HN Brief: 2026-07-04
Today’s Hacker News was split between a satirical startup parable that had everyone nodding in painful recognition and a quieter but persistent thread of skepticism about AI coding tools—whether it’s the cost of running huge models locally, the hype around confidence theater, or the realization that feeding an agent its own history doesn’t help. Alongside that, a surprising number of stories turned on physical infrastructure: Costco’s in-person retail model, factories as just rooms, a valve open-source e-ink faceplate, and the terrifying return of the screwworm.
The threads most worth your time: "Half-Baked Product" for the cathartic-but-contested look at startup dysfunction that split the room; "Valve open-source the Steam Machine e-ink screen" for a rare example of corporate hardware generosity and the community’s glee; "Costco is the anti-Amazon" for a sprawling, affectionate debate about retail efficiency and cultural quirks; "Espionage Against the European Parliament" because hacking a sitting investigator of spyware is a blatant act of suppression; and "The Fall and Rise of Screwworm" for a horrifyingly fascinating story of a flesh-eating parasite’s return after a failed containment barrier.
Half-Baked Product [comments]
1253 points · 376 comments · weli.dev · 23h ago
The article is a satirical parable about a startup building an oven with a broken algorithm, where the founder over-promises to VCs and a big customer, the engineer caves to feature requests from sales, and the whole thing collapses under the weight of technical debt and misaligned incentives. HN mostly ate it up — a ton of people chimed in saying it’s “uncomfortably accurate” and that they’ve lived that exact story as the engineer or Mario, with several calling it the best blog post they’ve read. But there was real pushback too: a vocal minority argued the piece is just engineer-worship that glosses over real complexity, that it panders to the HN audience by painting a one-sided “businesspeople bad, engineers good” narrative, and that Tesla’s long game shows hype can work if you have enough capital. The split was essentially between readers who found it cathartic truth-telling and those who found it a shallow caricature that ignores how some startups actually succeed through the same messy dynamics.
Valve open-source the Steam Machine e-ink screen so you can make your own [comments]
553 points · 101 comments · www.gamingonlinux.com · 19h ago
Valve open-sourced the e-ink faceplate for the Steam Machine – they’re calling it the “Inkterface” – so anyone can build their own using off-the-shelf Adafruit parts and an ESP32 over Bluetooth. The thread quickly got into the weeds on e-ink refresh rates: one former reMarkable firmware engineer explained that while full refreshes take about four seconds, partial updates can go much faster, but you need periodic maintenance cycles to avoid permanent ghosting or damage. A lot of people wished Valve had just included the screen in the base model, though others pushed back hard, arguing component costs already justify the $1,049 price and that the swappable, non-invasive design (battery-powered, Bluetooth, no modding needed) is the next best thing. There was also a split over the Steam Machine’s value versus building your own PC – the consensus being you can’t match its size, quietness, and features like HDMI-CEC and controller wake-from-sleep for the same money, even if raw performance per dollar is lower. Overall, the community is genuinely thrilled Valve handed the community the specs rather than locking it away, even if they’re skeptical about third-party vendors actually delivering pre-built versions.
Costco is the anti-Amazon [comments]
391 points · 366 comments · phenomenalworld.org · 16h ago
The article argues that Costco’s deliberately limited selection, in-person shopping, and high-wage model represent a smarter and more socially responsible alternative to Amazon’s endless-assortment, doorstep-delivery model. HN largely bought the thesis, but the discussion quickly turned into a sprawling, affectionate chest-thump about Costco’s cultural quirks and operational details—the $1.50 hot dog price that hasn’t changed in 40 years, the 6% turnover rate, and the comparison to Buc-ee’s as a purer distillation of American capitalism. Several people pushed back on the logistics comparison, noting that Amazon delivery vans consolidate dozens of stops per route, making last-mile delivery potentially more efficient per item than individual car trips to a warehouse. Others pointed out that Costco and Amazon barely compete head-to-head—Costco is for bulk groceries and household goods, while Amazon dominates for one-off items, small electronics, and anything you can’t find in a store with only 4,000 SKUs. A recurring side tangent was the surprise that e-commerce is still under 17% of U.S. retail, with commenters breaking down how cars, gas, groceries, and clothing keep people in physical stores.
Espionage Against the European Parliament [comments]
350 points · 86 comments · citizenlab.ca · 11h ago
Citizen Lab published evidence that Stelios Kouloglou, a former MEP who served on the European Parliament’s committee investigating Pegasus spyware abuses, was himself infected with Pegasus twice during key moments of the committee’s work — once while in a Greek hospital meeting a fellow targeted journalist. The thread quickly veered away from the specific findings into a broader fight about EU surveillance and American tech dominance, with one side arguing that EU dependence on US platforms and GDPR’s compliance burden have crippled local alternatives, while others pushed back hard that this is a transparent attempt to redirect blame from the actual spyware scandal. A major subthread zeroed in on the detail that Kouloglou apparently missed or forgot Apple’s in-app threat notifications, leading to debate over whether those warnings are too easy to dismiss, could have been suppressed by the malware, or whether he simply lied to avoid embarrassment — and whether Apple should maintain a persistent notification log. Some commenters dismissed the whole thing as just another day in global espionage where everyone spies on everyone, but others countered that hacking a sitting committee member actively investigating the spyware is a blatant act of obstruction that should dwarf the usual geopolitical shrugs.
Jamesob's guide to running SOTA LLMs locally [comments]
333 points · 150 comments · github.com · 17h ago
The linked guide documents one person's extreme DIY build to run large language models (like a 594B-parameter GLM model) on four NVIDIA RTX PRO 6000 GPUs with 384GB of VRAM using a custom PCIe switch to get low-latency peer-to-peer communication between the cards. The Hacker News thread mostly orbited around the gap between that $40k setup and more accessible options, with lots of pushback on whether the build makes practical sense. Several people argued that dual RTX 3090s for about $2k actually match the key bottleneck (memory bandwidth) and give you fast enough token generation for useful work, while dismissing MacBooks as having far too little bandwidth for the price. Others countered that for many tasks, a single 24GB GPU at 4-bit quantization works fine, and that the MEASURE of "SOTA" is misleading here — the expensive "lobotomized" large model may not actually outperform a smaller, higher-bitrate one, especially for coding, and context window gets uncomfortably tight. A persistent sub-thread debated whether cloud hosting is just cheaper and more practical, but the local crowd pushed back hard on that too, arguing that the freedom from sharing data, token limits, and vendor lock-in makes even a slow local model more valuable in practice.
Alibaba to ban Claude Code in workplace over alleged backdoor risks, source says [comments]
324 points · 274 comments · www.reuters.com · 23h ago
The article reports that Alibaba is banning Claude Code internally over alleged backdoor risks, but the thread quickly pivots to questioning how Alibaba engineers can even use Claude in the first place, given that Anthropic blocks access from China — the answer being that the CLI tool works with any model and that Alibaba has global offices in Singapore, Hong Kong, and North America. The "backdoor" claim gets heavy pushback: many point out that the recent controversy was about Claude Code sending data like timezone info without explicit approval, which is sloppy and hostile but not a true backdoor, while others argue that Anthropic remotely pushing code without user consent crosses a line regardless of the label. A major split emerges over intellectual property — some defend Alibaba's model distillation as fair use of publicly available outputs, while others call it Chinese reverse engineering that disrespects Anthropic's investment in RLHF and safety research, leading to a broader ideological fight about whether IP law hinders progress or whether the CCP's history of economic mismanagement makes it untrustworthy with superintelligence. The whole thread is less about the ban itself and more about the tangled web of corporate espionage, VPN evasion, and the moral double standard where both sides accuse each other of stealing.
Wordgard: In-browser rich-text editor from the creator of ProseMirror [comments]
294 points · 97 comments · wordgard.net · 23h ago
Wordgard is a new open-source rich-text editor from Marijn Haverbeke, the creator of ProseMirror, built as a schema-based, modular alternative with a stronger API and better collaborative editing support. The thread immediately split over why this needed to exist as a separate project rather than a ProseMirror 2.0 — the author explained that incompatible interface changes would create confusion, and that he had a pile of new design insights that justified a clean break. A lot of commenters dug into the React interop pain points, with one side arguing that React’s DOM ownership model is the real constraint, while others pointed out that Wordgard’s transaction-based updates should eliminate the weird redraws that plague ProseMirror in React apps. Mobile bugs got aired on Android Chrome and Firefox, with reproducible crashes reported, but the author was responsive and filed issues. There was also a side conversation about the self-hosted code repository being down for some, which the author defended as a move away from GitHub’s Copilot trajectory and sanctions-based blocking, though a few pushed back that US sanctions law isn’t optional.
60% Fable cost cut by converting code to images and having the model OCR it [comments]
263 points · 89 comments · github.com · 16h ago
The submission is a tool called pxpipe that cuts Claude Code costs by up to 60% by converting bulky text (system prompts, tool docs, history) into PNG images, exploiting the fact that image tokens are priced by pixel dimensions rather than character count, so dense content like code packs far more characters per token as an image. HN immediately split on whether this is a clever compression technique or a pricing arbitrage loophole that Anthropic will inevitably close—some argued it’s a genuine insight from DeepSeek’s research on visual token embeddings being more efficient than text tokens, while others countered that it’s lossy, that the model misreads hex strings and confabulates, and that the real cost is just shifted to the OCR compute on the model’s side. Several commenters pointed out the readme’s honest but extensive caveats about exact-recall failures and the need for a verbatim-risk guard, and there was a strong side thread criticizing the LLM-generated, overly verbose prose of the README itself, calling it hard to parse and inconsiderate of readers. The consensus seemed to be that this is a neat hack for price-sensitive, non-byte-exact workloads, but not a free lunch—accuracy drops on tasks requiring exact recall, and the pricing gap may not last.
Zuckerberg 'Admits' Meta's Layoffs Were Ineffective [comments]
236 points · 217 comments · eshumarneedi.com · 19h ago
The article is a critique of Zuckerberg’s acknowledgment that Meta’s mass layoffs, driven by a bet on AI agents replacing human programmers, didn’t pay off as expected. The HN thread mostly sidestepped the article’s specific timeline and instead turned into a broader referendum on whether Zuckerberg has ever successfully built anything new since Facebook’s early days—with many pointing out that Instagram, WhatsApp, and Oculus were all acquisitions, and that Reels is a TikTok clone. A strong split emerged: one side argued that Meta’s core ad business is a cash machine, so “failure” only matters from an innovation lens, while the other side countered that the company’s dominance came from buying or crushing competitors, not from original vision. Several commenters pushed back on the article’s “dimwitted or evil” framing, arguing that Zuckerberg is neither—he’s simply a 22-year-old who never grew up, insulated by wealth and surrounded by yes-men, which explains his vibes-based management and inability to see beyond the next panic. Another thread debated whether AI progress is truly inevitable, with some calling it a cargo-cult justification for reckless CEO moves.
Factories are just rooms [comments]
232 points · 95 comments · interconnected.org · 16h ago
The article is Matt Webb's account of giving a talk to 7-year-olds about manufacturing, arguing that factories shouldn't inspire awe but instead should feel accessible — "factories are just rooms." HN ran with this in two directions: a bunch of people loved the demystification angle, saying that showing kids the actual messy process of making stuff is more powerful than the slick "thousand products a second" videos. But a much longer thread veered hard into a detailed argument about Shenzhen's manufacturing ecosystem — people who've actually worked there described how garage-sized shops coordinate through sourcing agents to build bespoke megawatt-scale electric motors in days, while others pushed back hard that this only works at tiny scale and that quality control is a nightmare once you try to grow. There was also a split over whether the "anyone can do it" message actually works on kids, with one thoughtful take saying that "boring" is the real killer for that age group — making something feel too easy can be just as demotivating as making it feel impossible.
Please stop the AI confidence theater [comments]
228 points · 244 comments · www.elenaverna.com · 19h ago
The article is a plea to stop the "AI confidence theater" — the exaggerated claims that AI is life-changing, when most people’s real use cases are just saving 15 minutes on email summaries. The HN thread largely agreed with the critique, but many pointed out the irony that the post itself is sponsored by an AI scraping tool (Firecrawl) and the author promotes their own AI double, which undercuts the argument. Several commenters pushed back on the idea that marketing, finance, and HR were "spared" from AI job impacts, noting they’ve been gutted too. A strong thread argued that the real problem isn’t just hype but the endless cycle of grifters moving from crypto to AI to quantum computing, and that the only fix is to stop amplifying the noise. Some shared genuinely useful but modest AI wins — side projects, faster debugging, quick electronics advice — that don’t need the "life-changing" label to be valuable.
Markets are competitive if and only if P != NP [comments]
221 points · 152 comments · arxiv.org · 16h ago
The linked arxiv paper argues that competitive markets require P != NP, because if firms had the computational power to solve hard problems, they could efficiently detect and punish collusion, making it stable — but that also means the widely believed P != NP is what protects competition. HN immediately pointed out that the site's title mangled the core claim (it read "P = NP"), and then spent most of the thread debating whether the paper's assumptions hold in practice. Several commenters pushed back on the idea that detecting secret price-cutting is the main barrier to collusion, arguing that in real markets the real problem is firms having no effective way to punish defectors even when they know they're being undercut. Others brought up the RealPage rent-fixing scandal to argue that algorithmic collusion doesn't need superhuman compute — it just requires shared information and a fig leaf. The thread was split between those taking the paper as a neat theoretical curiosity and those dismissing its "spherical market in a vacuum" framing as too detached from how actual pricing works.
Performance per dollar is getting faster and cheaper [comments]
219 points · 68 comments · www.wafer.ai · 10h ago
This post from Wafer claims they’ve served GLM-5.2 on AMD MI355X GPUs at 2,626 tokens per second per node and 213 tok/s single stream, all at over 2x lower cost than NVIDIA’s Blackwell — thanks to aggressive MXFP4 quantization and custom tuning on AMD’s ROCm stack. The HN crowd immediately flagged that the “lossless” quantization claim doesn’t hold up: they show small accuracy drops (GPQA down 1.9%, GSM8K down 1%) and multiple people argued that FP4 is functionally lobotomized in practice, not close to frontier quality. The performance-per-dollar framing also drew pushback on performance-per-watt — AMD’s MI355X pulls 1,400W versus B200’s 1,200W, and real datacenter cooling and power constraints make that a meaningful gap. A few folks dug into the benchmarks to note the 2,600 tok/s is aggregate across many concurrent streams, not per-user throughput, and that the numbers rely heavily on a 60% cache hit rate and speculative decode workarounds that required patching ROCm — underlining that AMD’s software story still lags despite this clever engineering.
Holes [comments]
214 points · 37 comments · xkcd.com · 14h ago
The linked article wasn't available to this summarizer; from the discussion, it's the latest XKCD comic cataloguing famous holes around the world (mines, boreholes, sinkholes, caves) with "oops" tags for the ones that caused disasters. HN immediately jumped in to share the real-world videos and Wikipedia rabbit holes behind each entry — the Pantai Remis mine collapse, Lake Peigneur being drained into a salt dome, the terrifyingly deep Mponeng gold mine, and the Kola Superdeep Borehole. A strong split emerged over whether the Kola hole counts as "massive" (it's just a narrow drill hole) versus Mponeng, which is a vast commercial mine with thousands of workers. People also corrected the comic's implication that the deepest boreholes are exactly ~12 km by pointing out that Al Shaheen's 12km record is horizontal length, not depth, and debated whether that depth coincidence is meaningful or just coincidence. The thread ended up less about the comic itself and more about people enthusiastically one-upping each other with obscure geological trivia, like Lake Baikal's 7 km of sediment sitting on a continental rift, or the salt mine under Detroit.
Leanstral 1.5: Proof abundance for all [comments]
204 points · 46 comments · mistral.ai · 9h ago
Mistral released Leanstral 1.5, a small 6B-parameter open-weight model that specializes in formal proof engineering for Lean 4, claiming it saturates benchmarks like miniF2F and uncovered five previously unknown bugs in open-source repositories. The HN thread zeroed in on that bug-finding example—a simple overflow in a varinteger library’s sign function—and called it marketing fluff, since property-based testing and fuzzing would catch that edge case in seconds, and the bug had already been filed by someone else a week prior. Several people ran quick comparisons showing GPT-5.5 and even Claude Code find the same bug trivially, though supporters countered that the real point is the model’s tiny size (fits on commodity hardware) and the method—automatically proving the absence of bugs, not just finding one. A tangent emerged around whether Lean is the right tool for software verification versus Isabelle or Dafny, and a few commenters debated Europe’s AI competitiveness, with one arguing this release proves European labs can still punch above their weight in niche domains.
SearXNG: A free internet metasearch engine [comments]
201 points · 52 comments · github.com · 11h ago
SearXNG is a free metasearch engine that aggregates results from DuckDuckGo, Google, Brave, and others while promising not to track or profile users. The HN crowd is broadly enthusiastic—many have been self-hosting it for years and prefer it over Google or DDG, especially for privacy—but the complaints are consistent: engines regularly break or get blocked (Google stopped working a few days ago, others throw CAPTCHAs), it’s slower than using a single engine, and the results can be worse. A major tangent came from the original creator of Searx, who dropped in to promote his new project Hister, a full-text indexer of your own browsing history, and sparked a whole separate debate about whether searching your past visits replaces a metasearch engine. The thread also dug into using SearXNG with AI agents—several people detailed how to wire it up via MCP servers, though there’s friction over missing built-in MCP support and the need for JSON API workarounds.
I Wasn't Allowed Prompting ChatGPT During My Chalk Talk: This Is Discrimination (2025) [comments]
192 points · 93 comments · inpreparation.substack.com · 14h ago
The piece is a satire from an academic parody site, depicting a postdoc who argues that being barred from prompting ChatGPT during a chalk talk is discrimination because she does 85% of her science by editing LLM output. The HN discussion immediately split: some readers didn’t catch the satire and took the article at face value, leading to a long meta-debate about Poe’s Law and whether text-based humor is becoming impossible to signal reliably online. Others recognized the satire but argued it’s uncomfortably close to real attitudes they see in software engineering and academia, where people treat AI-generated content as their own work. A smaller faction pushed back on the premise entirely, insisting that even if AI is a useful tool, the chalk talk exists to test whether the researcher actually understands their domain well enough to steer the model effectively.
Giant trees have no trouble pumping water to top branches: new research [comments]
182 points · 89 comments · news.exeter.ac.uk · 9h ago
The new study claims that the world’s tallest tropical trees have evolved water-transport adaptations that fully compensate for height, so they aren’t actually more drought-vulnerable than shorter ones. HN immediately zeroed in on the wording “pumping” — a good chunk of the thread is a physics debate about whether trees really pump (they don’t; it’s transpiration-driven capillary tension) and whether the 10-meter suction limit of a vacuum would even apply here. Several people pushed back hard, pointing out that if water transport were no issue at all, we’d expect trees way taller than the observed ~130-meter cap, so something else — compressive strength or something biomechanical — must be the real limiter. Others brought up that many actual giant trees, especially coast redwoods, rely on fog and moss to get water to the top, suggesting the study’s “no trouble” claim might only hold for rainforest dipterocarps in humid conditions.
Memorizing session transcripts isn't useful [comments]
177 points · 153 comments · 12gramsofcarbon.com · 16h ago
The article argues that feeding AI coding agents their own past session transcripts doesn't improve performance — artifacts like docs, commit messages, and PR descriptions are what actually matter, and indexing transcripts just adds noise and token waste. The HN thread overwhelmingly agrees, with many saying they've ditched memory systems entirely in favor of a simple PLAN.md or AGENTS.md file that the agent reads and updates as a save state. A few people push back, arguing that engineered context still helps smaller models or that a routing layer for injecting information at the right time could be valuable. The discussion also spins into broader skepticism about the whole AI-coding-product ecosystem, with one commenter calling the author's story "influencer spam" and wondering why anyone should care about someone who no longer writes code by hand complaining about the prediction machine they handed their job to.
PostgreSQL and the OOM killer: Why we use strict memory overcommit [comments]
167 points · 99 comments · www.ubicloud.com · 19h ago
The article argues that PostgreSQL should run with Linux's strict memory overcommit (vm.overcommit_memory=2) to avoid the OOM killer corrupting shared memory and forcing a full crash recovery — and they even found a kernel bug that inflated the commit counter by 648GB on an 8GB machine. The thread quickly split: several experienced operators warned that mode 2 can break apps like Go runtimes that reserve huge virtual memory, causing random ENOMEM failures instead, and recommended testing aggressively or using oom_score_adj and careful ratios instead. Others pushed back hard, saying disabling overcommit just trades a rare OOM for daily crashes in programs that can't handle allocation failures (with PostgreSQL being a notable exception that handles ENOMEM gracefully). A long sidebar emerged comparing Windows (which never overcommits and kills the right process) to Linux's "random" OOM killer, with multiple people insisting Windows's default is actually stricter and more predictable. The consensus felt like: strict overcommit is great *if* you control every process on the box and they all handle allocation failures — otherwise you're better off isolating PostgreSQL on its own machine and leaving the kernel's heuristic alone.
International chess federation sanctions Kramnik [comments]
151 points · 88 comments · www.fide.com · 15h ago
FIDE officially banned former world champion Vladimir Kramnik for two years (one year active) after his ethics commission found he violated multiple codes by publicly accusing players like Daniel Naroditsky and David Navara of cheating without evidence. The thread immediately filled in the context the article danced around: Kramnik has a years-long pattern of losing online blitz games to younger players and then publicly declaring they must be using computer assistance, based on laughably bad statistical reasoning that a professor of mathematical statistics from Waterloo publicly demolished. The discussion pivoted hard to the tragic death of Daniel Naroditsky, who openly talked about the psychological toll Kramnik's harassment took on him before dying from a combination of heart issues and drugs—many in the thread squarely blame Kramnik for pushing him to that state, though some push back that the official cause wasn't ruled suicide. A running side theme was how top chess grandmasters often have zero domain knowledge outside chess, with people sharing clips of them flubbing basic trivia to explain why Kramnik genuinely believes his own nonsense. The consensus was that the ban is a slap on the wrist but at least establishes a precedent that publicly weaponizing cheating accusations has consequences.
Africans Are Turning to Starlink [comments]
149 points · 181 comments · www.economist.com · 10h ago
The Economist article looks at how Starlink is becoming a pricey but necessary stopgap for internet in Africa—Nigeria's Ekiti state, for example, is using it because running fibre up hills from Lagos is too expensive and mobile networks are overloaded. The HN thread quickly turned into a comparison with rural America, where people chimed in that they face the exact same calculus: no wired options, cellular hotspots that are slow and expensive, and Starlink at $55/mo that actually works. A big split emerged over whether fibre theft is a real obstacle in Africa—some argued thieves cut cables hoping they're copper and destroy fibre, others pushed back that once they realize it's worthless they stop, and one commenter pointed out that failed social institutions, not technical challenges, are the root cause. The technical crowd dove deep into Starlink's architecture: whether inter-satellite laser links are actually used for consumer traffic (apparently not yet, at least on land), how many ground stations Africa really has (only a handful), and whether the economics of launching thousands of satellites actually beats running fibre 20 km to a single house.
The Fall and Rise of Screwworm [comments]
143 points · 59 comments · www.construction-physics.com · 19h ago
The article traces the history of the screwworm, a flesh-eating livestock parasite that was eradicated from North America using a sterile-male fly factory, and explains how the barrier at the Darien Gap failed around 2023, allowing the pest to march back into the US. The thread quickly zeroes in on why the eradication program was never pushed into South America—the consensus is that the Amazon’s size and lack of infrastructure made it prohibitively expensive, though one commenter pointed out that opportunity ended around 2010, and another argued that leaving it as a half-measure was a mistake that only needed one defunding event to unravel. Several people pushed back against the idea that modern insecticides could handle the problem, with the original doubter correcting themselves after reading the old USDA reports. The discussion also split on the role of recent DOGE cuts: some argued the cuts made a bad situation worse, while others noted the outbreak started in 2023, well before those cuts, so the real failure was letting the barrier program atrophy over years of success. A few tangents veered into the surreal horror of the fly factory itself—described as "a 76,000-square-foot artificial wound"—and the one-in-a-million risk of radiation resistance evolving in the sterile flies, which the thread largely dismissed because the factory population never sees radiation.
FreeBSD ate my RAM [comments]
129 points · 45 comments · crocidb.com · 12h ago
The article is a deep dive into why FreeBSD memory reporting tools like btop and fastfetch show wildly different numbers, tracing the discrepancy to how each tool handles ZFS's ARC cache and the kernel's page queues. The author found that btop was using a 32-bit integer that overflowed on systems with more than 4GB of RAM, making its memory usage numbers completely wrong on modern hardware, and also wasn't accounting for the ARC cache at all. The HN discussion quickly split into two camps: one side appreciated the detailed investigation into memory reporting heuristics and the author's successful patches to btop and fastfetch, while the other side was dismissive, arguing this is just "ZFS cache, the end" and that anyone running an unfamiliar server OS with an enterprise filesystem shouldn't be surprised. A deeper thread emerged about the historical swap reservation differences between FreeBSD and Linux, with one person finally understanding why the old "swap should be 2x RAM" rule existed, and another pointing to the "worse is better" essay to frame the BSD vs. Linux philosophical divide.
America, 1926: A forgotten 100-year-old report [comments]
125 points · 173 comments · www.derekthompson.org · 16h ago
Derek Thompson dug up a forgotten 1,500-page report from 1929 called *Recent Social Trends*, commissioned by Herbert Hoover, and uses it to compare America in 1926 to America in 2026 — finding eerie parallels like technology-driven anxiety over "mechanical power then vs. AI now," fears of immigration, and a stock market boom sitting on top of widespread unease. The thread immediately latched onto the cyclical-history angle, with many pointing out that the "typical American" described in the piece — a 26-year-old in 1926 — would go on to live through the Great Depression and World War II, prompting a debate over whether that actually maps onto today's generational fate. A big split emerged around Strauss-Howe generational theory: some see it as a useful explanatory framework for why we feel like we're heading into a crisis, while others argue it's been hijacked by extremists to justify power grabs. The comments also spent a lot of energy arguing whether World War III has already started (Crimea 2014? The current multi-front proxy wars?) or whether that's a dramatic overstatement, with the pushback being that a real world war would be far worse than what we're seeing now.
Steam Controller Auto-Charge – pilot to magnetic charging puck using CV [comments]
125 points · 25 comments · github.com · 9h ago
This project uses an overhead webcam and computer vision to autonomously guide a Steam Controller across a desk to dock with a magnetic charger—by vibrating the controller's haptic motors to make it crawl. The HN thread quickly focused not on the code, but on the fact that the README buries the lede: it never plainly says "the controller moves by vibrating," instead using phrases like "internal dual Linear Resonant Actuators" and "pilots the controller," which several people argued is a dead giveaway that the README was AI-generated. A link to a video showing the controller scooting across the desk went around, and one person pointed out that a human would have opened with "vibrate to magnetic charging puck with computer vision." The discussion also branched into frustration over Steam Controller availability, with multiple people reporting reservation dates as far out as 2027 and debating whether Valve's randomized delivery is better than letting scalpers set a market price.
Odin, Wikipedia and engagement farming [comments]
124 points · 165 comments · katamari64.se · 8h ago
The piece is a blog post dissecting the recent deletion of the Odin programming language article from Wikipedia, framing it as a case study in how Wikipedia’s notability guidelines and reliance on traditional secondary sources fail niche but active software communities — and accusing the language’s creator and other prominent figures of performative outrage and “engagement farming” on Twitter. The HN discussion split sharply: a large contingent argued Odin is genuinely obscure outside game-dev and systems-programming bubbles, that the deletion was procedurally correct, and that the real problem is fans mistaking Discord buzz for encyclopedic notability. Others pushed back hard, pointing out that Wikipedia happily retains articles for decades-old esolangs like Malbolge or near-dead Scheme implementations, making the standard seem arbitrary and hostile to languages that haven’t landed corporate press releases or academic papers. The thread also dug into the deeper tension — that reliable-source rules were designed for traditional media and map poorly onto programming languages where the best documentation is the compiler repo, a Discord server, and industry adoption rather than a New York Times feature. A minor but telling tangent emerged around whether Odin influenced or was influenced by Jonathan Blow’s Jai, with the consensus landing firmly that the claim of Odin influencing Jai is backward.
The Life and Times of Maxis, Part 1: SimEverything [comments]
117 points · 15 comments · www.filfre.net · 16h ago
The linked article from Filfre (The Digital Antiquarian) traces the history of Maxis through its early "SimEverything" era, arguing that massive mainstream hits like *SimCity* (and *Myst*, and *Trophy Bass*) were often dismissed by hardcore gamers but sold to a much wider, non-identifying audience. On HN, the discussion veered hard into personal nostalgia—people recounting hundreds of hours in *SimCity*, *SimEarth*, and *SimAnt*, with several noting the inscrutability of those dense manuals as a formative experience. One commenter dropped a link to a 1996 Stanford talk where Will Wright demonstrated an early *Sims* prototype (then called *Dollhouse*) loaded directly from a *SimCity* save file, which got more traction than the article’s core thesis. Another voice pushed back against the article’s framing by asking whether Wright was influenced by the West German board game *Ökolopoly* and the cybernetics of Norbert Wiener—a genuinely unexpected historical angle. The thread also split over the UI direction Maxis took after *SimCity 2000*, with one person lamenting the shift from a detailed Win32 GDI interface to "big dumb buttons" and a more kid-friendly aesthetic.
Farmer, marketer at odds over sales of white nectarines [comments]
115 points · 118 comments · apnews.com · 13h ago
A California farmer, locked in a lawsuit with a fruit company over exclusive rights to a white nectarine variety, is giving away 100,000 pounds of fruit rather than letting it rot. The thread immediately split into a bitter fight over plant patents and intellectual property on food crops, with many commenters arguing that no company should hold rights over a living thing and that patents on food are evil—while others shot back that decades of selective breeding need private incentive, and the quality of modern produce is proof the system works. A few people punctured the moral outrage by pointing out that the article itself says this nectarine variety isn’t patented at all; it’s a straightforward contract dispute where the farmer agreed to sell only through one packer and then tried to sell elsewhere. The conversation kept circling back to Monsanto and terminator seeds, but the sharpest takeaway was that the loudest anti-patent arguments were attacking a case that isn’t even about patents.
Applied Category Theory Course (2018) [comments]
102 points · 7 comments · math.ucr.edu · 11h ago
This is a full online course on applied category theory taught by John Baez, based on the _Seven Sketches in Compositionality_ book. A few people immediately recognized Baez's name and praised his ability to explain tough ideas clearly, but a couple of others poked fun at the phrase "applied category theory" as an oxymoron. The most substantive exchange came from a programmer who said the book hit a good balance between abstract nonsense and practical use, and asked for more recommendations — getting pointed toward Lawvere's _Conceptual Mathematics_ and Eugenia Cheng's _The Joy of Abstraction_. Overall the thread was thin, more celebratory and slightly skeptical than deep, with the real value being a concrete pointer for anyone looking to bridge category theory and programming.
Generated 2026-07-04 08:29 UTC
Generated by Sauron from Hacker News discussions and linked articles.