AI Resume Judge Trips Over Itself

A test of HackerRank’s open-source hiring tool showed the same resume could score 90, then 74, then 88. That is a terrible look for AI screening. If a bot cannot grade one CV the same way twice, why should anyone trust it with jobs?

Giant AI Chip Ditches Memory Bottlenecks

The proposed Sophon PFG-1 chip promised a wild shortcut around today’s HBM crunch: pack an enormous 330 GB of memory right onto the die. Even as a bold pitch, it captured the mood of an industry desperate for cheaper ways to feed AI.

Ford Brings Veterans Back

Ford said it rehired 350 veteran engineers after AI and automated quality systems failed to catch enough manufacturing issues. For all the boardroom talk about replacing experience, the factory floor just voted for people who know where the squeaks live.

China Chases the Chip Crown

China’s push to build an ASML-like machine stayed one of the day’s biggest industrial stories. With export controls squeezing access, the country is pouring effort into homegrown chip tools, even if catching the Dutch giant still looks brutally hard.

New OS Lets Kernels Compete

The new QSOE release turned heads by pitching one operating system with selectable kernels, borrowing ideas from QNX while targeting modern hobbyist and embedded work. In a world of endless apps, seeing fresh OS ambition felt downright refreshing.

Chinese Model Tops Claude on Security

Semgrep said GLM-5.2 from Zhipu AI beat Claude Code on its security benchmark, a result that cut straight through the usual frontier-lab pecking order. The bigger story is how fast strong rivals are closing in.

Google Rations Gemini for Meta

Google reportedly limited how much Gemini capacity Meta could use, a deliciously awkward twist in Big Tech’s AI triangle. Renting brains from a rival always looked risky, and now the supply chain drama is showing up in plain sight.

Labs Learn from Black Boxes

A paper on knowledge distillation argued smaller models can learn plenty from powerful black-box systems like GPT-4 without seeing their internals. That keeps the cost war alive and suggests frontier secrets leak through behavior more than branding.

Codex Still Lacks a No Peek Switch

An open issue asked for a .codexignore-style way to stop OpenAI Codex from reading sensitive files. That sounds like a basic safety belt, yet it is still missing. Agent tools clearly shipped at speed while privacy controls lagged behind.

Claude Reads an MRI Report

One writer used Claude Code to read an MRI report as a kind of second opinion, with cautious caveats. It is equal parts useful and nerve-racking: the tool can translate medical jargon, but nobody wants bedside confidence from software that still hallucinates.

Congress Eyes Web Age Checks

Congress looked ready to move on the KIDS Act, which would push sites and apps toward age checks before people can browse or message freely. It was sold as child safety, but the practical result looks a lot like ID gates for the open web.

EU Revives Chat Scanning Fight

The EU’s Chat Control fight flared again, with critics warning that lawmakers were trying to advance message scanning behind closed doors. Once governments treat private chat as inspectable by default, secure messaging stops feeling very secure.

Street Cameras See More Than Plates

The spread of Flock cameras kept raising the same cold question: if the system can spot far more than a plate number, how long before everyday driving becomes one more searchable behavior trail. Convenience is the pitch; surveillance is the product.

Apple's New Disk Format Gets Opened

A deep dive into Apple’s new ASIF sparse image format showed the usual magic trick in Cupertino land: new file tech arrives, and outsiders have to pry it open bit by bit. Reverse engineering matters because closed formats quietly shape who gets to interoperate.

Memory Prices Tell a Brutal History

A giant timeline of memory prices from 1960 to 2026 turned storage history into a chart of collapsing costs and new bottlenecks. It makes today’s AI hardware scramble easier to read: cheap bits built the boom, scarce fast memory now taxes it.

Meta faces ugly surveillance lawsuit

Meta’s nightmare week got worse after the author of Careless People said the company tracked her for a year to keep her quiet. The case lands right on the ugliest possible mix of surveillance, power, and Silicon Valley ego.

Fords AI shortcut hits a wall

Ford tried the classic boardroom magic trick: swap humans for AI and hope quality stays put. Instead, the plan reportedly helped create a mess serious enough to fuel fresh doubts about automation being treated like a cheap miracle cure.

Texas wants app stores checking ages

Texas’ App Store Accountability Act is being framed as child safety, but critics see a giant age-check machine aimed at every phone user. If it sticks, Apple and Google could end up policing identity far beyond Texas.

Mystery GitHub account dumps fresh exploits

An anonymous GitHub account started dumping alleged 0-days like candy, and that is exactly as bad an idea as it sounds. The stunt thrilled chaos fans and terrified defenders, because unpatched bugs do not care who finds them first.

Silicon Valleys old voice falls silent

The death of Om Malik at 59 felt like losing one of the few people who could explain Silicon Valley without sounding hypnotized by it. His work at Gigaom taught the web to cover tech as culture, power, and business all at once.

Asian rivals chase the Mythos buzz

Chinese and Japanese players are rolling out Mythos-like models, a blunt reminder that the AI race is not waiting for a tidy US-only script. If one lab finds a hot niche, rivals everywhere will clone, tune, and ship.

Cybersecurity people pour cold water on Mythos

After the hype around Claude Mythos Preview, security people are already doing the adult thing and calming down. The takeaway is less robot apocalypse, more faster workflows, sharper attackers, and one more tool nobody can ignore.

AI invents radio chips humans miss

Researchers say AI is producing radio chip layouts that humans would never sketch, which is thrilling and a little unnerving. The machines are not just writing emails now; they are wandering into deep engineering territory.

Writers now must prove they wrote

When writers have to prove a sentence was not made by an LLM, the internet has officially entered its trust crisis era. Tools meant to help people write are also making honest work look suspicious, which is a rotten bargain.

Smart router picks local or cloud AI

Wayfinder Router promises a practical AI trick people actually want: send easy prompts to local models and harder ones to the cloud without asking another model what to do. It is the kind of cost-cutting common sense this market needs.

Forgotten IBM graphics chip gets cracked open

A reverse-engineering deep dive into IBM’s MCGA chip turned obscure old hardware into catnip again. It is part detective story, part preservation mission, and a reminder that yesterday’s cheap graphics hacks still deserve serious respect.

The Apple II card that meant business

The Videx VideoTerm card helped turn the Apple II from hobby toy into office machine, and recreating it on FPGA shows how much hidden craft lived inside early personal computing. Old expansion cards are getting the grand biography treatment.

Linux gives rejected PCs another shot

The latest guide to reviving abandoned PCs with Linux lands right as Windows 11 keeps pushing older machines off the table. People are clearly tired of being told a perfectly fine laptop is e-waste because a checklist says so.

OpenTTD keeps the train empire rolling

OpenTTD 16.0 beta proves the forever game still has plenty of track left. Big community projects like this keep winning because they age better than a lot of glossy modern releases, with player freedom beating cinematic clutter every time.

Web MIDI fights a stubborn 1983 synth

Getting Web MIDI to play nicely with a 1983 Yamaha DX7 sounds niche until you remember the modern web keeps colliding with beloved old gear. The result is a charming battle between browser ambition and hardware that refuses to be rushed.

Engineers Grieve the Job They Loved

A blunt essay on developer grief struck a nerve. The old promise of stable, meaningful software work looks badly shaken as layoffs, AI tools and constant churn turn proud builders into anxious survivors. Software careers suddenly feel far less solid.

AWS Builds Safer Boxes for Wild Code

AWS unveiled Lambda MicroVMs so companies can run user-made or AI-generated code inside isolated, stateful environments. It is a clear sign that cloud platforms now treat untrusted code as the main event, not a weird corner case.

Micron Sells the Memory Crunch

Micron reportedly locked customers into unusually high DRAM and NAND prices for five years, a reminder that the AI boom is not just about chatbots. Memory is becoming a hard power lever, and cheap hardware is looking like yesterday's dream.

Big Firms Rally to Protect Open Source

A new effort called Akrites wants major companies to coordinate fixes for fragile open source components the whole world quietly depends on. The message is sober and overdue: the digital plumbing is critical, exposed and under-defended.

AMD Finally Frees HDMI on Linux

AMD moving toward full open HDMI 2.1 support on Linux felt like one of those long overdue wins. It matters for gaming handhelds, desktops and anyone tired of paying modern hardware prices while still living with odd display compromises.

LLM Costs Hit the Wall

One argument dominated: today's LLM economics look wildly out of line. If every useful AI task needs a premium model and mountains of compute, the industry is building a very fancy bonfire of cash. Something simpler has to give.

AI Backlash Leaves Tech Exposed

The warning is blunt: AI is becoming politically unpopular, and tech cannot hand-wave that away forever. People are less dazzled by demos when they mostly see job fears, spam, surveillance and products that still feel half-finished.

AI Money Floods the Election Map

AI money is now pouring into U.S. elections, turning model makers and investors into a serious political force. That drew quick suspicion, because an industry already accused of moving too fast is now trying to shape the rules around itself.

Washington Wants a Say on ChatGPT

Reports said the U.S. government may decide who gets the latest ChatGPT upgrade, a sharp turn from the old move-fast mood. Once governments start rationing frontier tools, AI stops looking like a normal app market and starts looking strategic.

Open Models Keep Chasing the Giants

The race between open weights and closed models is still tighter than the loudest marketing suggests. Closed labs keep the edge, but the gap is no longer cartoonishly huge, which keeps hope alive for cheaper and more open alternatives.

Security Team Dissects a Failed State Hack

A detailed breakdown of a failed suspected nation-state intrusion was irresistible reading for security people. The real lesson was not movie drama, but patient forensic work showing how much messy human effort hides behind flashy attack headlines.

Amazon Tries to Fix Multiplayer Hosting

Amazon opening up GameLift Servers got a warm reaction because multiplayer hosting is usually pain wrapped in invoices. Anything that makes online games easier to launch and scale without a small ops army feels like a very real win.

Space Force Wants Satellites on Speed Dial

The U.S. Space Force wants new satellites deployed in weeks, days or even hours instead of years, and that pace shift is striking. Space is starting to look less like slow prestige hardware and more like fast-moving infrastructure with uniforms.

Rust Database Beats H100 with Gamer GPU

A Rust database using gaming GPU ray tracing cores to beat an H100 on spatial work is exactly the sort of benchmark story that makes hardware people sit upright. It hints that clever software can still embarrass eye-watering spending.

AOL's 1996 Meltdown Still Feels Familiar

The retelling of AOL's 1996 collapse read like ancient history with painfully modern lessons. Overload, brittle systems and confused operations are not relics at all. The logos changed, but reliability failures still rhyme embarrassingly well.

EU puts cloud giants on notice

Europe is moving toward calling AWS and Azure digital gatekeepers, which could force new rules on pricing, switching, and bundling. Cloud used to feel untouchable; now Brussels looks ready to kick the server room door open.

IBM shows off tiny chip future

IBM rolled out a flashy sub-1 nanometer chip claim built around stacked 3D design. It is still research, not tomorrow's laptop, but the message is loud: the chip race is not done, and physics still has a few wild tricks left.

Deno chases desktop without Electron

Deno 2.9 wants web developers to ship desktop apps without dragging around the usual Electron weight. A single binary and less boilerplate sound great, and the pitch lands because plenty of people are tired of simple apps eating half a laptop.

Windows 10 gets a quiet extra life

Windows 10 was supposed to be over, then Microsoft quietly found one more year of updates for regular users. That tells you everything about how many PCs are still stranded there - and how messy forced upgrade plans keep looking in the real world.

Framework's fast Ethernet reveals USB-C chaos

A look at Framework's 10G Ethernet module turned into a perfect little horror story about USB-C. One tiny port still hides power limits, cable quirks, and chipset drama, which is why buying the right dongle somehow remains a full-time job.

Why the AI backlash keeps growing

The anti-AI mood is no longer a niche grumble from artists and teachers. This argument ties together job fear, bad products, and a general sense that executives keep selling inevitability while everyone else is asked to swallow the downside.

AI kids books turn into nightmare fuel

Cheap AI books for kids are flooding online shelves with creepy art, broken logic, and the same recycled mush. It is funny for about ten seconds, then bleak, because the whole mess shows how easily low-cost slop can crowd out trust.

The AI underclass fear goes mainstream

The fear that AI could create a permanent class of people with no bargaining power is moving from sci-fi panic to serious argument. Once the tools look good enough, learn to code stops sounding helpful and starts sounding like a cruel joke.

OpenAI keeps Wall Street waiting

OpenAI may wait until next year before trying an IPO, which is a neat way of saying the hottest company in tech still has plenty to sort out. Between structure, power, money, and expectations, this story keeps looking more complicated than the hype.

Hackers swarm an AI assistant test

One developer invited the internet to jailbreak an AI assistant and got exactly the stampede you would expect. More than 2,000 attempts turned into a useful stress test, and a reminder that putting secrets near a model is still asking for trouble.

Wikipedia staff push first global union move

Workers at the Wikimedia Foundation in Britain are seeking union recognition in a first for Wikipedia staff anywhere. For a site built on ideals and volunteer spirit, this is a sharp reminder that mission-driven tech jobs are still jobs.

LastPass gets burned again

LastPass users are dealing with yet another breach notice, this time tied to an outside partner. At some point third-party incident stops sounding like an excuse and starts sounding like the business model is held together with wet tape.

Curl's ancient bug closet bursts open

Researchers found six new curl flaws, including what looks like the oldest reported issue in the tool's history. That is unsettling precisely because curl is everywhere, quietly moving data around the modern internet like plumbing nobody checks.

Om Malik leaves a huge media void

Tributes poured in for Om Malik, one of the rare tech writers who could explain Silicon Valley without sounding dazzled by it. His work shaped how the industry understood itself, and the grief hit because that kind of voice is now painfully scarce.

Bohemia opens a Cold War time capsule

Bohemia Interactive put the remastered Cold War Assault source code on GitHub, turning a 2001 military sim into a fresh playground for tinkerers. Old game code dumps are catnip for preservation fans, and this one lands with real historical weight.

Wikipedia staff push for a union

Wikipedia’s UK staff are moving to force union recognition, a first for the famous online encyclopedia. It is a sharp reminder that even idealistic tech institutions still run into old fights over pay, power, and working conditions.

Cloudflare opens OAuth plumbing to everyone

Cloudflare rolled out self-managed OAuth more widely, making it much easier to connect outside apps to its platform without messy custom work. It is the kind of quiet infrastructure update that saves developers time and reduces future lock-in.

Bunny makes DNS free

bunny.net made Bunny DNS free, gambling that better basic internet plumbing will win customers later. Free DNS is not flashy, but faster lookups and one less bill are exactly the sort of move that gets engineers to pay attention.

Hotter cooling tackles thirsty AI data centers

NVIDIA-linked partners showed a 45°C liquid cooling design for AI facilities that pushes water use close to zero. As giant AI racks spread everywhere, the boring machinery behind the scenes is suddenly starting to look like front-page tech news.

GTA 6 goes pricier and disc-free

GTA 6 will cost $80, and the boxed version will not even include a disc. That landed like a warning shot for consumers: bigger game budgets now mean higher prices, less ownership, and one more shove toward an all-digital future.

OpenAI builds its own chip

OpenAI showed off Jalapeño, its first custom inference chip with Broadcom, making it clear the company wants less dependence on rented GPUs. The AI race now looks like a full-stack brawl, from model weights down to raw silicon.

Anthropic alleges massive Claude siphoning

Anthropic says Alibaba-linked operators used 25,000 accounts to pull answers from Claude and help improve Qwen. Whether you call it distillation or digital shoplifting, the fight over copying AI behavior just got a lot uglier.

Gemini starts clicking around for you

Google added computer use directly to Gemini 3.5 Flash, letting the model interact with tasks instead of just chatting about them. It is another nudge toward AI helpers that actually do office work rather than merely describing it.

GLM-5.2 shakes up open agents

Excitement around GLM-5.2 suggests open models are no longer just cute alternatives for agent work. The mood has shifted toward real competition, especially for teams tired of paying premium prices to a tiny club of giant AI labs.

Cheap open models raise eyebrows

A pricing look at DeepSeek V4 showed just how cheap some open-weight models have become next to Anthropic and OpenAI. That huge gap is turning curiosity into action as more builders test whether cheaper is now good enough to win.

Europe revives the encryption nightmare

A fresh push for chat control in Europe put encryption back in the firing line. The fear is simple: once governments demand scanning inside private messages for safety, private communication stops feeling very private at all.

Programmers ask where the job goes

A raw Ask HN thread about the future of programming captured the industry’s nerves in plain language. With LLMs writing more code and teams staying lean, people are wondering if the craft is changing faster than careers can adapt.

Elastic cuts staff again

Elastic said it will lay off 7% of employees, another reminder that even established software firms are trimming hard while talking up AI and automation. The corporate version is efficiency; the worker version is more empty desks.

Tesla crash reignites autonomy doubts

A Tesla smashed through a Texas home, reviving the endless question of where driver error ends and self-driving responsibility begins. The promise of autonomy still keeps arriving with a frightening amount of ambiguity attached.

CAPTCHAs look cooked

A blunt argument made the rounds that CAPTCHAs have been losing to machines for twenty years, and AI agents may finally finish the job. If that is right, one of the web’s most hated rituals is giving us far less protection than advertised.

F3 Tries to Rewrite Data Files

A new F3 format pitched itself as a cleaner, faster answer to Parquet, with better structure and more room to grow. It sounds like dull plumbing until you remember this is exactly the layer that quietly decides how modern data stacks behave.

iPhone Boot Code Gets Cracked

Researchers disclosed usbliter8, a SecureROM exploit for Apple A12 and A13 devices. That is the sort of hardware flaw people truly fear, because bugs this deep can survive for years and leave owners with very few good escape routes.

Private Web Passes Push Forward

The Pact proposal aims to prove a user is legitimate without turning the web into a nonstop CAPTCHA gauntlet or demanding ID everywhere. It felt like a direct pushback against creepy web gatekeeping, and that gave it real spark.

Oracle Cuts Deep for AI

Oracle revealed roughly 21,000 job cuts while reshaping around AI and cloud demand. The message was painfully plain: the AI gold rush is still minting big corporate stories, but plenty of workers are being asked to pay the entry fee.

OpenAI Wants Bots Patching Bugs

OpenAI expanded DayBreak with GPT-5.5-Cyber, aiming to discover and patch serious software flaws at machine speed. The pitch is thrilling and slightly chilling, because defenders now need automation simply to keep pace with attackers.

LLMs Still Miss the Obvious

The Reversal Curse paper showed a nasty blind spot: teach a model that A is B, and it may still fumble B is A. For all the confident demos and polished chatter, this was a cold splash of water on the whole understanding debate.

Qwen Chases World Model Agents

Qwen-AgentWorld pushes the idea that language models can build internal world models to plan and act more effectively. It is ambitious stuff, the kind that makes labs sound closer to digital brains even when the benchmarks still need side-eye.

Gemini Slips Into Thinking Loops

Users complained that Gemini sometimes resets, forgets context, or gets stuck circling its own thoughts. Nothing drains faith faster than a paid assistant acting like it lost the thread midway through the job and refuses to admit it.

Claude Has a Rough Day Online

Anthropic reported elevated errors across claude.ai, the API, and Claude Code, a reminder that AI tools now sit directly inside real workdays. When the model hiccups, developers, teams, and deadlines all get dragged into the mess.

Offline Swipe Typing Gets Real

FUTO rolled out FUTO Swipe, an offline gesture typing model for Android that tries to match the big keyboard apps without the constant data vacuum. In a market built on collection first, plain old privacy suddenly felt rebellious again.

Rhombus Reaches Version One

The Rhombus language hit 1.0, bringing a new design from the Racket world and another serious attempt to rethink how programming feels. New languages appear every week, but this one landed with the rare smell of real craft.

ASCII Art Goes Full 3D

A Show HN demo rendered textured 3D scenes in plain ASCII inside the browser, without canvas or WebGL. It is gloriously impractical, which is exactly why people loved it: weird, clever, and proof the web can still surprise you.

Steam Machine Nostalgia Fires Up

Fresh Steam Machine chatter had hardware romantics getting emotional about Valve’s living-room dream all over again. With SteamOS gaining credibility, the idea no longer feels like a punchline, just a beautiful gamble people want to believe in.

Valve brings PC gaming downstairs

Valve’s new Steam Machine looks like a serious attempt to put PC gaming back in the living room without the usual box-of-cables chaos. With SteamOS and beefy AMD hardware, the old dream suddenly feels less like a rerun and more like a launch.

The face scan fight turns nasty

The loud warning against facial scans hit a raw nerve because the pitch is always the same: safety, children, convenience. Then one day your face becomes the price of entry for the web, and that bargain looks rotten from every angle.

Microsoft buys decades of power

Chevron signing a 20-year power deal for a Microsoft data center made the AI boom look even hungrier. This is no longer just about faster chips and flashier chatbots. It is about who can lock down electricity before everyone else does.

Apple supplier breach sparks secret fears

A claimed cyberattack on Tata Electronics rattled nerves because this is not some obscure supplier. It sits near the heart of hardware manufacturing for names like Apple and Tesla, so any leak whispers ugly things about the whole supply chain.

Open model GLM picks a fight

The arrival of GLM-5.2 set off the usual online stampede, but this time the excitement had teeth. An open model being weighed against Claude Opus says the pecking order is shifting, and the closed giants no longer look comfortably untouchable.

GLM rushes onto local machines

The guide to running GLM-5.2 locally poured fuel on the fire. A massive model with a 1M context window landing in the hands of tinkerers turns the story from lab demo into home experiment, and that is when the real chaos usually starts.

Tiny model talks big on reasoning

A 3B-parameter model claiming reasoning results above Opus 4.5 is exactly the kind of headline that makes big-budget AI labs sweat. If the training recipe matters more than brute size, then the small-model revolution just got much louder.

Codex bug hammers local SSDs

The Codex logging bug was the kind of problem that makes people clutch their laptops. Reports of runaway writes filling local SSDs with giant SQLite logs are a sharp reminder that AI helpers can still behave like toddlers with root access.

Claude thinking looks more like theater

A close look at Claude Code logs argued that its much-touted extended thinking is really a neat little summary rather than raw inner thought. That does not kill the feature, but it does puncture the mystical aura that marketing loves to float.

Polymarket hype videos smell fake

The Polymarket promo machine looked a lot grimier after reports that viral 'big win' videos were made with fake bets on cloned sites. It is a perfect modern mess: slick creators, shady growth tactics, and viewers left holding the most obvious bag.

Memcached gets its overdue flowers

The case for memcached read like a love letter to boring software that simply works. While newer tools keep demanding migrations, modules, and drama, this old cache keeps doing its job quietly, which suddenly feels almost rebellious.

Deno chases the desktop dream

Deno Desktop wants to turn a TypeScript project into a desktop app without the usual packaging headache. That pitch lands because developers are tired of giant stacks for simple tools, and anything promising fewer moving parts gets instant attention.

Linux secure boot hits a nasty date

The warning about Linux and expiring Secure Boot certificates was dry on paper but nasty in practice. Few things ruin your day faster than a machine that refuses to boot after an update, especially when the deadline was hiding in plain sight.

Your smart TV may moonlight online

Research saying nearly half of LG smart TV apps contain residential proxy software is the sort of detail that makes a living room feel creepy. People bought screens to watch shows, not to wonder whether the TV is moonlighting as network infrastructure.