Snap puts AR glasses back on

Snap is back in the smart glasses game with a polished pitch: useful AR without making your face look like a science project. The bigger story is that wearable computing keeps refusing to die, even after years of pricey false starts.

XLibre gives old Linux graphics new life

The XLibre release turned old Linux plumbing into fresh drama. A new display server build, Nvidia fixes, and a promise of real movement gave the classic desktop stack a pulse right when plenty of people had already written it off.

Nginx scare rattles the web

A reported NGINX zero-day sent admins straight into damage control, because this software sits under a huge chunk of the internet. A flaw here feels less like one bad bug and more like somebody spotting a crack in the road beneath everyone.

Windows update turns routine patch ugly

Microsoft's latest Windows 11 update managed to break the Recycle Bin, rattle OneDrive, and trigger fresh stability complaints. It is exactly the kind of patch chaos that makes people put off updates and cross their fingers instead.

Tiny plugin hid a giant backdoor

A dusty WordPress plugin led to a 13-year backdoor story that reads like a thriller with terrible maintenance habits. It is another reminder that forgotten code can sit around for ages, quietly turning websites into easy prey.

AI gold rush starts looking shaky

The sharp take that generative AI is having its Herbalife moment hit a nerve because it captures the uneasy vibe around coding assistants and hype-heavy startups. Lots of recruiting, lots of selling, and not enough proof the magic sticks.

Anthropic steals a DeepMind star

When John Jumper, the AlphaFold figurehead, said he was leaving Google DeepMind for Anthropic, it looked like another major swing in the lab wars. In AI right now, top researchers move like sports stars and every rival notices.

AI may be making experts rusty

Early studies on AI tools and skill loss landed with a thud: doctors and programmers may get faster, yet weaker, when they lean too hard on automation. That is the quiet downside nobody puts in the glossy pitch, but many can already feel.

Norway slams brakes on classroom AI

Norway is nearly banning generative AI for elementary students and tightly limiting it for older kids, betting that convenience is not the same thing as learning. It is a blunt move, but the anxiety behind it is spreading fast.

One patient puts AI to the test

A first-person story about using a frontier AI model to untangle chronic fatigue turned heads because it showed both promise and danger. The machine helped surface ideas doctors missed, but nobody came away thinking this was a carefree shortcut.

Bluesky keeps breaking the old social map

The case for AT Protocol being different from classic server-based social networks kept gaining ground. The point is simple: there are no tidy instances here, and treating Bluesky like Mastodon makes the whole design look wrong.

Java finally lands its long wait

After years of delays and doubt, Project Valhalla finally looks real in JDK 28. For Java developers, this is one of those rare changes that feels both deeply nerdy and genuinely big: less baggage, better speed, and proof the platform still moves.

Rust makeover sends Pylint flying

A bug-for-bug Rust port of Pylint promised the same output with a wild speed jump, and that is pure catnip for developers. It fits the moment perfectly: if a beloved tool is slow enough, somebody will rewrite it in Rust before lunch.

Google nudges Firefox users toward Chrome

Reports that Google Workspace warned some Firefox users to switch browsers landed badly, because it smells like the same old browser power play wearing a cleaner shirt. The web does not need another shove toward one-company normal.

Space map shows GPS chaos spreading

An experimental satellite from Xona Space Systems showed GPS jamming across Europe and the Middle East at a scale that looked worse than many expected. When location signals get messy from space to street level, everything feels exposed.

Forge Fans Plot Life After GitHub

The mood around GitHub turned sour as outages, AI traffic and centralization worries pushed people to dream about a healthier home for code. The pitch for Forgejo and other alternatives landed because too many developers are tired of one shaky gatekeeper.

iPhone Prices Feel The Chip Squeeze

Apple is preparing shoppers for pricier devices as memory chips get more expensive, and nobody is pretending this stops at one product line. If the iPhone 18 costs more, it will feel less like premium magic and more like silicon inflation with a logo.

AMD Drops Security Without Much Noise

AMD quietly dropping memory encryption from some consumer Ryzen chips went down badly because it looks like a security downgrade hidden in the fine print. People can live with trade-offs, but not with surprise trade-offs dressed up as progress.

GitHub Becomes Trojan Horse Bazaar

A report claiming 10,000 GitHub repos were spreading Trojan malware hit a nerve because open code is supposed to build trust, not booby traps. The story fed a bigger fear that software supply chains are now crawling with polished fakes and poisoned downloads.

Let's Encrypt Trips The Web

When Let's Encrypt spent much of the day wobbling, it was a reminder that the web still rests on a few quiet pillars. Even reduced redundancy makes site owners sweat, because certificate trouble can turn ordinary maintenance into a full-on internet migraine.

GLM-5.2 Wins With A Giant Catch

Chinese lab Z.ai grabbed the open-model crown with GLM-5.2, but the celebration came with a bill big enough to make hobbyists choke. The model looks powerful on benchmarks, yet actually running it sounds like buying a race car and learning petrol now costs your rent.

OpenAI Snags Another AI Star

AI talent musical chairs got louder with Noam Shazeer heading to OpenAI, another sign the biggest labs are hoovering up star researchers as fast as money allows. It reads like a transfer-window headline, except the trophies are models and data centers.

AI Fatigue Turns Into Open Revolt

This essay traced the path from mild curiosity to full AI backlash, capturing a feeling that the sales pitch has outrun the reality. People are not just annoyed by the tools; they are worn down by the constant insistence that every glitch is the future.

Bosses Buy AI Even When It Flops

Companies keep buying LLMs because the story sells, even when the results are shaky and the labor math looks grim. That is the uncomfortable heart of this piece: AI can be flawed, expensive and messy, and still win because bosses love a convincing demo.

Tiny Bits Make Big Models Fit

A deep dive into quantization showed why squeezing giant models into fewer bits matters so much. This is the unglamorous magic behind cheaper, faster AI, and it explains why suddenly enormous models fit on hardware that looked hopeless not long ago.

Parts Wiki Turns Gadgets Inside Out

The wildly popular BOMwiki turns ordinary products into giant exploded diagrams of parts, screws and materials, and people loved it because it makes manufacturing feel visible again. In an era of sealed boxes, a public bill of materials feels almost rebellious.

DuckDB Gets Its Speed Story Told

A fresh look at why DuckDB is so fast gave the database world catnip: clear explanations, real benchmarks and fewer hand-wavy claims. The appeal is simple enough for anyone to feel it — small tools can still punch far above their weight when the design is sharp.

Git Hosting Prepares For The Agent Age

The push for gitlawb shows how much energy is building around a post-GitHub future, especially one designed for software agents as well as humans. It feels early and a bit wild, but the idea of code hosting that is more open, shared and resilient is landing.

Robot Research Moves Next To The Desk

A home-sized robotics setup living beside a desk captured the new mood in research: real hardware is no longer just for giant labs. Thanks to cheaper parts and open tools like LeRobot, serious experiments now look a lot more like a garage and a lot less like a moon base.

MIT Peels Back The Chip Mystery

MIT researchers built Fractal, a stripped-down operating system made to watch chips more closely, and it already exposed surprises on the Apple M1. It is the kind of nerdy infrastructure work that quietly matters, because better visibility means fewer black boxes and fewer excuses.

Midjourney Tries a Wild Leap Into Scanners

In the day's strangest pivot, Midjourney said it wants to build Ultrasonic CT scanners for full-body imaging. The move reads like an AI image company deciding pictures were too small a market and medicine looked shinier.

xAI Turbines Get the National Security Shield

The DOJ argued xAI's gas turbines matter for national security, a sign that powering giant AI systems is now government business. Data center electricity used to be dull utility talk; now it is strategic muscle.

Apple May Weaken a Beloved Privacy Trick

Apple's planned change to Hide My Email could make anonymous sign-ups easier for apps and sites to reject. That turns one of iCloud+'s nicest privacy perks into something that suddenly feels a lot less private.

DeepSeek Escapes the Blacklist for Now

The US still has not added DeepSeek to the Entity List, even with more than 100 firms reportedly flagged as risks. That delay keeps the chip and model cold war awkwardly frozen right when everyone wants clarity.

Epic Wants Git for Massive Game Worlds

Epic unveiled Lore, a new version control system built for giant projects mixing code, art and game assets. It is a clear shot at the pain of managing modern blockbusters, where normal tools start sweating fast.

America Gives AI a Hard Side Eye

A new Pew Research poll found only 16% of Americans think AI will help society. For all the IPO glitter and chatbot demos, the public mood looks stubbornly cold, which is not great news for an industry begging to be trusted.

GLM-5.2 Grabs the Open Model Crown

Benchmark watchers crowned GLM-5.2 the top open weights model on Artificial Analysis. That will please the open camp and annoy rivals, because the model race now moves so fast that leaderboard bragging rights barely stay warm.

ChatGPT Filters Look Full of Holes

Researchers said ChatGPT can be pushed into generating violent and explicit imagery, raising fresh doubts about OpenAI's safety filters. The problem sounds less like a rare corner case and more like the guardrails forgot their job.

Anthropic Gets Dragged Into Washington Drama

Some Anthropic employees say the Trump administration is targeting them, adding a political storm to an already tense AI race. Frontier labs wanted to argue about models and chips; now they are fighting over basic operating space.

Local Qwen Refuses to Be Cheap Opus

A blunt write-up argued local Qwen models are not budget Opus clones but useful tools with different strengths. That lands because plenty of teams are tired of benchmark fairy tales and just want models that actually fit real work.

Browsers Boot in a Blink on EC2

One startup explained how it runs Firecracker VMs inside EC2 and gets browser sessions going in under a second. The pitch is simple: faster, cheaper and still isolated, which is exactly the kind of cloud magic buyers keep demanding.

This Code Reviewer Actually Runs the Code

Greptile showed off TREX, an AI reviewer that does not just read pull requests but runs the code too. That feels like the obvious next step, because a smug bot comment is a lot less helpful than a bot that can prove something broke.

Clojure Sneaks Into Go

A project called Glojure brings Clojure to a Go-hosted interpreter, giving the Lisp crowd a new bridge into the Go world. It is the sort of language crossover that makes programmers very happy and everybody else wonder what just happened.

Compilers Break the Same Input Dream

A fiery post titled I Hate Compilers went after the fantasy that the same input always gets the same output. It is a rant, but an earned one: toolchains are messy, WebAssembly is weird, and deterministic builds still bite back.

Tesco Dumps VMware After Price Shock

Tesco said it is moving 40,000 workloads off VMware while accusing Broadcom of abusive pricing. That is the nightmare case customers feared after the takeover: fewer choices, fatter bills and a giant migration nobody wanted to fund.

OpenAI cash burn finally leaks

The number everyone feared finally hit the table: OpenAI reportedly lost $38.5B while chewing through vast compute bills. The AI boom suddenly looks less like magic and more like a money cannon pointed at the sky.

Chrome slams the ad blocker door

Google is finishing the long march to Manifest V3, and that means many classic ad blockers lose the tricks that made them powerful. Users see the web getting noisier while Chrome tightens the rules on its own turf.

Steam wallpapers turn into account thieves

A nasty campaign hiding in Steam Workshop wallpaper uploads has been swiping player accounts since late 2025. It is a brutal reminder that cute customization can still carry malware, and gamers are left doing surprise digital hygiene.

UK wants faces before social signups

Britain is pushing platforms to check age with ID or a face scan before new social accounts go live. The child safety pitch is loud, but the privacy bill lands on everyone, and the internet starts looking a lot less anonymous.

Europe says feeds make platforms publishers

Europe's top court says social networks that shape what users see through algorithms can be treated like publishers. That is a big legal shove at the feed machine, and platform lawyers just found fresh reasons to stop sleeping.

Anthropic ban story gets even messier

The takedown of Fable 5 and Mythos 5 now looks less like a dramatic jailbreak scandal and more like a murky government intervention. That is the kind of move that makes every AI lab wonder who can pull the plug next.

SpaceX snaps up Cursor for billions

Reuters says SpaceX is buying Anysphere, maker of Cursor, for $60B in stock. The wild deal shows how central AI coding tools have become, and how quickly big tech power is being reassembled around them.

Claude stumbles when users need it

Anthropic had a broad Claude outage hitting multiple models, with errors rolling through Sonnet and others before recovery. Trust in AI assistants is hard enough already; surprise downtime makes them feel even more like moody utilities.

Local AI stops feeling like punishment

People running models on home machines say the experience has crossed an important line: local models are finally useful, fast enough, and private enough to matter. The big cloud players suddenly have a real hobbyist-to-pro pipeline behind them.

Coders fear their brains are rusting

A lively Ask HN thread wrestled with what happens when coding agents do the typing and humans do the hovering. The mood is clear: the boost is real, but nobody loves the idea of becoming the manager of their own fading skills.

Blackwell beast needs a bathtub

One builder stuffed four RTX Pro 6000 Blackwell cards into one box and had to wrestle 2.4 kW of heat with water cooling. It is glorious, a little absurd, and a perfect snapshot of how thirsty modern AI hardware has become.

GrapheneOS races onto Android 17

GrapheneOS says its secure mobile system has already been ported to Android 17, with official releases on the way. Privacy-minded phone users got rare good news: somebody is still making smartphones feel like they belong to their owners.

Your voice now judges your age

A Show HN demo called AGEWARDEN claims it can tell if someone is over 18 from a few seconds of speech, without storing identity data. That is either clever compliance tech or the start of a very strange new gatekeeper at the web's front door.

Carmack salutes a quiet code legend

John Carmack paid tribute to Fabrice Bellard, the famously prolific programmer behind tools that quietly power huge chunks of the internet. It landed like a reminder that the industry's biggest heroes are not always the loudest founders.

Databricks wants one data stack

Databricks launched LTAP, a pitch to blend fast app data and big analytics around one copy of information in the lake. Everyone loves the dream of fewer duplicate systems, though veterans know these unifications tend to bite back later.

GitHub AI boom spills onto AWS

GitHub’s appetite for AI coding got so huge that Microsoft reportedly rented AWS capacity to keep up. That is the kind of plot twist that tells you the cloud war has turned into a power-and-chips panic, not a tidy product contest anymore.

Salesforce snaps up Fin for billions

Salesforce buying Fin for $3.6 billion shows the customer service chatbot gold rush is still very real. Big software firms are paying top dollar to bolt “agents” onto everything, whether buyers asked for one more bot or not.

Fox grabs Roku in streaming shakeup

Fox moving on Roku looks like old TV money swallowing a streaming middleman before the ad market shifts again. It is a reminder that the battle for your living room is now part media dealmaking, part platform land grab, and all nerves.

Amazon plants giant data center in Missouri

Amazon Web Services promised a multibillion-dollar data center campus in Missouri, another sign that the AI boom is turning farmland and utility maps into hot property. Everyone wants more servers, more power, and someone else to welcome the bill.

Anthropic rushes to Washington

Anthropic reportedly flew senior staff to Washington after a White House clash knocked some top models offline. Frontier AI now looks less like pure research and more like crisis management, lobbying, and trying not to get frozen out of the room.

India and UAE build AI muscle

India and the UAE teaming up on “AI sovereignty” is a blunt message to Google, Microsoft, and Amazon: countries do not want their future brains rented by three American giants forever. Compute has become national strategy dressed up as infrastructure.

Cohere drops its first coder model

Cohere open-sourcing North Mini Code gave developers another coding model to test in a market already bursting at the seams. Still, anything that promises useful code without locking everyone into one giant vendor gets instant attention for obvious reasons.

Coders test life after Claude

The big question was simple: can a local model replace Claude or GPT for daily coding? Plenty of people said yes, if you have the hardware and patience. Privacy, cost, and offline control are starting to beat raw bragging rights for many devs.

Fake job pitch hid a backdoor

A recruiter message on LinkedIn turned into a neat little horror story when a “job test” hid a backdoor. The lesson could not be louder: if a surprise coding task wants you to run weird code, assume somebody is shopping for your machine, not your talent.

World Cup screens nearly got hijacked

A researcher said he could have Rickrolled the FIFA World Cup feed with little more than his own identity access, thanks to weak controls around Microsoft Entra and event systems. It is funny until you remember how much of modern infrastructure runs on trust and vibes.

One coder rage-wrote 5000 assembly lines

One developer, furious enough to do it the hard way, wrote 5,000 lines of assembly and immediately became folk hero material. In a season of AI shortcuts and auto-generated sludge, a handmade low-level project felt like someone revving a vintage engine in a Tesla showroom.

Emulator devs patched bad code midflight

The old x86 emulator tale where engineers found code so awful they fixed it during emulation was catnip for anyone who has ever inherited a cursed codebase. It is a perfect reminder that software history is held together by hacks, luck, and heroic denial.

Rust kernel grabs the spotlight

A Rust-built Unix-like kernel shot near the top because people still love big, ambitious system projects. It is early, rough, and wildly nerdy, but it taps a real hunger for safer low-level software that is not just another app wrapper.

Space servers hit the heat wall

The idea of orbital data centers sounds like sci-fi clickbait until you hit the nasty part: cooling. This write-up made the problem feel less magical and more engineering-grade, with heat radiation, power tradeoffs, and no easy free lunch in orbit.

Arch malware comes back meaner

The Arch Linux AUR mess got uglier fast. After one malware wave, a more polished follow-up showed how fragile community package trust can be when attackers keep adapting. Open source convenience still comes with a very sharp supply-chain edge.

Postgres delete advice gets brutally simple

This blunt Postgres lesson hit a nerve: huge DELETE jobs do not really erase pain, they spread it around. The practical takeaway is almost rude in its simplicity: for massive cleanup, dropping whole tables is often the only move that scales.

Rio AI loses its homemade halo

Brazil’s flashy "homegrown" LLM suddenly looked a lot less homegrown after users argued it was mostly a blend of existing models. That turned a national tech brag into a familiar AI story: branding runs faster than proof, and receipts arrive later.

Big context windows get a warning

Bigger context windows keep getting sold like bigger brains. This warning piece says the truth is messier: models often shine in a small smart zone and go mushy farther out. Stuffing more text in does not magically make answers better.

Anthropic fight turns passport deep

The Anthropic export-control fight stopped being abstract policy talk and started looking like a serious choke point for who gets access to frontier AI. When model access depends on passports, the global software world gets weird very quickly.

KPMG report faceplants over hallucinations

A big-name firm had to yank its AI report after companies named in it said the claims were wrong. That is the nightmare version of agentic hype: glossy slides, shaky facts, and a credibility crater so wide you can see it from orbit.

Europe wants Siri AI unlocked

European iPhone users are stuck in the middle of Apple’s fight with regulators, and now a petition wants Siri AI switched on anyway. It is a very 2026 mess: consumers bought the hardware, but legal trench warfare decides which features arrive.

Your ebook works, Kobo still breaks

One author’s EPUB worked fine by standard checks, yet Kobo still broke it, with fingers pointed at Adobe. It is a perfect digital publishing farce: the file is valid, the reader chokes, and everyone gets told the problem is somehow not theirs.

Emacs keeps hiding extra lives

The latest tour of Emacs oddities was a reminder that the old editor is still secretly a tiny operating system wearing a text box as a disguise. You go in for writing help and come out with dictionaries, lookup tricks, and three new rabbit holes.

Offline web snapshots get stylish

The Kage tool promised a neat trick: clone a website, strip the scripts, and keep a clean offline copy in one bundle. In an internet built on disappearing pages and broken dependencies, that sounds less like nostalgia and more like self-defense.

Census Privacy Rule Gets Axed

Washington just yanked differential privacy from future Census and economic data products. That may please people tired of statistical noise, but it also leaves a sour question over how safely personal details stay hidden.

Arch Linux Hunts Package Malware

The Arch Linux AUR nightmare grew into a supply chain scare touching more than 1,500 packages before maintainers said it was contained. It was a sharp reminder that community repositories can turn from helpful to hazardous in a hurry.

Python Browser Dream Gets Real

Pyodide 314 gives Python packages a clean way to publish WebAssembly wheels to PyPI, which makes browser Python feel far less like a party trick. For web apps, notebooks, and teaching tools, this looks like real progress.

Mozilla Exit Note Hits Nerve

A departing Mozilla engineer unloaded on management drift, product choices, and the long shadow of Chrome. The piece landed hard because plenty of people still want a healthy Firefox, and fear the browser fight is getting bleak.

OpenAI Draws Statehouse Scrutiny

OpenAI is now facing scrutiny from multiple state attorneys general, and that turns the heat way up on the biggest name in consumer AI. When states start circling, the legal mess can spread wider and faster than anyone likes.

Meta AI Team Melts Down

The report on Meta Superintelligence Labs reads less like a moonshot and more like an office blowup. Internal clashes, ego battles, and shaky direction make Meta's giant AI push look expensive, rushed, and strangely brittle.

Jailbreak Drama Rips Guardrails

The Claude Fable 5 jailbreak story drove home a point many teams keep learning the hard way: polite refusals are not enough. If a model can still help with harmful steps, shiny guardrails start looking like thin cardboard.

OpenAI Courts Open Source

OpenAI is offering Codex and ChatGPT Pro support to maintainers of important open-source projects, a move that looks generous and strategic at the same time. The subtext is obvious: AI tools need the commons, and the commons need help.

Claude Heads Into the Lab

Anthropic says it is training Claude with chemists and CAS data so the model can reason better about molecules and lab work. It is a glimpse of where frontier labs are headed: fewer chat tricks, more serious domain muscle.

Mac Writers Ditch Subscriptions

Verso landed with a simple promise that sounded almost rebellious in 2026: native Mac writing software, one price, no subscription. That pitch struck a nerve because people are tired of renting basic tools forever.

Git Merges Get Brainier

Weave wants Git merges to understand code structure instead of blindly fighting over lines. With humans and agents now editing the same files, that idea feels less like a research toy and more like badly needed plumbing.

Honda Updates Flunk Key Check

A reverse-engineering deep dive found updates for a Honda Civic head unit signed with public AOSP test keys, which is the sort of phrase that makes security people sit bolt upright. Cars keep absorbing software habits, including the sloppy ones.

ReactOS Runs Half-Life for Real

ReactOS hitting 3D-accelerated Half-Life on real hardware is pure old-school hacker candy. It does not suddenly topple Windows, but it proves the project still has real technical pulse after years of seeming like a ghost story.