HN Brief: 2026-10-05
A roving argument about AI's place in making things dominated the front page, with the vibe coding proxy war flaring across Rust Adobe clones, Mac search tools, and VB6 emulators alike. A deeper fracture over trust converged from three directions—cars that leak through a Faraday cage, an Anthropic reviewer flagging a user's diary, and an open-source maintainer burying a security patch for six months. The counterweight was a streak of nostalgia for the tangible: a rescued film-studio archive, a deliberately useless conveyor-belt keyboard, and a eulogy for the instant feedback loop of classic RAD tools.
The bind of trusting a closed AI with private thoughts is laid bare in 'Anthropic reported diary entry to police, woman faces felony charge'. The day's hottest proxy war is 'ArtCraft Apps – open-source Adobe compatible suite written in Rust', pitting AI-generated ambition against professional domain engineering. The most perfectly circular debate belongs to 'Improper redaction reveals Google Data Center water and electricity usage', where a genuine scoop gets swallowed by a philosophical melee over consumption metrics. A single dark turn in 'Nearly 200 people under observation after Irkutsk lab worker dies from plague' lets a discussion of lab safety wander into the realization that bubonic plague is already endemic in the US. And the most surgically effective detail of the day comes from 'Car is a smartphone on wheels. Here's who's listening', where a Faraday cage fails to stop the leak.
Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s [comments]
769 points · 343 comments · github.com · 19h ago
Strata is a new tool that lets you run the 125B-parameter Qwen 3.8 Flash Next on a consumer GPU like the RTX 4090, hitting over 100 tokens per second locally. People are reporting it works shockingly well on modest hardware—even a 3080 with 10GB VRAM can run the coder variant—sparking direct comparisons that for coding, this MoE model handily beats Qwen 27B and Claude for a lot of people. The main split wasn't about the speed, but the quality of the heavily quantized models, with a particularly sharp exchange between those arguing from "vibe checks" and those demanding proper benchmarks against denser 4-bit models. A smaller but notable thread broke out around the security of the README telling AI coding agents to run setup scripts, drawing a clear line between that practice and the familiar `curl | bash` complaints everyone knows.
Turn off Apple Intelligence on macOS 27 and get its disk space back [comments]
536 points · 354 comments · github.com · 12h ago
The submission is RemoveMacAI, a tool that disables Apple Intelligence on macOS 27 and deletes its model files to recover disk space, since Apple doesn't provide a single toggle to get that storage back. The discussion almost entirely ignored what the tool actually does to instead wage a proxy war over the curl | bash installation method in its readme. One camp argued that piping a script from the web is a genuine security risk, as a server that sees the connection can deliver targeted malware without leaving a static artifact for anyone to audit. The other camp countered that the outrage is mostly security theater, pointing out that macOS is a closed-source binary, the Homebrew tap relies on the same trust chain, and the same critics are unlikely to audit the code they pip or brew install either. The thread sacrificed any real examination of RemoveMacAI's internals entirely to the latest skirmish in the eternal debate over running remote scripts directly in your shell.
Improper redaction reveals Google Data Center water and electricity usage [comments]
369 points · 474 comments · www.1011now.com · 12h ago
The article is a local Nebraska news station finding Google data centers botched the redaction of their state reports, exposing specific electricity peaks, water consumption up to 547 million gallons per facility, and a combined $117 million in expected tax refunds. The HN thread almost entirely ignored the transparency story and the tax breaks, pivoting ferociously to debate whether the water numbers actually mean anything. A heavy contingent did the math to show the water use is a rounding error compared to local agriculture, ethanol, and golf courses, while another faction shot back that this exact “look over there” framing is a successful industry deflection away from energy use and the subsidy question entirely. The argument was airtight and completely disconnected from the local reporting that sparked it, turning into a generic philosophical melee over what level of resource consumption deserves protest.
Car is a smartphone on wheels. Here's who's listening [comments]
209 points · 140 comments · automatictransmission.khoury.northeastern.edu · 16h ago
A paper out of Northeastern and Consumer Reports tore down 21 modern cars and 30 companion apps, visualizing a data pipeline that funnels driving habits and phone data straight to data brokers and insurers with zero consumer opt-out. HN seized on the finding that sealing the car in a Faraday tent couldn't stop the leak—it just switched from cellular to Wi-Fi—as proof the surveillance is a mandatory hardware feature. The thread split between those applying the standard "you're the product" diagnosis and a grimmer faction pointing out that even paid subscriptions don't buy privacy, opt-outs just brick navigation. The argument widened into a debate over whether public apathy is complicity or exhaustion, but the sharpest turn came when someone admitted that proposing GDPR-style enforcement on HN itself gets downvoted for being "anti-business," forcing the community to confront a blind spot in its own ideology.
Nearly 200 people under observation after Irkutsk lab worker dies from plague [comments]
201 points · 173 comments · www.themoscowtimes.com · 5h ago
A lab worker at an anti-plague institute in Siberia died from pneumonic plague after reportedly breaking a test tube containing the live bacteria, putting nearly 200 contacts under medical observation. The immediate split lands on whether that broken-test-tube story holds up—the official statements are conspicuously vague and the reporting leans on unverified Telegram posts, so a strong camp argues the narrative is still unproven. A fierce cultural argument erupts over whether this is a uniquely Russian safety-culture catastrophe or just a universal human problem, with the UK smallpox lab death and the Wuhan COVID leak pulled in as evidence that every country drops the ball. Below the geopolitics, the sharpest criticism hits the practical details: why glass test tubes are still standard at this biosafety level, how narrow the window for effective post-exposure antibiotics actually is, and the sobering reality that bubonic plague is already endemic in US prairie dogs, making the entire “lab leak” angle feel like a sideshow to a much bigger, overlooked threat.
A browser-native classic Visual Basic VB6 IDE [comments]
190 points · 66 comments · wieslawsoltes.github.io · 13h ago
This is a browser-native recreation of the classic VB6 IDE that can load real `.vbp` project files and run the forms directly in the browser. The thread split hard over whether the project was a genuine labor of love or an LLM-generated wrapper, citing the developer's impossibly prolific portfolio of similar retro recreations and a commit history that looks more automated than human. Serious technical pushback focused on the missing Win32 API, the total lack of debugging, the HTML-only export, and the faux IntelliSense that doesn't match the original context-aware help. The conversation deepened into a eulogy for the lost RAD paradigm, arguing the property grid and instant feedback loop were peak DX and wondering if AI agents could ever bring that immediacy back. A deeper wormhole traced VB6's BASIC lineage straight into production nightmares like Sage 100, still running on 1970s MAI Basic Four code under modern systems.
Powerless F1 drivers frustrated by Bahrain F1 software glitch [comments]
183 points · 146 comments · www.motorsport.com · 6h ago
The article covers the Bahrain GP, which was derailed by a software glitch that locked the cars into an idle loop during the wet formation lap, forcing a 50-minute delay for the FIA to patch every car. The thread immediately got into the weeds on the technical debate, explaining that the universal firmware exists specifically to police the ban on traction control, and a bug in that very safety/control software brought the whole grid to its knees. Hardly anyone defends the glitch itself, but the comments split sharply over whether this proves F1 has become a tech overreach nightmare as the drivers claim, or simply an unforgivable software quality failure in a system that serves a genuine purpose. An unexpected tangent spun out of the fact that the FIA patched the cars during the red flag, with someone joking they "vibe coded" a fix in an hour, which opened up a meta-discussion about treating car safety firmware like a web app deployment. Regardless of where you fall on the engineering vs. driver skill spectrum, the consensus is that letting an FIA server cut the throttle on a 200mph race car should never be an option, and that the whole firmware layer needs a hard boundary.
In Ukraine, distributed renewables foil Russia's assaults [comments]
167 points · 192 comments · energytransition.org · 23h ago
The article covers how Ukraine is using distributed solar and battery systems to keep critical water pumps running despite Russia systematically targeting centralized power plants. The HN thread was mostly a battle over whether this is a genuine resilience win or a cost-imposition trap, with one side arguing that forcing Ukraine into an expensive decentralized maintenance premium is exactly Russia's game, and the article is naive for ignoring that. Direct pushback pointed to UN-documented strikes on kindergartens and shopping centers as evidence that Russia's strategy isn't a cool calculus of budget drains but deliberate terror against civilians, which the counter-arguer dismissed as irrelevant to the hypothetical economics of the grid. The thread then swerved into a surprisingly deep meta-debate about how to use hypotheticals in arguments, sidelining any deep technical discussion of Ukraine's energy transition but perfectly capturing the fundamental split between rational-actor analysis and atrocity-focused perspectives on the war.
Show HN: AI search for every photo and every frame of video on macOS [comments]
151 points · 67 comments · github.com · 22h ago
SCM wants to be a local, AI-powered search engine for every frame of video and every photo on your Mac, letting you describe memories in plain language. People immediately grilled the choice of Tesseract for OCR instead of Apple's built-in Vision framework, which is vastly faster and more accurate on Apple Silicon. This suspicion quickly snowballed into a broader debate over whether the whole project was "vibe coded" with an LLM, with one side arguing it produces slop that domain experts can spot instantly and the other side countering that iterative LLM development makes software better. A practical contingent tried to steer the conversation toward the math of processing thousands of videos locally, where scene-detection sampling was the only sane path to avoiding days-long indexing. The author defended the Electron/Node stack by pointing to a desire for portability, but largely conceded the criticism was fair and hinted at a native Swift/MLX rewrite.
Self-hosted HTTP tunnels with SSH and Nginx [comments]
129 points · 31 comments · vincent.bernat.ch · 9h ago
The article walks through building a self-hosted ngrok alternative out of stock OpenSSH and nginx, using SSH remote port forwarding and secure link tokens for access control. The core debate was whether this was a beautifully minimal composition of boring tools or a wildly overcomplicated setup that carelessly exposes arbitrary localhost ports, with the author arguing the security module addresses that risk explicitly. Existing projects like sish (a dedicated SSH tunnel server) and iroh (which uses direct p2p connections without port forwarding) were brought up as cleaner alternatives, though the article deliberately stuck with OpenSSH to avoid requiring special server software. The wider discussion questioned the premise itself, splitting between those who find a plain home server perfectly adequate and those worried the methodology creates fresh security policy nightmares for networks.
Bill Draper has died [comments]
118 points · 37 comments · archive.ph · 19h ago
The NYT obituary for Bill Draper is the story of a pioneer venture capitalist who drove through orchards to knock on doors and backed Skype and OpenTable while famously passing on Apple. The thread seized on that orchard detail, quickly pivoting into a sharp, tech-tinged debate over California's water rights, the export of water disguised as almonds and rice, and the land-use tradeoffs of being the heart of the global tech industry. Personal recollections painted Draper as genuinely humble and decent, a counterpoint to the thread's sharpest split: a debate over the real-world impact of his venture philanthropy, where skeptics were met with detailed Audacious Project and Skoll Award metrics from the foundation's defenders. Beyond the man, the conversation kept circling back to the unique dynasty he anchored—a third-generation venture family—making his death feel like the closing of a specific chapter in the history of concentrated institutional capital in the Valley.
In the wake of Tippett Studios’ closure, a digital archive appears online [comments]
118 points · 14 comments · filmstories.co.uk · 11h ago
An anonymous fan scooped up a binder of 90 CD-ROMs at the liquidation auction of Phil Tippett's shuttered effects studio and dumped the entire contents onto the Internet Archive for everyone. The thread immediately recognized Tippett's legendary status—the "Dinosaur Supervisor" gag from Jurassic Park got its knowing nods—and shifted from mourning the closure to celebrating the thrill of the data rescue. There was a brief, sharp moment of pedantry when someone questioned calling CD-ROM files "digitized," a line of questioning that was met with immediate, exasperated pushback. The real energy, however, was in the urgent consensus that hosting on Archive.org is step one, but mirroring the torrents immediately is step two, because it's only a matter of time before the copyright lawyers realize the auction bypassed them and demand the take-down. The overall vibe was pure gratitude that something destined for private collections stayed public, paired with a sharp awareness of just how fragile that victory really is.
VGHF Digital Archive passes 5000 magazines. Here's what's next [comments]
106 points · 16 comments · gamehistory.org · 22h ago
The Video Game History Foundation’s digital archive just crossed 5,000 scanned game magazines, and they’re marking the milestone by expanding into Japanese publications, starting with the full run of Neo Geo Freak rather than just relying on community scans. The thread immediately swerved past the good news to fight about the archive’s no-download policy, with the preservationist side arguing that a refusal to let people grab the files is a failure of the entire proliferation ethos—a hedge against enshittification or a control tactic, depending on who’s talking. Backlash to that position was equally firm: the VGHF is a properly funded, legally airtight museum with deep industry backing, not some random wiki that might vanish overnight, and its value as a curated, searchable research tool justifies the access model. A sprawling, heartfelt comment from someone who learned English and discovered the wider world through smuggled British gaming magazines in the 80s grounded the whole fight in exactly why this level of care and permanence matters so much.
Rejection Sensitivity in Gifted and Twice-Exceptional Children [comments]
102 points · 79 comments · teachyourkids.substack.com · 20h ago
The submitted article is a detailed Substack guide framing rejection sensitivity in gifted and twice-exceptional kids as a nervous system intensity problem rather than a character flaw, complete with scripts and book lists for parents. A significant portion of the thread immediately recognized the patterns from their own lives, finding the explanations for perfectionism, catastrophic social slights, and defensive anger validating and practically useful. The major friction came from a rigorous pushback against the article's intellectual foundations: critics pointed out that the author isn't a clinician, the section on "Why It Happens" has zero citations, and the piece presents contested frameworks like Dabrowski's overexcitabilities as settled neuroscience. That skepticism ricocheted into a broader tangent about the history of bad educational science going mainstream, using the whole-language reading debacle as the textbook example, and questioning whether the article's style is just highly polished "vibe writing" that an LLM could easily replicate. The core split effectively came down to people who found the practical framing immediately useful versus people who saw it as a well-written but evidence-light piece of popular psychology building an audience for the author's consulting practice.
ArtCraft Apps – open-source Adobe compatible suite written in Rust [comments]
100 points · 114 comments · getartcraft.com · 9h ago
The linked article presents ArtCraft Apps, an audacious open-source suite written in Rust that aims to replace the entire Adobe creative ecosystem—Photoshop, Illustrator, Premier, Lightroom, and InDesign—with native desktop apps that run locally and avoid cloud lock-in. The HN thread immediately split into a nasty proxy war over "vibe coding," with a significant faction convinced the entire project is AI-slop generated by someone who doesn't understand the code, while the project's creator directly engaged, admitting it's extremely early alpha but passionately arguing that using coding agents is just how modern software gets built. Technical skeptics piled on with specific domain expertise, warning that maintaining something as finicky as correct PDF font subsetting or a SolidWorks-grade boolean geometry kernel takes years of hard human effort far beyond what LLMs can brute-force. A darker tangent emerged warning the creator they're poking a bear with very deep pockets, pointing out the project's pricing page for other AI tools suggests commercial ambitions that undermine the "just a FOSS absolutist" defense. Overall, the thread captured the current AI-du-jour moment perfectly: a clash between genuine excitement for collapsing proprietary software margins and deep weariness over the dismissiveness of established open-source labor.
How to scale intent, quality, and artistry with AI [video] [comments]
82 points · 27 comments · www.youtube.com · 23h ago
This is a video of Stripe’s Katie Dill arguing that AI doesn’t have to produce generic “zombie UI” if teams bake strong design intent and artistry into their use of the tools. HN was sharply split on the premise, with a big chunk of the thread digging into the word “scale” and arguing the whole concept of scaling artistry is a category error. The defenders of the talk pounced on that critique to point out the video is a Trojan horse that actually advises you to *not* scale at all, and instead stay intentional while using AI’s speed for iteration. The most pointed pushback wasn’t aimed at the talk’s idealism but at its economics—taking the time to do careful, high-quality work puts you at a clear speed disadvantage against anyone just hammering out acceptable slop, and the market largely doesn’t reward the distinction. That led into a side debate on whether the industry naturally bifurcates into cheap mass-production and high-touch craft, with the smart bet being to stop trying to win on the scale axis entirely.
Blindsight (Watts Novel) [comments]
79 points · 73 comments · en.wikipedia.org · 15h ago
The linked article details Peter Watts’ *Blindsight*, a 2006 hard sci-fi novel about a crew of transhuman specialists sent to encounter an alien intelligence that is hyper-intelligent but completely lacks inner awareness. The thread was spectacularly polarized: one camp called it a landmark of philosophical SF that eerily predicted the consciousness-vs-intelligence debates around modern LLMs, while another found it heavy-handed and undermined by the pulpy gimmick of a genetically resurrected vampire captain. Defenders pushed back hard that the vampire was thematically essential and scientifically grounded in the author’s appendices, but it clearly remained a dealbreaker for many critics. The dominant tangent ran straight into current AI discourse, with sharp disagreement on whether the book’s non-conscious aliens map neatly onto LLM behavior—and the split ran deep, leaving no consensus on whether it’s a prescient masterpiece or just philosophically flashy.
A 40ms Go garbage collector pause caused by swap [comments]
78 points · 51 comments · frn.sh · 6h ago
A developer found that running swap to absorb memory spikes in a Go service caused their garbage collector's stop-the-world pause to jump from ~50 microseconds to 40 milliseconds, because the kernel evicted the GC's own metadata pages to swap. The thread split sharply between people saying "swap is incompatible with latency-sensitive GC, just disable it" and kernel experts arguing that disabling swap just kicks the paging problem to code pages, making `mlock` the only real fix. A few people with deep GC theory chops jumped in to clarify that this specific failure mode isn't inevitable and pointed to decades of "on-the-fly" GC algorithms that avoid STW pauses entirely, noting that Go deliberately optimizes for throughput over latency. The practical takeaways boiled down to pinning the metadata pages in memory with `mlock`, or, as Discord famously did in their 2020 blog post explaining why they switched from Go to Rust, ditching the runtime entirely for low-latency paths. A brief tangent advocating for reference counting got firmly grounded in decades of evidence that it suffers from the exact same predictability problems (cascading deallocation, concurrency lock contention) that people blame tracing GCs for.
Xray-core concealed a certificate verification bypass vulnerability [comments]
78 points · 10 comments · github.com · 14h ago
The linked article is a full disclosure, chronicling how Xray-core maintainers buried a certificate verification bypass in their `pinnedPeerCertSha256` option for half a year before silently patching it. The thread immediately fixated on the stark hypocrisy of a project that regularly shamed users for the insecurity of `allowInsecure` yet effectively forced them into a subtly broken replacement and then hid the fix. A broader tangent explored who Xray's users really are—people in China, Iran, and Russia staking their connectivity on tunneling through state firewalls, making a silent MITM vulnerability a dangerous breach of trust. The strongest operational consensus was that the core failure wasn't just the software bug but the complete absence of a security advisory, leaving anyone who updated with no way to know if they were still exposed, with the recommended mitigation being to offload TLS handling entirely to a reverse proxy.
Homa: The end of TCP for AI clusters [video] [comments]
70 points · 36 comments · www.youtube.com · 12h ago
John Ousterhout's Homa proposal argues that TCP's byte-stream model and sender-driven congestion control are fundamentally mismatched for the latency-critical, small-message coordination workloads that increasingly dominate AI clusters, and the solution is a clean-slate, receiver-driven, message-oriented transport protocol. The thread split sharply: one camp argued Ousterhout is attacking TCP as a strawman, since QUIC already delivers message boundaries, request multiplexing, and encryption without Homa's inability to detect a fully dropped RPC, while the other camp insisted the whole premise is off-base because real training clusters run on InfiniBand or RoCE where Homa's advantages over the existing complexity are unclear. A lively tangent connected the argument to space-based compute, noting that if Ousterhout is right that intra-rack tail latency is the new bottleneck, putting compute nodes hundreds of kilometers apart makes the problem fundamentally intractable, limiting space datacenters to single-node models. What no one disputed was the underlying diagnosis: the shift from throughput-heavy gradient dumps to latency-sensitive inference coordination forces a reckoning with TCP's deep legacy assumptions, and whether Homa or something else wins, that reckoning isn't going away.
All I wanted was a custom domain email [comments]
69 points · 101 comments · jacobg.co · 14h ago
The post walks through a developer’s slog to get a cheap custom domain email working with Apple Mail, ultimately landing on Foundermail. Obvious recommendations like iCloud+ and Fastmail got shouted down by those who've dealt with iCloud silently deleting inbound messages or the per-user costs that violate the "unlimited addresses" condition. Self-hosting drove the fiercest divide, with one camp swearing deliverability is easy with proper records and the other countering that getting through to Gmail from your own server is a constant, losing fight against the big providers. FounderMail’s creator showed up to defend the service's new-address approval gate, but the gesture backfired when people noticed the entire website was AI-generated—a hilariously ironic look for a tool whose whole pitch is that it makes it impossible for a bot to operate.
The evolution of effective altruism [comments]
68 points · 182 comments · www.economist.com · 18h ago
The Economist piece tracks how effective altruism evolved from a fringe do-gooder philosophy into a world-conquering movement, while warning that its longtermist wing's obsession with AI might actually steer humanity toward the extinction it fears. The HN crowd was deeply split, but the loudest argument treated EA as a moral fig leaf for the tech elite to accumulate power with "patient philanthropy" as a convenient excuse to hoard wealth indefinitely, a critique fueled almost entirely by the FTX collapse and the perception that leaders like Will MacAskill escaped accountability while sitting on mansions. Defenders pushed back hard, insisting this caricature ignores the billions flowing to malaria nets and cash transfers, and that the controversial "earning to give" and AI cultism are a loudly vocal minority, not the core evidence-based giving community. A separate but persistent line of argument rejected the "social movement" label entirely, calling it a capital-driven sect or an unelected power bloc that democracy should be taxing into irrelevance, not platforming for its vibes. The thread's real fight wasn't over the article's thesis, but over whether the movement's failures are baked into its ideology or just human nature running amok inside a very wealthy, very online clique.
Anthropic reported diary entry to police, woman faces felony charge [comments]
62 points · 55 comments · www.techspot.com · 2h ago
A Florida woman is facing felony charges after a human reviewer at Anthropic flagged her Claude diary entries describing a plan to shoot up a sheriff's office. The reaction immediately split between seeing this as a justified safety intervention and a massive red flag for AI-empowered thought policing. A lot of the pushback zeroed in on the Florida statute's specific wording—"another person may view it"—questioning whether a private chat log reviewed internally by a contractor qualifies as an intentional public threat. The discussion also cast this as an impossible bind for closed-source AI companies, damned for reporting this while getting sued for *not* reporting similar threats, proving that any claim of privacy is contingent on human review. The resounding consensus was a blunt warning: unless you're running a local open-weight model, assume everything you type to a big AI chatbot is being read by a human, effectively ending the era of a truly private digital diary.
Incentives in Academic Research [comments]
48 points · 44 comments · www.msoos.org · 14h ago
The linked article from msoos blog argues that academic research incentives have drifted so far that researchers no longer care to correct known errors in their published work, because the rewards—promotions, fame, grants—have already been collected. The HN discussion largely accepted this diagnosis but split on the cause: some pointed to Goodhart's law, where proxy metrics like citation counts have become the actual goal and are now gamed precisely because that's what gets rewarded, while others took a more cynical view that academia was never a pure pursuit of truth but always a human system shaped by funding sources, institutional politics, and prestige-seeking. A handful of participants pushed back, claiming that modern tools like blogs and AI replication actually increase scrutiny and correctness, but that was met with examples of bogus machine learning papers that sail through top conferences and never get retracted. An unexpected thread emerged debating whether uncorrected sloppiness essentially constitutes fraud—if you stand behind a paper despite knowing it's wrong, it's hard to distinguish from outright dishonesty—and a parallel was drawn to software engineering, where deploying a quick "fix" that later breaks everything gets rewarded in the moment while the long-term consequences fall on someone else.
Replacement of petroleum based products with plant-based materials (2025) [comments]
48 points · 18 comments · onlinelibrary.wiley.com · 2h ago
The linked article wasn't available to this summarizer; from the discussion, it's an academic review of replacing petroleum-based fuels and plastics with plant-based materials. The HN thread immediately vaulted past the paper's specifics into a raw debate about the political economy of a post-petroleum world. A deep split formed between those who see consumer greed and the corporate profit drive as the root problem demanding a cultural shift, and those who counter that demand is infinite, business isn't a monolith, and the real bottleneck is finding technically viable polymer substitutes people will actually tolerate. Domain-aware voices consistently steered the conversation away from biofuels—widely written off as an inefficient, subsidy-fueled greenwashing scheme—toward the intrinsically harder problem of plant-based plastics, highlighting successes like nylon 11 derived from castor oil. The paper itself was dismissed as lacking depth, but the argument it sparked perfectly crystallized the central tension between betting on technical substitution and demanding a systemic reduction in consumption.
Google Japan shows off conveyor-belt keyboard with keys that move to fingers [comments]
37 points · 12 comments · www.tomshardware.com · 13h ago
Google Japan’s latest fully open-source gag is a conveyor-belt keyboard that slides keys to your stationary fingertips so your arm never needs to cross the desk. The HN thread ditched the Tom’s Hardware article for the actual Google blog and GitHub repo, treating the open-source nature as the real headline rather than the novelty of moving keys. Someone framed it as *chindogu*—the Japanese art of elaborate uselessness—which instantly reset expectations from “is this practical?” to “how mechanically cool is this?”. The only real tension was between people enjoying the ballet of keys and the ergonomics crowd who actually build Japanese split keyboards like the Corne, who argued the wrist posture in the demo looks terrible and that this sits firmly in art-project territory. The most surprising tangent was the observation that Japanese speech-to-text still isn’t good enough to be the obvious alternative, which is quietly the only reason this physical hack feels necessary rather than purely absurd.
Show HN: Build with Python – a beginner course where your code draws [comments]
35 points · 12 comments · scimigo.com · 13h ago
Based on his daughter’s struggles, the creator built a browser-based Python course where code draws pictures, removing setup friction and giving immediate visual feedback. The thread quickly flagged that the first exercise was already complete (a leftover demo), which the creator fixed on the spot, but then the debate shifted to the $49 price, with several people calling it AI slop and a grift, while others dismissed pricing gripes from non-customers. Alongside the pricing fight came detailed pedagogical critiques about poorly scaffolded lessons and unhelpful hints, with CMU’s free CS Academy held up as a better example. The creator admitted heavy AI use but said the content came from his teaching experience, and the thread ended up sharply divided on whether the course offers real value or not.
I asked Claude build a physically accurate O'Neill cylinder you can walk around [comments]
30 points · 29 comments · island-three.gruberbuilds.workers.dev · 12h ago
This is a WebGPU simulation of an O'Neill cylinder that an AI model generated, promising a walkable, physically accurate space habitat in your browser. The thread was a battlefield between the ten minutes of genuine awe and the overwhelming wave of crashes, single-digit frame rates, and hard browser lockups across Mac, Windows, and Linux that made the experience unworkable for most people. The small group who got it running smoothly argued the AI-built NPCs and simulated physics made it a genuine breakthrough demo, but critics countered hard that the experience looked decades old, ignored established WebGPU incompatibilities, and featured clouds that sparked a whole physics sideline about whether they made sense in the habitat. The real tension wasn't about the glitches themselves, though—it was about the sales pitch gap between "walk around a cylinder" and a site that mostly generated error messages, which some argued actively harms the credibility of AI-generated experiences.
Protest against housing crisis in Spain [comments]
29 points · 55 comments · www.theguardian.com · 23h ago
The Guardian piece covers the massive protests in Spain triggered by the eviction of an 87-year-old woman and the opposition blocking the government's housing decrees. The discussion immediately fractured over whether the answer is simply to build more housing—with one side arguing the protesters are just protecting their own property values—or whether that’s a naive supply-side fantasy in a country absorbing 500,000 immigrants while building barely 90,000 homes a year. People familiar with the Spanish market kept pointing out that any new units just get scooped up by investment funds and turned into Airbnbs anyway, making regulation and rent control look like the only real check on the system. Meanwhile, a strong strain of political cynicism emerged, arguing that Sánchez has been in power for nearly a decade and these protests feel conveniently timed to justify calling a snap election he can blame on his rivals. One of the sharpest points cut through all the policy noise by asking whether the evicted woman, who had been paying a pittance for a prime Madrid apartment for decades, is really the fair housing martyr everyone wants her to be—a question that exposed just how intractable this problem really is.
RuneScape's Position on Gen AI [comments]
28 points · 66 comments · www.reddit.com · 16h ago
Jagex, the studio behind RuneScape, acknowledged on Reddit that their Dragon Slayer II trailer used Gen AI art from a trusted partner who had explicitly confirmed it wouldn't, complete with a six-fingered hand that made the lie impossible to ignore. The Hacker News thread split hard on whether a total ban on Gen AI in game development is realistic, with one camp arguing it's already standard for code pipelines and economically naive to reject it outright, while the other side saw this as more evidence of vendors poisoning the supply chain by sneaking in slop against contracts. A recurring focus was the weird double standard that gives code a pass for being indirect but draws a firm line at art because it's directly exposed to players, even though both are part of the craft. The discussion also pivoted to whether this intense anti-AI sentiment is a Western phenomenon, pointing out that Asian gaming communities barely blinked at similar Gen AI use in recent trailers.
Generated 2026-10-05 08:14 UTC
Generated by Sauron from Hacker News discussions and linked articles.