HN Brief: 2026-08-02
Today’s HN had a current of quiet fury running through it—multiple threads documented how big platforms (Google, Cursor, BMW) quietly degrade their own products, and the crowd was in no mood to forgive. Alongside that, a nostalgic countercurrent surfaced: people mourning the pre-enshittification web, sharing RSS setups, and celebrating a 15-year-old’s cycloidal gearbox or a calculator running Linux. The AI threads were split as always, but the debate shifted from “is it useful” to “is it corrupting”—between the fake bug reports and the Stross manifesto, the unease felt different today.
Dive into “How Google helped destroy adoption of RSS feeds” for the communal lament of the open web’s lost era, and “Cursor removed cost information from the usage page” if you want to watch power users fight opaque pricing with receipts. “A Surveillance Treaty in Disguise” unpacks a quiet cross-border data grab Canada signed without consultation, while “The Silicon Valley Founder Meat Grinder” delivers a lurid cautionary tale that half the thread insists is real and the other half calls a tall tale.
How Google helped destroy adoption of RSS feeds (2023) [comments]
509 points · 175 comments · openrss.org · 13h ago
The article lays out a methodical case that Google deliberately embraced, then killed off RSS support across Chrome, Reader, FeedBurner, and Google News in order to drive people toward its own ad-ridden, centralized platforms. The HN crowd came out swinging in full agreement—nobody is defending Google here, and the shutdown of Google Reader in 2013 is repeatedly called out as the moment the company tipped from “great” to “kinda evil.” The main pushback is less about the article’s facts and more about the motives: people argue Reader was killed to push Google+ (which also died), and that the real reason was ad revenue, since RSS cuts out the middleman and can’t easily be monetized. A strong undercurrent of nostalgia runs through the thread, with folks mourning the pre-enshittification internet and pointing out that Google has a cultural inability to nurture products that aren’t billion-user moonshots. Several people share their own RSS setups (like NetNewsWire or FreshRSS) as a quiet rebellion, but the consensus is clear: Google’s pattern of embrace-extend-extinguish has done real damage to the open web.
Cursor removed cost information from the usage page and CSV export [comments]
323 points · 146 comments · forum.cursor.com · 16h ago
A forum post from a Cursor user blew up about the company silently switching the usage dashboard from dollar amounts to token counts, killing the CSV cost export and even zeroing out historical cost data in the API. The thread is furious: power users say they relied on the dollar-per-request breakdown dozens of times a day to track budget and compare model efficiency, and they see this as a deliberate move toward opacity to hide what they're actually being charged. A Cursor employee confirmed the change is intentional for non-enterprise plans, claiming dollar amounts “led to confusion” because included usage looked expensive—which the thread dismisses as a transparent excuse to obscure costs. The conversation quickly broadens into a referendum on whether Cursor is worth it in 2026, with a split between people who still praise its integrated IDE and Composer 2.5 model routing, and a vocal camp who’ve moved to Codex, Claude Code, or Zed and say Cursor’s value has evaporated now that it’s owned by SpaceXAI and its UI keeps degrading.
Diátaxis [comments]
312 points · 39 comments · diataxis.fr · 11h ago
The link is to Diátaxis, a framework that splits software documentation into four distinct categories—tutorials, how-to guides, reference, and explanation—to keep authors from mixing teaching with reference material. Several people in the thread admitted they found it most useful as a prompt to throw at an LLM for a quick first pass at docs, which prompted a mix of agreement and eyerolls. There’s a persistent split between those who swear by it for organizing sprawling codebases and those who argue it forces extra clicks—one frustrated dev specifically called out that hiding API docs under a “reference” tab turns a one-click lookup into a two-click annoyance. Some commenters pointed out that the framework isn’t new, with one noting it’s essentially a refined version of Divio’s older system (the author of Diátaxis even chimed in to agree that the earlier Divio version got things wrong). A separate camp argued the real problem is that the technical writing profession never popularized its own deep knowledge, leaving a lightweight framework like this to become the default thinking tool for developers who don’t know what they don’t know.
Seedance 2.5 [comments]
311 points · 159 comments · seed.bytedance.com · 11h ago
ByteDance released Seedance 2.5, a video-generation model that can produce coherent 30-second clips with multimodal references and editing controls. The HN thread immediately split over whether these models have any legitimate use beyond misinformation and spam, with some arguing they're only useful for slop and shady ads, while others pointed to filmmakers and educators who can finally realize scenes that would otherwise be impossible. A former filmmaker chimed in with a more ambivalent take — they'd dreamed of tools like this but now feel strangely disinterested, comparing it to pirating a ton of games and playing none, because the collaborative struggle of filmmaking was part of the art. Others pushed back hard on the "harm-only" narrative, noting that the quality here is genuinely impressive and we're close to climbing out of the uncanny valley, while a separate thread went deep on the ethical line between harmless fetish fodder and deepfake revenge porn.
A Surveillance Treaty in Disguise: Canada Signs UN Cybercrime Convention [comments]
290 points · 158 comments · www.michaelgeist.ca · 17h ago
Canada quietly signed the UN Cybercrime Convention, which Michael Geist argues is really a cross-border surveillance and evidence-sharing pact—one the government itself opposed less than a year ago and that Russia, China, and Iran helped draft. The thread quickly splits: some people point out that Australia, the EU, and the UK also signed, and that signing isn’t ratification, so the practical impact might be limited, but others push back hard that signing signals an intent to ratify and that the treaty’s procedural powers turn repressive domestic laws into triggers for international data grabs. A huge chunk of the discussion pivots to American digital rights and libertarianism—complete with calls for a new Bill of Digital Rights and arguments about whether millionaire tech donors actually want privacy—which frustrates Canadian commenters who want to focus on the treaty’s implications for Canada’s lawful access agenda and the danger it poses to diaspora communities and security researchers. The consensus among the informed voices is that the government signed in mid-summer without public consultation, that the treaty’s marginal value lies in cooperating with authoritarian states, and that it effectively subverts existing safeguards like those in Canada’s mutual legal assistance framework.
AI financial advice is surprisingly good, especially if you ask right questions [comments]
289 points · 255 comments · mitsloan.mit.edu · 9h ago
A new MIT Sloan study found that LLMs give surprisingly good financial advice—encouraging higher savings, diversified stock investing, and age-appropriate risk reduction—but the quality improves sharply when users write structured, academic-style prompts rather than vague ones. The Hacker News thread quickly pushed back on the idea that "good advice" just means generic platitudes like "save more and diversify," with several people noting that the real test is whether AI can handle tricky edge cases like tax law, local jurisdiction quirks, or volatile leveraged ETFs (TQQQ drew a whole side debate on whether holding it long-term is actually fine if hedged properly). A few commenters shared concrete positive experiences feeding their own YNAB or Tiller transaction exports into Claude, getting genuinely useful budgeting and spending-pattern analysis, but others warned that AI confidently recommends S-corp conversions without knowing you live in NYC, where it’ll cost you more. The strongest split in the thread was between the technical optimists (who think AI is already threatening financial advisors’ jobs) and the people who argue that the hard part is behavioral and emotional—talking clients out of panic-selling during dips—which no LLM can do.
NetBSD 11.0 [comments]
269 points · 121 comments · blog.netbsd.org · 14h ago
NetBSD 11.0 finally shipped after a long delay, with the release notes openly acknowledging unresolved security issues and promising a 11.1 follow-up within two months. The thread immediately split over whether NetBSD is viable as a daily driver — one camp pointed to Bluetooth support that OpenBSD lacks (NetBSD has had it for 20 years, using code that eventually got ported to and then ripped from OpenBSD), while others dismissed the whole idea, saying it's nowhere near Linux for desktop use and that your hardware determines everything. A long tangent erupted over the continued inclusion of CD-ROM images: defenders cited support for ancient 486 and VAX hardware that can't boot USB, plus the enduring availability of CD-R discs and the fact that .img files exist for USB users anyway, making the complaint feel like missing the point of supporting obscure architectures. On the GCC front, someone asked why NetBSD ships with GCC 12.5 when 16.1 is latest — the answer wasn't GPLv3 fear (that died with GCC 4.2.1) but rather that the branch was cut a year ago and bumping the compiler risks breaking VAX and other exotic ports that newer GCC struggles with.
Google has abandoned Google News? [comments]
268 points · 193 comments · elgan.com · 12h ago
The article argues that Google News has become nearly unusable—filters for date, language, and region are routinely ignored, and results are now cluttered with social media posts and foreign-language content. Many in the thread immediately agreed and pointed to paid alternatives like Kagi and Ground News as working replacements, though one journalist said those don't handle ultra-specific keyword tracking. A vocal skeptic couldn't replicate the reported bugs and accused the author of sloppy reporting—cropping screenshots and conflating Google Search with Google News—while others noted that search results are famously non-deterministic anyway. The conversation quickly expanded into a broader lament about the enshittification of consumer software over the last decade, with side tangents about Windows peaking at 2000 Pro and a SoftBank investor presentation literally about golden geese.
RipGrep musl binaries occasionally segfault during very-large searches [comments]
268 points · 181 comments · github.com · 19h ago
The submission is a detailed GitHub issue reporting that the musl-built ripgrep binary intermittently segfaults when searching very large directory trees with high concurrency, with the crash traced to a heap-metadata assertion inside musl's `calloc` called from `opendir`. The thread immediately splits into two camps: one side argues that the analysis linked in the issue—a long write-up—is obviously AI-generated slop, verbose and incoherent, and that it adds negative value by wasting expert time on plausible-sounding nonsense; the other side counters that the reproduction case is real and valuable, and that an AI's ability to gather raw observations and point human experts at the right interaction is a net positive even if the conclusions are wrong. A systems dev with experience debugging similar kernel-level races offers a concrete alternative theory—a bug in direct page table zapping exposing stale cached entries—and notes that the AI’s explanation is nonsense. The real tension isn’t about the bug itself (which everyone agrees exists) but about whether AI-written bug reports should be taken seriously when the analysis is opaque and possibly hallucinated.
RamenHaus [comments]
235 points · 119 comments · ramen.haus · 23h ago
The article is a curated photo directory of ramen shops worldwide, organized by city and country. HN immediately turned it into a referendum on whether ramen outside Japan is worth the hype—locals defended London's Kanada-Ya chain while others insisted only the real thing in Japan, or at least Düsseldorf's Japanese expat scene, delivers. The economics of ramen in Western cities got roasted as overpriced for what it is, with comparisons to $20 burgers and a longer argument breaking out about whether lingering for three hours over dinner is normal or ridiculous. There was also a meta tangent about why a food site made the front page, answered by the "intellectual curiosity" guideline and a wave of nostalgia for simple, JavaScript-free enthusiast sites from the 1990s, though one person pointed out that no JS made navigating the gallery cumbersome.
The Silicon Valley Founder Meat Grinder [comments]
221 points · 147 comments · zaksa.zip · 11h ago
The article tells the story of “Jim,” a charismatic but unfocused bootcamp grad who rockets into Y Combinator and a string of founder roles before crashing out in a haze of drugs, orgies, and financial ruin. HN immediately split on whether this is a real archetype or a tall tale: several commenters insisted they’ve never seen the drug-fueled sex parties described, while others shot back that the “hacker house” scene and EA/rationalist circles absolutely have that reputation, with one linking to a Futurism piece about OpenAI-related orgies. A big chunk of the thread zeroed in on what the author framed as “lifestyle creep”—a $10k table, home brewing, a dog and cat—and argued that those are cheap hobbies, not signs of recklessness, and that the real problem is the expectation that founders should have no life outside work. Others defended the “fake it till you make it” strategy as a survival tactic in a system that rewards first-to-market grifters, while a separate tangent dug into widespread amphetamine use in SV, with people noting that “80-hour weeks” is often just code for stimulants.
The Art of 64-bit Assembly [comments]
221 points · 101 comments · nostarch.com · 17h ago
The submission is a book announcement for *The Art of 64-bit Assembly*, which promises to teach low-level Windows assembly by building constructs like vtables, exceptions, and coroutines from scratch in MASM, arguing that AI-generated explanations paper over real ABI details. The thread immediately derails into a deep technical argument about Windows vs Linux calling conventions and vtable ABI standards — people swap nitty-gritty details about MSVC vs Itanium ABI, whether COM vtable layouts are actually defined independently of C++, and how Cygwin and MSYS2 mix conventions by switching to SysV internally. A parallel debate erupts over whether anyone still writes assembly by hand in the age of LLMs: some defend hand-coded asm for critical code like interrupt handlers or real-time MCUs, while others point out that LLMs can be useful as iterative translators or for exploring SIMD optimizations, though they often produce buggy or slower code. A recurring pushback is that assembly is often a hobby or a way to achieve true mastery, and that compilers haven’t eliminated the need for hand-tuned hot loops or crypto-timing guarantees.
Go 1.27 Interactive Tour [comments]
181 points · 52 comments · victoriametrics.com · 6h ago
The article walks through Go 1.27’s new features—generic methods, promoted field keys in struct literals, a UUID package, and post-quantum crypto—with runnable examples. The thread immediately got sidetracked by the old generics debate: one person was “absolutely shocked” Go finally added them, but a former Go team member shut that down by saying it was never about opposition, just about finding a design everyone liked, and another commenter pointed out that the same core people were involved the whole time. Someone tried to pivot to whether generics could fix Go’s error handling, but multiple people argued it’s a non-starter—the Go team has explicitly closed the door on syntax changes for errors, and the real problem is people not checking them, not the pattern itself. A more practical worry came from someone who flagged the automatic draining of HTTP response bodies on Close as a subtle behavioral shift that could break code relying on an early Close to abort large downloads, though the team added a conservative limit and a keepalive opt-out.
A directory of people who love RSS [comments]
170 points · 85 comments · andrewshell.org · 15h ago
Andrew Shell launched a directory for people who love RSS, letting them add their site and display a live blogroll. The thread quickly derailed into a deep Unicode rabbit hole when people noticed the heart in the title actually renders in color on some devices—sparking a debate about whether HN filters emoji, whether that character is really an emoji, and how OS-level font rendering and variation selectors work. Elsewhere, the usual RSS-versus-Atom fight flared up: Atom fans argued it’s technically superior but conceded that RSS 2.0 won because podcasts killed Atom and “good enough” prevailed. Several commenters pushed back against the idea of yet another directory, calling it a profiling goldmine for big tech, while others countered by sharing their own RSS projects and pining for the return of indie blogging without today’s content-mill incentives.
Linux desktop market share has hit over 10% in North America [comments]
160 points · 74 comments · old.reddit.com · 4h ago
The linked article wasn’t available to this summarizer; from the discussion, it’s about StatCounter data claiming Linux desktop share has crossed 10% in North America. The thread immediately picked apart the numbers: Cloudflare’s own radar shows only about 6% if you filter for likely-human traffic, and a recent jump in Linux share neatly coincides with ChromeOS suddenly dropping—suggesting a ChromeOS update may have started misidentifying itself as Linux in user-agent stats. People also slammed StatCounter for still listing “OS X” and “macOS” as separate categories (despite being the same thing), which makes the entire methodology look sloppy. Still, several commenters argued that the *trend* is real—gaming on Linux (Steam Deck, Proton) and Windows enshittification are driving genuine adoption—even if the exact percentage is nowhere near 10%.
Anime User Interfaces [comments]
157 points · 42 comments · animeuserinterface.tumblr.com · 9h ago
The Tumblr post is a grab-bag of anime screenshots showing old-school futuristic interfaces—cockpits, control panels, data readouts, the works. A chunk of the discussion immediately turned into a semantic fight over what counts as a UI: some people said half the images have no interface at all, while others argued that any information display or control surface, even a book's page layout, is a user interface. The more interesting split was about why those dense, button-laden designs vanished: one side says high computer literacy let creators assume users could handle complexity, the other counters that user-hostile dumbing-down was a deliberate business move, not progress, and that "AI" might finally force interfaces to show more information again. Several folks called out missing classics like the ZERO System from Gundam Wing or the paper computer from Golden Boy, one person flagged that a WarGames clip snuck in, and a few commenters got suspicious that the whole thread was a stealth ad for an Evangelion-themed Wacom tablet.
Postmortem for Kernel Soundness Bug #14576 [comments]
143 points · 53 comments · leodemoura.github.io · 13h ago
The postmortem details a Lean kernel bug that allowed a fake AI-generated "disproof" of the Collatz conjecture to pass verification. HN focused on the uncomfortable implications: the bug was only reachable via metaprogramming, which the frontend normally catches, and it took two independent bugs in two kernels (Lean and nanoda) for the exploit to work, so updating both closes the door. Several commenters pushed back against the idea that this undermines formal verification, pointing out that you can only trust a verifier as far as you trust any software, and that the real safety net is having multiple independent implementations—Metamath's approach with a tiny kernel got held up as an alternative, though people noted even Metamath's verifiers have had bugs. There was also a split over whether the "disproof" framing was flashy gimmick that added noise, with maintainers wishing it had just been a simple proof of false instead. The broader takeaway: AI-generated proofs that exploit kernel bugs are a concrete risk, but the Lean team is already hardening and tasking security models to find more.
Kenji/Serious Eats – 30-Min Pressure Cooker Pho Ga [comments]
139 points · 79 comments · www.seriouseats.com · 16h ago
The linked article is Kenji López-Alt’s Serious Eats recipe for a 30-minute pressure cooker chicken pho, which uses the sealed pot to extract flavor fast while keeping the broth clear. The thread quickly split into two camps: people who wanted to talk about pho versus ramen as comfort foods, and people who questioned why a recipe was on HN at all. Defenders pointed out that Kenji is an MIT grad who applies food science, making his pressure-cooker hacks a genuine technical hack—especially his note that electric pressure cookers maintain the same pressure regardless of altitude, which generated its own sidebar on how the thermostats actually work. Others flagged that the recipe is from 2015, not fresh, and griped about SEO-driven “updated on” dates, while longtime fans vouched for the recipe’s reliability and swapped tips for adapting it to electric models that run at lower psi.
On the non-use of AI in my writing process [comments]
130 points · 122 comments · www.antipope.org · 19h ago
Charlie Stross posted a manifesto explaining why he refuses to use LLMs in his fiction writing — citing stolen training data, his involvement in the Anthropic class action, and the tools’ fundamental lack of embodiment or understanding. Hacker News immediately zeroed in on his claim that “generalized adversarial networks” are the key tech behind LLMs, which is technically wrong (LLMs are transformers, not GANs), and a long thread erupted over whether that mistake undermines the whole piece or is just a minor aside from a novelist who shouldn’t need to be an engineer to have a valid opinion. Some argued the error is ironically a sign of human authorship — an LLM would have caught it — while others used it as a cudgel to dismiss Stross entirely, prompting pushback that his main arguments about copyright and the authorial craft stand independent of that technical glitch. The thread then spiraled into agentic systems’ ability to plan long-range narrative (with one camp insisting modern tooling already handles that) and a deeper, more philosophical debate about whether people opposing AI are really struggling to articulate a secular version of “the human spirit” — and whether that’s a more honest framing than quibbling over water usage or capability ceilings.
Show HN: I'm a 15 Year Old Wannabe Engineer, This Is a Cycloidal Gearbox I Built [comments]
120 points · 22 comments · github.com · 5h ago
A 15-year-old posted a cycloidal gearbox they designed and built, along with the Python script that generates the CAD models. HN was broadly supportive and impressed, with multiple people telling them they're already an engineer for doing the work, not just wanting to be. A few commenters warned about posting age online, and there was a split over AI involvement — some pointed out that the project could have been vibecoded with AI and advised the kid to make something without it, while others pushed back that the craftsmanship speaks for itself regardless. The thread also took a minor detour into Hack Club, with its founder inviting them in and a reply sharing nuanced concerns about the pressure Hack Club can put on teenagers to accelerate their lives.
Running Kimi K3 on MI355X at Better Performance per Dollar Than B300 [comments]
112 points · 19 comments · www.wafer.ai · 3h ago
The article claims Wafer.ai ran the 2.8T-parameter Kimi K3 model on AMD MI355X GPUs and achieved better performance per dollar than Nvidia's B300, positioning AMD as a viable, cheaper alternative despite software hurdles. The HN thread savaged the post as barely-disguised AI slop, pointing out that the article left em-dashes intact and used cherry-picked pricing ($2.50/hr for MI355X vs $6.00 for B300) that doesn't reflect real-world rental or ownership costs. Several people noted the B300 beat the MI355X on every raw throughput metric, and the only MI355X advantage was cost per hour on a single aggregation site — with one commenter calling the whole thing a shoddy, exaggerated comparison from a company that previously rug-pulled a token subscription plan. The thread also devolved into the usual debate over calling open-weight models "open source," with most agreeing the term is misleading when training data and tools aren't released.
YC founder asks desperate job seekers to tattoo themselves for an interview [comments]
105 points · 62 comments · sfstandard.com · 11h ago
The story is about a YC-backed AI startup CEO who had a tattoo artist at a party and offered job interviews to anyone who got the company’s logo permanently inked, then apologized after the inevitable backlash. HN mostly reacted with disgust, calling it a ghoulish exploitation of desperate job seekers and a sign that the startup scene has eaten itself, though a few pushed back that the article made it sound more sinister than a simple party with free tattoos and a poorly worded social media stunt. The thread split over whether this reflects YC’s broader culture or is just one freak—but several people pointed to other YC companies with company tattoos among early employees and a job listing that romanticized polar-exploration levels of hardship for $40k. Others dismissed it as a pure publicity play (“any press is good press”) and noted the apology itself read like AI slop, while a smaller camp argued the real story is how seven people were desperate enough to take the offer.
Explorative modeling: Train on the best of K guesses [comments]
99 points · 25 comments · alexiglad.github.io · 16h ago
The article proposes "Explorative Modeling," a training trick where a generative model makes K random guesses for each data point and only backpropagates through the best one, turning the old averaging-into-blur problem into a way to scale generative capacity. HN immediately split into two camps: one calling it a breakthrough that “makes every prior image model obsolete” if the numbers hold, and another pointing out strong similarities to existing methods like Importance Weighted Autoencoders, GRPO, and minibatch optimal transport—arguing the claimed novelty is oversold. Several domain experts pushed back hard on the article’s framing, saying the author misunderstands that diffusion and autoregressive models don’t just “factor generation” to avoid blur; they output distributions, not point estimates, so the whole premise is itself confused. Others noted that the added training cost (K forward passes) and the equal-likelihood sampling behavior are real downsides, though some thought those could be fixed with joint prediction. The most interesting tangent was the comparison to GANs, which one commenter said felt more like sampling real modes than diffusion’s slop—but that thread got derailed by a garbled reply.
Tell HN: I hate your fuzzy search
94 points · 56 comments · news.ycombinator.com · 13h ago
The post is a rant against fuzzy search everywhere, arguing that exact match should always be an option and that sites break this by burying it or removing it entirely. The thread largely agrees, with people piling on specific offenders—Gmail’s mobile search that loses results after you press enter, Spotify’s refusal to show exact artist matches, Facebook Marketplace serving unrelated garbage, and even YouTube surfacing lowest-common-denominator trash. But several people point out that HN’s own search (via Algolia) already supports exact match if you use double quotes, which the submitter apparently missed; a few counter that fuzzy search is what makes Google useful for normal people who can’t remember exact names, and suggest the real fix is to show exact results first, then fuzzy, with clear toggles. A deeper split emerges between power users who want deterministic, literal results and engineers who argue that fuzzy-first is cheaper to implement and serves the majority who don’t type queries precisely.
But can your calculator run Linux? [comments]
91 points · 9 comments · raymii.org · 12h ago
The article is a detailed guide to getting Linux running on an HP Prime G2 calculator, complete with a custom kernel, touchscreen driver, and Doom—because of course. The thread mostly took it as a fun technical exercise, but the conversation quickly turned to whether graphing calculators still have a real use case at all: one camp argued that phones and laptops have made them obsolete, while others pushed back hard on that, pointing out that the HP 48/50 series’ unit handling and RPN workflow are still unmatched for quick, keyboard-driven calculations in a way no phone can replicate. Someone also jumped in with the old NetBSD-toaster joke, only to have another person correct the history—the joke was about BSD running on a toaster, and it was never literally possible with the original toaster hardware, just a hobbyist hack. Overall, the thread was split between nostalgia for the golden age of HP calculators and a pragmatic shrug that, while Linux on a calculator is neat, the stock OS with Python and physical buttons is actually more usable day-to-day.
How to Do Great Work (2023) [comments]
89 points · 68 comments · paulgraham.com · 23h ago
The linked article is Paul Graham's 2023 essay laying out his recipe for doing great work: find something you're deeply curious about, get to the frontier, notice gaps, and explore. HN spent most of the thread questioning whether PG himself qualifies as having done great work—some pointed to YC, Bayesian spam filtering, and Hacker News, while others argued his definition is essentially "work that generates large investment returns," making the whole thing a veiled startup manifesto. A significant contingent dismissed the essay as generic advice for the already-privileged, pointing out that structural barriers like poverty or lack of free time make this kind of self-directed ambition impossible for most people. A separate strand was a direct recommendation to skip PG's essay entirely and read Richard Hamming's "You and Your Research" instead, with several people calling that the superior guide. There was also a weird detour into whether an AI voice reading the essay is neat or cringe.
CISA Alert: Water Sector PLC Targeting [comments]
89 points · 63 comments · censys.com · 13h ago
A CISA alert and a Censys report detail how threat actors are targeting internet-exposed programmable logic controllers in water utilities, locking out operators and forcing boil-water notices. The HN thread largely sidestepped the specific vendor counts and instead lit into the utter failure to secure critical infrastructure after 15 years of federal warnings, with many arguing that connecting PLCs to the raw internet with default passwords crosses into deliberate negligence. A deep split emerged over whether the root cause is incompetent leadership that should face jail time versus systemic underfunding and inability to attract skilled staff in small municipalities—one side insisting the federal government should be auditing every utility, the other countering that there are 150,000 separate water systems and regulation can’t scale that way without understanding the problem. Some commenters with government experience pushed back hard against claims of a purely political, no-accountability bureaucracy, arguing that federal employment is merit-based and that the real issue is project-based, lowest-bidder upgrades that never include security.
Plug-in solar is coming. Plug-in batteries should follow [comments]
88 points · 104 comments · www.regen.co.uk · 5h ago
The UK article argues that plug-in batteries—portable units that charge and discharge through a standard wall socket—should be legalized to let renters and flat dwellers cut electricity bills via time-of-use tariffs, just as plug-in solar recently was. HN immediately zeroed in on the fire risk: the core worry is that a battery discharging through a wall socket bypasses the fuse box, letting a circuit draw more current than its wiring is rated for, though defenders note UK ring mains and fused plugs on every appliance make that far less dangerous than it sounds. A substantial chunk of the comments veered into a California vs. everyone-else rant about net metering, with several people claiming utilities are scamming solar owners with retroactive fees and surcharges, while others shot back that NEM 3.0 fairly reflects grid costs and that rooftop solar at retail rates was never sustainable. The thread split between Brits debating whether 800 W plug-in batteries are a real safety hazard or a solved problem, and Americans using the opportunity to air grievances about utility monopolies and Prop 13.
BMW Is Showing Commercials on Their Car's Dash Screens as a Treat [comments]
88 points · 94 comments · www.theautopian.com · 16h ago
The article reports that BMW is showing full-screen Spider-Man ads on dashboard displays when you start the car, pitching it as a "festive animation" even though it’s just a promotional banner you can click to play. The HN thread is broadly furious—calling it a breach of the car-as-sanctuary covenant and another step in BMW’s enshittification after the heated-seat subscription debacle—but a few people point out that the ad is a one-time optional pop-up tied to a movie partnership, not a permanent feature. The discussion quickly expands into predictions that every connected car will eventually show ads, with some arguing this is inevitable as long as automakers need to grow revenue, while others counter that the real problem is that BMW is destroying its own luxury cachet by treating $60k owners like free-to-play marks. A separate vein of the conversation veers into comparing Chinese versus European car quality and sanity, suggesting the market pressure from Chinese EVs is what’s driving desperate monetization like this.
GitHub has alternatives, but no replacement [comments]
86 points · 122 comments · lalitm.com · 16h ago
The article argues that while there are plenty of Git hosting alternatives, none reproduce GitHub's social layer—shared identities, conventions, and discovery—so there's no real replacement in practice. HN mostly agreed that the lock-in is real, but the dominant reason cited wasn't social features: it's the free Windows and macOS CI runners, which people bluntly said they can't afford to replace for hobby projects, and which Microsoft is effectively subsidizing to the tune of millions. A strong contingent pushed back on self-hosting as impractical for community contributions, echoing the article's point that even filing a bug on a stranger's Forgejo instance is too much friction. The thread also split on Tangled, with some calling it promising and others warning it's VC-funded and therefore suspect, while a separate meta-debate erupted over whether people confusing "Git" with "GitHub" was the real problem—or just a convenient excuse.
Generated 2026-08-02 08:07 UTC
Generated by Sauron from Hacker News discussions and linked articles.