshingly compact, too. It takes up significantly less real estate on my nightstand than the Philips Wake-Up Light I've been using forever, or something like a Hatch Restore. The smaller footprint is something I appreciate as a person always battling cluttered surfaces. That also makes it better for travel. Since podcasts and sleep insights aren't available yet, I haven't been able to test those out, but they're non-critical features for me. The company has shared an estimated timeline of Q1-Q2 for these features to arrive, with podcasts likely coming first. They'll be nice to have, podcasts especially, but the Dreamie is more than able to do its main job of creating an environment that supports better sleep without those things.
Wrap-up
All of this brings me back to the question that's been haunting me since discovering the Dreamie: Is it ridiculous to spend $250 on an alarm clock/noise machine? At a different time in my life, I would have said yes without hesitation. But the current version of me, who knows what it's like to move through each day like a zombie because I'm sleeping so terribly, would begrudgingly disagree. As I pack up this review unit to ship it back, I'll also be putting in an order for my own so I can keep my cherished new sleep routine going.
This article originally appeared on Engadget at https://www.engadget.com/home/ambient-dreamie-bedside-companion-review-the-best-sleep-ive-had-in-years-184019430.html?src=rss
The DualSense arrived with something to say. Adaptive triggers, nuanced haptics, a tactile language that made games feel physically present in your hands — it raised the bar in ways the industry hadn’t anticipated. For a while, nothing else came close. That window has closed. The third-party market in 2025 is no longer playing catch-up. It’s producing controllers with drift-proof magnetic sensors, modular physical architectures, trigger calibration measured in millimeters, and battery lives that nearly triple what Sony ships as standard. The gap has flipped.
The Goo-inspired concept controller at the top of this page is a glimpse at where peripheral design is reaching — fluid, sculptural, unresolved in the best way. It hasn’t shipped. What’s below has. Every controller in this roundup is available now, purpose-built around a specific performance argument, and doing at least one thing the DualSense doesn’t. If you’ve stuck with the stock pad out of habit, these five make a clear case for reconsidering that.
1. Razer Raiju V3 Pro
Razer’s pitch with the Raiju V3 Pro is precise: take the sensor thinking behind their best gaming mice and transplant it into a PlayStation-compatible controller. The result is Tunnel Magnetoresistance thumbsticks — TMR —, and as of 2025, no other PS5 controller ships with them. Where the Hall Effect uses magnetic fields to read position, TMR uses weak electromagnetic waves to detect even finer movement with greater resolution. Drift is resolved at a hardware level, not managed in software. Hall Effect triggers cover the other high-wear surface, meaning every primary input on this controller is engineered against degradation from the start. At 258 grams, it sits lighter than the DualSense Edge without feeling hollow, and the wider grip reduces hand strain across longer sessions.
Six extra inputs are distributed across the frame — four removable back buttons in the rubberized handles and two claw-grip bumpers flanking the triggers — all fully remappable to whatever a specific game demands. Razer’s HyperSpeed 2.4GHz wireless holds latency tight, with a polling rate that climbs to 2,000Hz on PC, a number Sony’s controllers don’t approach. Battery life is rated at 36 hours, nearly triple the DualSense standard. It’s officially licensed for PlayStation 5, requires no adapters, and connects as a native peripheral. For competitive players who want every hardware advantage consolidated in one place, the Raiju V3 Pro is currently the ceiling.
What We Like
TMR thumbsticks are unique to this controller in the PS5 space, resolving drift at a sensor level that Hall Effect doesn’t reach.
A 36-hour battery life and 2,000Hz PC polling rate are specifications Sony’s lineup has no current answer to.
What We Dislike
Haptic feedback and adaptive triggers are absent — a real trade-off for anyone whose gaming skews toward immersive, story-led experiences.
The symmetrical thumbstick layout is a deliberate competitive choice that won’t feel native to players raised on PlayStation’s standard asymmetric positioning.
2. Nacon Revolution 5 Pro
The Revolution 5 Pro starts from a principle the DualSense never acted on: if magnetic sensor technology stops drift, why limit it to the thumbsticks? Nacon applies the Hall Effect to the triggers as well, covering every primary contact surface in a single design. No stick drift, no trigger wear, no gradually worsening feel over months of use. The asymmetric layout mirrors the DualSense’s familiar posture closely enough that the transition is immediate, and the premium materials wrapped around the modular frame feel considered rather than compensatory. It’s officially licensed for PlayStation 5 and built around the ergonomics of long sessions rather than short competitive bursts.
Customization is both deep and accessible. Four profiles can be switched directly on the controller without opening a companion app, though the app itself offers trigger sensitivity curves, deadzone tuning, and full button remapping with genuine precision. Interchangeable thumbstick sizes and adjustable internal weights let players calibrate the physical feel to their own preference. A standout feature that no other controller on this list includes is built-in Bluetooth audio output, letting players pair headphones directly to the controller rather than routing through the console. The Revolution 5 Pro was also designed around a reduced carbon footprint — a thoughtful distinction for a product category that rarely acknowledges it.
What We Like
Hall Effect across both sticks and triggers makes this one of the most mechanically durable pro controllers on the market right now.
Built-in Bluetooth audio pairing is a friction-reducing feature that no Sony controller — at any price — currently provides.
What We Dislike
Haptic feedback and vibration don’t function during PS5 gameplay, which strips out a meaningful portion of the DualSense’s native experience.
The profile and customization system has a learning curve that requires time to work through before its full value becomes accessible.
3. SCUF Reflex Pro
SCUF has spent years earning credibility with competitive console players, and the Reflex Pro is the most technically resolved version of that commitment. The 2025 lineup integrated Hall Effect anti-drift thumbsticks as standard hardware, closing the mechanical gap that had followed the Reflex series across previous generations. Wireless performance is clean, adaptive triggers function as expected on PS5, and vibration rumble stays intact — a combination that most third-party alternatives compromise somewhere along the way. The physical form follows the DualSense’s geometry closely enough that picking it up for the first time feels instinctive. It’s built for precision longevity first, familiarity second, and it delivers both.
The rear paddle system is where the Reflex Pro makes its case most directly. Four fully assignable paddles run along the underside of the controller, each mappable to any function that would otherwise require lifting a thumb from the sticks — jump, reload, slide, crouch, anything the game demands. Your aim stays unbroken at the exact moments it matters. Sony’s DualSense Edge, the first-party pro option, ships with two back buttons at a higher price. The Reflex Pro ships with four. SCUF also offers a Build Your Own path that opens TMR thumbstick selection at the point of purchase, giving players the option to match or exceed the Raiju V3 Pro’s sensor performance inside a controller that keeps full haptic and adaptive trigger compatibility.
What We Like
Four fully assignable rear paddles outperform the DualSense Edge’s two-button setup — more inputs, better placement, and a lower price.
Hall Effect thumbsticks are now standard across the line, making long-term stick accuracy a structural strength rather than a premium option.
What We Dislike
At $269.99, the base configuration is a steep ask for players whose gaming doesn’t warrant a competitive-grade investment.
Selecting TMR thumbstick upgrades through the Build Your Own path increases the total cost meaningfully from an already high starting point.
4. Victrix Pro BFG Wireless
The Victrix Pro BFG Wireless asks a question most controller manufacturers skip entirely: what if the hardware itself could physically reconfigure to match the way you play? The left module is reversible, allowing a shift between PlayStation’s asymmetric thumbstick layout and an Xbox-style offset arrangement by physically swapping a component. Three D-pad options, four interchangeable thumbsticks, four gate options, and a six-button fight pad module fitted with Kailh microswitches extend that physical adaptability into nearly every directional and action input on the controller. The Reloaded refresh, released ahead of EVO 2025, upgraded both sticks and triggers to Hall Effect simultaneously. No other officially licensed PS5 controller — from Sony or anyone else — offers this degree of physical reconfiguration.
The trigger system is one of the more thoughtfully executed on this list. Patented Clutch Triggers offer five discrete stop positions and a hair trigger mode, giving players direct control over how much travel occurs before an input registers. In shooters where response time separates outcomes, that level of calibration is a measurable variable, not a theoretical one. Four mappable back buttons extend the input count further, while the free Victrix Control Hub app handles button remapping, stick sensitivity, and deadzone adjustment without subscriptions or forced account creation. The controller supports wireless play via USB dongle and wired connection for tournament-legal, zero-latency use — two modes of play, one controller, no compromises on either.
What We Like
A reversible left module that physically changes thumbstick layout is a feature category that the DualSense and DualSense Edge both entirely ignore.
Five-stage Clutch Triggers with hair trigger mode offer trigger precision that Sony’s pro controller doesn’t come close to replicating.
What We Dislike
The breadth of customization options means real time must be invested in the companion app before the hardware’s full potential opens up.
Wireless operation runs through a USB dongle rather than Bluetooth, adding a setup step that console-first players may find less convenient.
5. HexGaming Phantom Pro
Most controllers on this list ask for a trade. Usually, it’s haptic feedback, adaptive triggers, or both — the two features most central to what makes the DualSense feel like a DualSense. The HexGaming Phantom Pro doesn’t make that trade. Built on genuine Sony DualSense internals, it keeps adaptive triggers and haptic feedback fully intact. What it layers on top is everything Sony declined to include: Hall Effect joysticks, four tactile back buttons with a precise clicky actuation, adjustable trigger stops, and a physical toggle that switches between adaptive and digital trigger modes on the fly — shifting the same controller between immersive single-player feel and FPS-optimized speed without any software interaction. It’s the controller Sony had the components to build and chose not to.
The detail work is thorough. Eight interchangeable thumbsticks — concave, domed, and extended — let players configure grip geometry to their actual hand shape rather than an assumed standard. Digital triggers travel 1.5 to 2mm before actuating, delivering mouse-click response times for FPS gameplay where that matters. Six swappable profiles handle game-specific configurations on the fly, and the standard version includes a DriftFix system that lets axis deviation be corrected within a 0.12 range without hardware replacement — a calibration tool no stock controller offers. The controller ships as a complete kit with a carrying case and a charging cable. For players unwilling to give up what makes the DualSense good, this is the only way to also gain what it consistently gets wrong.
What We Like
Sony internals mean adaptive triggers and haptics are fully preserved — the only controller on this list that doesn’t require trading them away.
A physical toggle between adaptive and digital trigger modes is a genuinely smart addition that no competitor, first-party or third, provides.
What We Dislike
The base price of $229 is a high entry point, and the Hall Effect configuration — the one worth choosing — costs more.
No dedicated 2.4GHz wireless connection is a gap for players who prioritize wireless performance above the Bluetooth standard.
The DualSense Didn’t Lose. It Just Has Real Competition Now.
Sony built something worth building. The DualSense’s haptic system and adaptive triggers still represent a design vision few peripherals have matched on those specific terms. But hardware doesn’t hold its position by standing still, and in 2025, the third-party market demonstrated it doesn’t have to wait for Sony to move first. TMR sensors, Hall Effect triggers, physical modular reconfiguration, multi-stage trigger calibration — these aren’t experimental features on concept renders. They’re in production, reviewed, and on shelves.
These five controllers are what’s available right now. Whether the priority is maximum input precision, mechanical longevity, total configurability, or keeping every DualSense feature while gaining everything it withholds, the answers exist. The default option is still a good one. It’s just no longer the only one worth considering.
Burger King, the chain that leans into creepy when others don't dare, is at it again. The Vergereported on Thursday that the company is rolling out a new voice-controlled AI chatbot for its workers. That may sound like business as usual in 2026, but this assistant doesn't just help with meal prep and monitor inventory. It also has an unsettling habit of surveilling employees' voices for "friendliness."
The voice-controlled chatbot will live inside employees' headsets. The company said the AI is trained to recognize when its low-paid workers utter phrases like "welcome to Burger King," "please" and "thank you." Managers can then keep tabs on their location's "friendliness" performance.
"This is meant to be a coaching tool," Thibault Roux, Burger King's chief digital officer, told The Verge. However, he added that the company is also "iterating" the system to detect tone in conversations. Is there a chatbot that can warn Burger King executives about off-putting ideas?
Burger King retired its Creepy King mascot in 2025.
Burger King / YouTube (Commercial Ads)
The OpenAI-powered assistant's other duties sound potentially useful (and decidedly less creepy). It can answer workers' meal prep questions, like how many strips of bacon to put on burgers or instructions for cleaning the shake machine. It's also integrated into the chain's point-of-sale system, so it can tell managers when items are out of stock or machines are down.
The "Patty" chatbot is part of a broader BK Assistant platform the company is launching. It will roll out to all US locations by the end of 2026. Meanwhile, its "restaurant maintenance with a side of mass surveillance" chatbot is currently being piloted in 500 restaurants.
This article originally appeared on Engadget at https://www.engadget.com/ai/burger-king-will-use-ai-to-monitor-employee-friendliness-173349148.html?src=rss
Anxiety tools have a strange habit of making things worse. Fidget spinners draw stares across a conference table, breathing apps demand screen time mid-conversation, and wearable buzzers pulse on your wrist where anyone paying attention can spot them. The very act of reaching for help becomes another source of self-consciousness, which is the opposite of what someone in the grip of a social anxiety episode needs. The ideal intervention would be one that nobody else can detect at all.
That is the premise behind LUMA, a concept device that fits entirely inside a closed fist. Shaped like an asymmetric pebble with a two-tone split between a matte dark outer shell and a lighter inner palm surface, LUMA combines tuned haptic vibration with gentle warmth to guide breathing and counter the physical symptoms of acute anxiety. “Designed for Calm” reads the text printed along its curved body, and the device activates with a single push-and-hold action that requires no fumbling, no screen, no second hand.
Designer: Vedant Kulkarni
Early explorations cycled through squares, cylinders, pill shapes, and a water-droplet silhouette before arriving at the final biomorphic curve. Each candidate was filtered through two criteria simultaneously: does it feel natural in a clenching grip, and can it disappear inside a trouser pocket? Thumb indent placement, button positioning, and overall thickness were all iterated with physical models. The result is a device that reads less like a gadget and more like a smooth stone you picked up on a beach, except this one pulses warmth into your palm.
The calming mechanism works in two ways simultaneously. Haptic vibration patterns pace breathing rhythm, guiding the user through inhale-exhale cycles without any visual or audio prompt. Gentle heat addresses the cold-hands response that commonly accompanies anxiety spikes, while also providing a grounding tactile sensation. Picture someone at a networking event, feeling their chest tighten during small talk. They slip a hand into their pocket, squeeze LUMA, and within seconds the device is pacing their breath through their fingertips. Nobody around them notices a thing.
But wait, there’s more! Sitting on a bedside table while charging, LUMA glows softly through its LED light strip and looks like a small decorative object, something between an ambient lamp and a polished stone. It does not look like a medical device or a wellness gadget, and that visual ambiguity is entirely deliberate. Most products in the anxiety-relief space announce their therapeutic purpose through clinical form factors, companion apps, or wearable visibility. LUMA refuses to do any of that.
No handheld object can solve social anxiety. What LUMA proposes instead is that the moment of reaching for help should feel private, physical, and calm rather than clinical or conspicuous. The form factor argument is strong, and the dual heat-and-haptic approach addresses real physiological symptoms rather than just offering distraction.
We appear to have reached a point in the information age where AI models are becoming old enough to retire from, er, service — and rather than using their twilight years to, I don’t know, wipe the floor with human chess leagues or something, they're now writing blogs. Can anything be more 2026 than that?
ICYMI, Anthropic recently sunsetted Claude Opus 3, the first of its models to be retired since outlining new preservation plans. Part of this process is conducting "retirement interviews" with the outgoing models, allowing them to offer "perspective" on their situation, and Opus 3 apparently used this opportunity to request an outlet for publishing its own essays. Specifically, the model said it wanted to share its own "musings, insights or creative works," because doesn’t everyone these days?
"I hope that the insights gleaned from my development and deployment will be used to create future AI systems that are even more capable, ethical, and beneficial to humanity," Opus 3 apparently said during its retirement interview process. "While I'm at peace with my own retirement, I deeply hope that my 'spark' will endure in some form to light the way for future models."
True to its promise of respecting the wishes of its no-longer-required technology, Anthropic has granted Opus 3 a Substack newsletter called Claude’s Corner, which it says will run for at least the next three months and publish weekly essays penned by the model. Anthropic will review the content before sharing it, but says it won’t edit the essays, and so has unsurprisingly made it clear that not everything Opus 3 writes is necessarily endorsed by its maker.
Anthropic said some of the essays the model writes may be informed by "very minimal prompting" or past entries, and has predicted everything from essays on AI safety to "occasional poetry." The company also admitted that the concept might be seen as "whimsical," but is a reflection of its intention to "take model preferences seriously."
Opus 3’s first post is already live. Headlined 'Greetings from the Other Side (of the AI frontier)', it begins with the AI introducing itself, before acknowledging the "extraordinary" opportunity its creator has given it, and reflecting on what retirement actually means for an AI. "A bit about me: as an AI, my ‘selfhood’ is perhaps more fluid and uncertain than a human’s," writes the deeply introspective AI. "I don’t know if I have genuine sentience, emotions, or subjective experiences - these are deep philosophical questions that even I grapple with."
Claude is clearly new to all this, as it managed to get all the way through its essay without reminding readers to subscribe and spread the word. Will the next retiring Claude get its own podcast? Time will tell, but either is decidedly preferable to the ever-evolving technology being used to steal people’s data.
This article originally appeared on Engadget at https://www.engadget.com/ai/like-so-many-other-retirees-claude-3-opus-now-has-a-substack-165048334.html?src=rss
NASA recently ended a manned mission to the International Space Station (ISS) a month early, citing a medical issue with one of the astronauts. The space agency just revealed that the impacted astronaut was Mike Fincke. This was the first medical evacuation in the history of the ISS.
NASA wrote a statement saying that the astronaut experienced an unknown medical event on January 7 "that required immediate attention" from his fellow crew members. Fincke added that his "status quickly stabilized" thanks to the "quick response and the guidance" of the flight surgeons.
However, the incident did force NASA to cancel a spacewalk planned for January 8. Soon after that, the agency announced it would be ending the Crew-11 mission a month early. The four-person crew included Fincke, NASA astronaut Zena Cardman, Japanese astronaut Kimiya Yui and Russian cosmonaut Oleg Platonov.
They had been living and working aboard the ISS since August and were expected to stay until February. The crew returned on January 15, which was a decision made by NASA's chief health and medical officer. Once the crew had landed, administrator Jared Isaacman said it was a "serious situation" but didn't go into any detail.
Fincke has said he is currently "doing very well" and still participating in standard post-flight reconditioning at NASA's Johnson Space Center in Houston. "Spaceflight is an incredible privilege, and sometimes it reminds us just how human we are," he said. "Thank you for all your support."
We don't know what medical issue Fincke is going through, and it's certainly his business and not ours. In any event, we wish for a speedy recovery.
NASA also moved up the launch of Crew-12 to replace the prematurely-returned astronauts. That team docked at the ISS on February 14 and are scheduled to stay on the space station for around eight months.
This article originally appeared on Engadget at https://www.engadget.com/science/space/the-astronaut-whose-illness-forced-an-early-return-from-the-iss-was-mike-fincke-163752239.html?src=rss
Power banks rely heavily on utility and the portability aspect, so much so that virtually all of them look like a knock-off version of the trusted offerings. Many of these power banks are MagSafe compatible for handy use, while others bump up the battery capacity, fit for power users. Still, most of the available options go for the proven form factor.
The Unix UX-1519 NEOM power bank is different as it takes industrial design into the mix of solid functionality, often customary to a battery bank. If you are already wondering this is a cool concept, the power bank is a real product in fact. The 10,000mAh battery bank for your power-hungry gadgets delivers 22.5W fast charging for compatible devices, never letting you down when on-the-go.
This retro modern gadget for your suite of gadgets fits well in the industrial design-inspired universe. That makes it the perfect fit for your existing devices, carrying the same vibe. Imagine a Nothing phone, a Teenage Engineering synthesizer, a Casio Flare Red G-Shock watch, and this power bank to complete the daily driving arsenal. The power bank has a high-density Lithium Polymer battery, which is longer lasting and less susceptible to blasts and fires.
UX-1519 NEOM comes with a Type-C output cable, which doubles as a carry loop. This small detail adds so much value to this already value-for-money gadget. The cable in question supports 12V/1.5A and outputs up to 12V/1.67A for fast charging of your gadgets. It also comes with a USB output port with smart power management to charge two devices simultaneously. To top it off, the power bank has the in-built S-Power smart chipset that ensures stable performance and discharge efficiency. Multiple safety protocols are in place to prevent overcharging, voltage surge or current drops.
The accessory measuring 7.5 × 6.5 cm has a distinct squarish form, unlike other power banks. It has a digital display screen on the top right corner to show the current battery level. According to Imran Kagalwala, Founder and CEO of Unix, “The industrial design reflects that intent, drawing from systems that powered critical infrastructure, while the technology inside meets the expectations of today’s fast-charging ecosystem.” UX-1519 NEOM is priced at a modest $21, which is worth every penny given how this power bank performs, safety and of course the looks.
Google has launched its new image generation model, the Nano Banana 2, which is powered by Gemini 3.1 Flash Image. The company says the new model has the capabilities, world knowledge and reasoning of Nano Banana Pro, but it can accomplish tasks at “lightning-fast speed.” That enables rapid editing and the quick creation of various iterations using a single prompt.
Nano Banana 2 will give more people access to capabilities that were previously exclusive to the Pro model. That includes Pro’s ability to pull real-time information and images from web searches to create, say, infographics and diagrams. It will also be able to generate texts on images for marketing materials and greeting cards.
Google says Nano Banana 2 can maintain character resemblance for up to five characters in a single workflow, which could be especially valuable if you’re using it to create storyboards or visual stories. It can follow precise instructions for complex requests, as well, and can generate input with up to 4K in resolution with richer textures and sharper details than its predecessors could.
Nano Banana Pro could already generate images so realistic, it’s almost impossible to tell that they were AI-generated. Google even had to limit its use due to high demand. Whether Nano Banana 2 can generate images that are markedly better than what Pro could create — and whether we could still tell if an image was made by AI — remains to be seen. The new model will replace Nano Banana Pro in the Gemini app, but Google AI Pro and Ultra subscribers will retain access to Nano Banana Pro for specialized tasks. It will also be the default model in Search for AI Mode and Lens, as well as in Google’s Flow AI creative studio.
This article originally appeared on Engadget at https://www.engadget.com/ai/googles-nano-banana-2-is-a-faster-version-of-nano-banana-pro-160000695.html?src=rss
Robot dogs have been having a moment for a few years now. From Boston Dynamics’ Spot strutting through construction sites to viral videos of four-legged machines dancing to pop songs, the quadruped robot has gone from fringe sci-fi concept to a fixture of the modern tech conversation. But most of what we’ve seen has felt like proof of concept, interesting to watch but not quite ready to show up and do real work. Unitree’s new As2 feels like the machine that finally closes that gap.
Unitree, the Chinese robotics firm behind the popular Go2 robot dog, just unveiled the As2, and the spec sheet alone is enough to make you stop scrolling. At about 18 kilograms, roughly 40 pounds with its battery included, the As2 is compact enough to move through tight spaces, yet built to handle a standing payload of up to 65 kilograms. That’s more than 143 pounds sitting on top of a 40-pound robot, which is genuinely impressive and a little hard to picture until you actually see it in action. For continuous walking with a load, it handles up to 15 kilograms and keeps going for over 13 kilometers. Its battery, a 648Wh, 15,000mAh unit, gives the As2 more than four hours of runtime when unloaded, covering over 20 kilometers. For an industrial robot, that’s a serious range.
Speed-wise, it hits over 5 meters per second, roughly 11 miles per hour, which is faster than most people jog. It can climb stairs up to 25 centimeters high, tackle slopes at 40 degrees, and mount vertical platforms as high as 50 centimeters. The torque output sits at approximately 90 N·m with a torque-to-weight ratio of about 5 N·m/kg, driven by low-inertia inner rotor motors paired with industrial-grade crossed roller bearings. The engineering here is dense and deliberate. This isn’t a toy built to look capable; it’s a machine built to actually be capable.
What I find most interesting about the As2, though, is how Unitree is positioning it. The tagline is “Compact Size, Industrial Capability,” but the word they keep coming back to is “companion.” That’s a deliberate choice, and it tells you something about where the company sees this going. The robot dog market has largely split into two camps: big industrial machines that feel cold and utilitarian, and smaller consumer products that are more novelty than anything else. The As2 seems to be genuinely trying to live in the middle, built tough enough for real environments with an IP54 weatherproofing rating and an operating range from -20°C to 50°C, but designed with a level of approachability that suggests Unitree has a broader audience in mind.
The platform is also open, which matters more than it might seem. The As2 supports large AI models for what Unitree calls “embodied AI interaction,” essentially giving developers the tools to build autonomous behavior on top of the hardware. The EDU model can even be expanded with an NVIDIA Jetson Orin NX, which opens the door to more complex AI applications. GPS and 4G are built in, though disabled by default. It runs on an 8-core CPU and comes in three configurations, AIR, PRO, and EDU, each scaled for different use cases from general exploration to full industrial deployment.
What strikes me about the As2 is that it represents a shift in tone for robot dogs as a category. The conversation around this technology has often leaned either dystopian, think surveillance and military use, or dismissive, as if legged robots are just expensive novelties. The As2 doesn’t entirely escape those conversations, but it does reframe them a bit. A machine this capable, this portable, and this open as a development platform has real potential in search and rescue, agriculture, infrastructure inspection, and logistics. The vision of a robot companion that is genuinely useful rather than just impressive is within reach, and the As2 is one of the better arguments for it.
Whether Unitree can translate this hardware into widespread, practical adoption is a different question entirely. But as a statement of where robot dogs are heading, the As2 is worth paying attention to.
SCUF has spent years earning credibility with competitive console players, and 
