HN Brief: 2026-06-28
Today’s front page was split between two anxieties: AI’s geopolitical race and the erosion of privacy. The DeepSeek speculative decoding paper sparked a debate about US labs hoarding advances while Chinese competitors publish open recipes, while the anonymous 0‑day dump and the IP webcam crawl each raised questions about who gets to control exposure. Meanwhile, Meta’s legal war on a whistleblower and a new anonymous chat widget for websites both highlighted how easily the tools of connection become tools of surveillance.
Threads most worth clicking into: “DSpark” for the political signal behind the tech and the real‑world price cuts it explains; “Zuckerberg's war on whistleblowers” for the escalating Streisand effect and the raw details from *Careless People*; “IP Crawl” for the uncomfortable line between public exposure and privacy violation; “Suspicious Discontinuities” for the real‑world pain of tax and benefit cliffs that can hit 100% marginal rates; and “Turn your site into a place people can bump into each other” for the brutal moderation reality check that undermines even the best nostalgic intentions.
Anonymous GitHub account mass-dropping undisclosed 0-days [comments]
785 points · 306 comments · github.com · 17h ago
An anonymous GitHub account dumped a consolidated archive of 23 proof-of-concept exploits for software like 7-Zip, Firefox, ffmpeg, and OpenVPN, claiming they’re all undisclosed 0-days. The thread immediately split into two camps: one arguing that “0-day” has lost meaning and that many of these are already fixed or public, and another defending the repo as legit open disclosure with the author specifically inviting people to report them for CVEs. A huge chunk of the discussion veered entirely off-topic into whether bank account numbers and SSNs should be treated as secrets at all, with people pointing out that checks literally print both numbers and that fraud is rare due to steep penalties. Another major angle was the author’s AI-fuzzing workflow—the crowd debated whether LLMs make security through obscurity look attractive again, countered by arguments that closed source is equally vulnerable to automated reverse engineering. The author’s snarky manifesto (”cybercrime is cringe,” admitting some PoCs are “kinda ass”) got a lot of meta-commentary, especially the Jamie Wolf bestiality law joke about the uselessness of abuse disclaimers.
DSpark: Speculative decoding accelerates LLM inference [pdf] [comments]
755 points · 317 comments · github.com · 22h ago
DSpark is a DeepSeek paper describing a speculative decoding technique that uses a semi-autoregressive draft model with adaptive verification length to accelerate LLM inference. HN immediately read the timing as a political signal—DeepSeek publishes detailed open-source papers while US labs like Anthropic and OpenAI are pushing for regulation and keeping their methods secret, and the thread split between those who see this as Chinese labs outcompeting through openness and those who argue the US labs simply have different priorities and access to the same techniques. Several commenters pointed out that this paper likely explains the dramatic 75% price cuts DeepSeek rolled out a month ago, linking the production deployment numbers in the paper to real user experience of fast, cheap inference. The technical discussion focused on how DSpark's approach—combining a parallel backbone with a lightweight sequential head to fix "suffix decay"—solves a known weakness in prior speculative decoding methods, and many noted that the hardware-aware scheduler is the key innovation for high-concurrency serving. The overall vibe was that DeepSeek continues to shift the Pareto frontier of cost-efficiency while publishing the recipe, and that the American labs' strategy of hoarding their advances is going to be hard to sustain.
Zuckerberg's war on whistleblowers [comments]
683 points · 242 comments · pluralistic.net · 17h ago
Cory Doctorow's piece details Meta's escalating legal war on whistleblower Sarah Wynn-Williams, who wrote a memoir about Facebook's role in the Myanmar genocide and internal cover-ups — she's been ordered to pay over $11 million for simply standing silent on stage, and her compliance only made Meta demand more. The HN thread immediately spun off into a heated historical debate about whether violent revolution or guillotine jokes are an appropriate response to billionaire impunity, with several people arguing that actual revolutions (France, Russia, China) just replaced one elite with another and killed off the moderates. A significant split emerged: one commenter flatly accused Doctorow of lying about the China data access allegation, claiming the book only said Meta *offered* to give China Chinese-user data (which never happened), while others pushed back that multiple employees resigned over Project Aldrain and the distinction is meaningless. The broader consensus was less about the article's specifics and more about the observation that extreme wealth and power inevitably corrupt, with several people pointing out that Zuckerberg's behavior fits a pattern of sociopathic self-justification common to billionaires, though one defender noted he rarely speaks publicly and every source on him has incentives to lie.
OpenRA [comments]
678 points · 131 comments · www.openra.net · 19h ago
OpenRA is an open-source engine that rebuilds classic RTS games like Red Alert and C&C with modern features—attack-move, fog of war, new random map generators, and a recent Dune 2000 overhaul. The thread quickly veered away from the release notes into a long, philosophical fight about whether game AI counts as “AI” at all, with one side arguing that RTS bots are textbook AI in the Russell & Norvig sense and the other insisting that cheating resource multipliers and perfect map vision make them a fundamentally different thing, just “bots” trying to appear intelligent. Several people dove into the practical side: how LLMs are too slow to play in real-time but are surprisingly good at writing AI scripts for custom games, and how even a fair, non-cheating bot built on simple heuristics can outplay most humans by never getting distracted. A few veterans pushed back on OpenRA’s balance, claiming the AI still cheats with sight bonuses and that the save/replay system is brutal for large maps—20 hours of game takes two hours of CPU to reload—while others countered that the original games were far worse and OpenRA is a dramatic improvement.
Fintech Engineering Handbook [comments]
544 points · 171 comments · w.pitula.me · 21h ago
The linked article is a free handbook distilling six years of fintech engineering experience into patterns for handling money—covering idempotency, double-entry ledgers, and the pitfalls of floating-point arithmetic. The thread swerved hard into a heated debate about how to represent monetary amounts: the handbook recommends sending money as strings or integers in JSON, but experienced practitioners clashed over whether minor-unit integers (e.g., cents) are safe or silently catastrophic when crossing systems with different decimal precision. Several commenters with HFT and payments backgrounds defended integer cents as an industry standard, while others warned that stablecoins, three-decimal currencies, or any cross-currency math will break that assumption and demand BigDecimal or rational numbers. The author replied when someone asked if it was AI slop, admitting ~20% AI assistance for grammar and completeness, which then sparked a side discussion on hyphen usage and whether that betrayed the AI touch—though most agreed the substance was solid and worth reading just for the idempotency keys section.
The case for physical media ownership [comments]
423 points · 272 comments · dervis.de · 20h ago
The article argues that “buying” digital media is often just renting a revocable license, and makes the case for physical discs, books, and cartridges as true ownership. The HN thread mostly agreed with the premise but immediately pivoted to a different conclusion: just pirate everything, since pirated copies are DRM-free, higher quality, and actually stay forever—users pointed out that the industry refuses to sell the product people want at any price, so piracy is a “product problem, not a price problem.” A few commenters pushed back on the ethics of taking from creators, but were met with the counterargument that the current licensing system is itself unethical and that pirates often still pay for the physical copy or support creators directly. Others corrected the article’s rosy view of physical media, noting that Blu-ray players phone home, keys can be revoked, and discs can stop playing on newer hardware—so even physical ownership comes with strings attached. The thread also split on practicality: some prefer the convenience and commitment of a physical shelf, while others argue that running a home NAS with Jellyfin is the only way to truly own and control your library.
IP Crawl: Living atlas of open webcams discovered on the public internet [comments]
290 points · 142 comments · ipcrawl.com · 12h ago
IP Crawl is a site that continuously scans the public internet for openly accessible webcam feeds—no hacking required—and presents them on a searchable map, with the stated goal of shaming manufacturers and cluing in negligent owners. The HN thread split hard on whether the project is a useful public service or just a bigger and more organized privacy violation, with one side arguing that if you leave a camera wide open it’s effectively public and the site is no different from a search engine, while the other side counters that compiling and streaming someone’s bedroom or living room goes far beyond what a simple port scan does, and that “it’s technically legal” doesn’t make it right. A lot of the pushback zeroed in on the difference between a camera being accessible and a website making it trivially browsable by category and location, with several people analogizing it to publishing a list of houses with open curtains. There was also a minor tangent when someone spotted what looked like a weed grow in the UK, which derailed a bit into long-wave radio towers in Droitwich.
Streaming services' obnoxiously loud ads become illegal on July 1 in California [comments]
266 points · 82 comments · arstechnica.com · 19h ago
California is making it illegal for streaming services to play ads louder than the content starting July 1, closing a loophole that the FCC already closed for broadcast TV years ago. The HN crowd mostly celebrated the news, but the discussion quickly split: some argued the streaming industry's whining about technical difficulty is nonsense—TV stations have normalized ads for decades—while others dove into the genuine technical quirks, like how server-side ad insertion and different encoding pipelines make loudness normalization harder than it sounds. A few commenters pointed out that loud ads aren't just a nuisance but can be genuinely painful for neurodivergent people, contrasting with an older generation that seems to tolerate or even prefer constant blaring background noise. The thread also veered into YouTube's aggressive ad placement and volume inconsistency, with many agreeing that the only real fix is an ad blocker or a different frontend, since YouTube's own "stable volume" toggle reportedly ruins audio dynamics.
Suspicious Discontinuities (2020) [comments]
232 points · 74 comments · danluu.com · 18h ago
Dan Luu's article catalogues "suspicious discontinuities" across tax credits, queue theory, college admissions, marathon finish times, and more — basically, sharp thresholds that create weird incentives or artifacts. HN latched hardest onto the tax-and-benefit cliffs, with people from the UK, Slovenia, and elsewhere piling on their own horror stories of marginal tax rates above 100% or losing thousands in childcare subsidies for earning two euros too much. Several commenters argued the real problem isn't the cliffs themselves but the compounding of multiple phase-outs across different programs, making it almost impossible to avoid >100% effective rates even if each individual cliff were smoothed. A recurring debate was why lawmakers don't just use gradients instead of hard cutoffs — the consensus being that voters don't understand interpolation, though others pushed back that this is a convenient excuse for a system that quietly lets politicians raise revenue through fiscal drag. The marathon and p-value discontinuities got nods as fun corroboration, but the thread mostly stayed grounded in how these artificial boundaries actively screw over real people.
Asian AI startups launch Mythos-like models [comments]
218 points · 167 comments · techcrunch.com · 18h ago
The article covers Asian AI startups Sakana AI (Japan) and 360 (China) launching models that claim to rival Anthropic’s export-banned Mythos, positioning themselves as local alternatives amid US restrictions. Hacker News was deeply split: many dismissed the new models as marketing hype, pointing out that Sakana previously had to walk back claims and that “Mythos-level” means little when most people can’t even benchmark it, while others argued the US brought this on itself through Anthropic’s years of fearmongering and regulatory capture, which backfired by handing the lead to Asia. A loud contingent saw this as karmic justice—Anthropic mined the internet, now others are bringing shovels—but pushed back against calling these “distillations” since they’re fundamentally model orchestrators, not copies. The thread also veered into a heated debate over whether superintelligence is genuinely dangerous or just another industrial revolution that will eventually raise living standards, with the anti-fear side accusing Anthropic of weaponizing safety rhetoric for profit. Underneath it all, a practical observation: Japan’s Sakana is an American-backed ally, not a Chinese adversary, so the real story is about sovereign AI hedging rather than a clean geopolitical shift.
Turn your site into a place people can bump into each other [comments]
215 points · 89 comments · github.com · 14h ago
The article introduces TownSquare, a tiny embeddable widget that populates your site with anonymous stick figures representing other live visitors, letting them see what page people are on, move around, and send ephemeral messages — a deliberate attempt to revive the old-web feeling of bumping into real people. HN largely loved the nostalgic idea, but the comments immediately split into two camps: one side was charmed by the serendipity and the author’s open-source generosity, while the other flagged the predictable abuse problems — within hours, people were posting slurs and violent messages in the TownSquare, and critics argued that anonymity without any persistence or identity actually loses what made the early web feel like a community of familiar faces. The author engaged directly, acknowledged the moderation difficulty (a basic filter and rate limits exist, and a telegram notification plugin was just added for site owners), and even entertained ideas like per-site persistent personas or a locational filter, though others pushed back that those would kill the simplicity or create an elite club. A few commenters surfaced memories of 90s precursors (Odigo’s site radar, PowWow, even the game Journey) and debated whether this kind of thing can survive a real userbase beyond a novelty spike — the consensus being that the idea is “cute” but the practical realities of anonymous chat moderation are a brutal, unsolved problem.
The best response to AI slop and online noise is from Robin Williams [comments]
193 points · 110 comments · jayacunzo.com · 6h ago
The article argues that Robin Williams' "Your move, chief" speech in *Good Will Hunting* perfectly captures why AI-generated content falls flat—it knows everything but has lived nothing, like Matt Damon's character. HN mostly agreed that lived experience matters but split hard on whether people actually take LLMs that seriously, with some pointing to kids losing confidence because AI cranks out passable beginner work and others noting the absurdity of treating statistical text generators as if they've tasted a strawberry. A long tangent erupted comparing LLMs to the *Blade Runner* "Tears in Rain" monologue, with multiple people arguing replicants are fundamentally different because they actually experience the world, while others rolled their eyes at the whole "meat brain" vs. "stochastic parrot" framing. The thread also lit up over LLMs using phrases like "I personally prefer" or "my favorite way to"—several people said this backfires hard and feels like a desperate attempt to fake humanity that only makes the machine more alien.
'Careless People' author claims Meta surveilled her for 12mos to enforce silence [comments]
186 points · 77 comments · fortune.com · 10h ago
A former Meta exec named Sarah Wynn-Williams is suing the company, alleging it surveilled her for a full year—sending reps to photograph her at public events, even taking issue with her sitting silently on a panel—all to enforce a gag order and non-disparagement agreement tied to her memoir *Careless People*. The thread mostly dug into the book’s specific allegations, with commenters trading stories about Zuckerberg throwing tantrums over losing at board games on the corporate jet, Sheryl Sandberg spending $13,000 on lingerie for assistants and asking them to snuggle in the plane’s bed, and the company knowingly fueling ethnic violence in Myanmar while ignoring employee arrests. A few skeptics argued the author signed a severance deal for a payout and is now cashing in years later, but that pushback was heavily outnumbered by people recounting details that made the executives look genuinely awful—including Zuckerberg forgetting the name of an employee he’d posted about as a martyr. Several readers said the thread itself proved the Streisand effect backfiring, with multiple people queueing up the book after seeing the surveillance story.
Marfa Public Radio Puts You to Sleep [comments]
185 points · 46 comments · www.marfapublicradio.org · 5h ago
Marfa Public Radio launched a sleep podcast that reads the station’s boring internal documents—FCC compliance forms, journalistic ethics codes—to lull you to sleep. The HN crowd mostly loved the concept, but immediately started trading recommendations for similar shows like *Sleep With Me* and *Boring Books for Bedtime*, with some noting those often end up *too* interesting to actually sleep to. A grammar fight broke out over “lay” vs. “lie” in the station’s tagline, quickly escalating into meta-snark about HN norms. A few people couldn’t even access the audio because the Cloudfront distribution blocks non-US visitors, and one listener pointed out the phone number on the page doesn’t match the one read aloud.
Michigan spent $1.8B and only created 602 jobs [comments]
163 points · 76 comments · www.msn.com · 10h ago
The linked article wasn't available to this summarizer; from the discussion, it covers a state report finding that Michigan spent $1.8 billion on business incentives and got only 602 jobs in return—about $2.5 million per job. The thread immediately zeroes in on the fact that almost all the money went to big public automakers, not small businesses, and several people flatly call it a laundering scam or corporate welfare, arguing that if the real goal were job creation, politicians would attach loan-contingency clauses or fund apprenticeships instead of handing cash to executives and shareholders. Others push back by noting that since the automakers' stock is held by pension funds, the money did indirectly benefit retirement accounts, but that gets shot down as trivializing—the point is the executives and shareholders got enriched along the way while the stated purpose failed. A strong split emerges: some see this as straightforward corruption that no one will be prosecuted for, while others argue it's not necessarily malicious but rather a predictable outcome of government being terrible at picking winners, with a side debate over whether the lesson is that we need better oversight or just lower taxes and less government spending entirely.
Ford hired AI and sacked humans. It backfired badly [comments]
161 points · 86 comments · www.the-independent.com · 4h ago
The linked article wasn't available to this summarizer; from the discussion, Ford tried to replace human workers with AI and the move backfired badly. The thread quickly turned into a broader lament about boardrooms still chasing the fever dream that AI will magically cut costs and eliminate “pesky humans,” with several people pointing out that executives are easy marks for AI vendors because they can brag about integration metrics while quality tanks. A recurring theme was how trivial it is to game an “AI mandate”—one person described burning tokens on absurdly overengineered queries just to hit usage targets, calling it corporate financial malpractice that will eventually blow up. Others pivoted to Ford’s long-standing quality decline, noting the irony of a company that once claimed “Quality is Job One” now flooding owners with recalls, and used that track record to argue that this isn’t a one-off mistake but a pattern of chasing cheap fixes instead of doing the hard work.
What Ozempic does to the gut-brain axis [comments]
153 points · 337 comments · www.psychologytoday.com · 10h ago
The linked article wasn’t available to this summarizer; from the discussion, the submission is about how GLP-1 drugs like Ozempic alter the gut-brain axis, but the thread largely ignored the neuroscience and instead became a sprawling, firsthand testimonial thread on what it actually feels like to be on the drug. The dominant theme is that these drugs don’t just suppress appetite—they quiet a constant mental “food noise” many people never realized was there, with some users reporting it also killed urges to online shop, drink alcohol, or binge other addictive behaviors. A sharp split emerged between those who say the effect is uniquely transformative (a switch flipping off) and those who insist that fixing diet or upping protein can produce the same result—with plenty of pushback from people who tried both and say the drug does something qualitatively different. Several autoimmune and ME/CFS patients chimed in to report dramatic symptom relief, while a minority described severe nausea and brain fog that made the experience miserable. The thread also wrestled with whether the Hadza hunter-gatherers and their low-baseline metabolic stress undermine the “modern lifestyle is broken” narrative—but mostly people just swapped protein and dairy tips.
Post-Mythos Cybersecurity: Keep calm and carry on [comments]
145 points · 55 comments · cephalosec.com · 17h ago
The article argues that Anthropic's hyped Mythos AI model, despite dramatic claims about automated zero-day discovery, is actually a modest incremental step—finding old bugs at enormous cost—and that existing security practices like defense-in-depth and vulnerability management remain sufficient. HN pushed back hard on the “keep calm” thesis, with many security practitioners arguing that technical debt and bad configurations are still the real threats, and that Mythos is just another tool—though some countered that equating it to a “tool” is like calling nukes “just another bomb.” A long political tangent erupted over the US government’s decision to restrict Mythos to American institutions, with one side calling it an authoritarian power grab that will create an asymmetric offensive capability, while others saw it as necessary control or even a sign of democratic revival. Several people noted the suspicious timing of Anthropic’s IPO filing right alongside the Mythos release and export restrictions, suggesting the whole affair is a pump-and-dump scheme before open-weight Chinese models catch up in six months.
Choosing a Public DNS Resolver [comments]
139 points · 42 comments · evilbit.de · 9h ago
The article is a deep interactive guide comparing 29 public DNS resolvers on privacy, security, speed, and filtering. The thread quickly splintered: some argued Quad9 with its malware blocklist is all you need, but others pointed out it has misfired on legitimate sites (a private torrent tracker, gist.github.com) and that the unfiltered Quad9 address lacks DNSSEC—though a correction noted Quad9 now enables DNSSEC on all its endpoints. A substantial faction pushed running your own recursive resolver (Unbound, AdGuard) as strictly better, while a counterpoint warned that self-hosting means uncached queries take >100ms and you lose the cache-sharing benefits of a large user base. The most contentious technical tangent was about EDNS Client Subnet: Cloudflare's refusal to send it reportedly breaks ISP caches for YouTube and Google Drive, routing traffic to distant servers and ruining streaming speed, a trade-off between privacy and CDN performance that many said the guide should flag more prominently. Meanwhile, the "bus factor" of single-operator resolvers and the jurisdictional creep of Chinese regulations on non-Chinese entries got serious scrutiny.
Feds Killed Polestar and Spared Volvo [comments]
138 points · 92 comments · www.thedrive.com · 6h ago
The article argues the U.S. government arbitrarily killed Polestar while sparing its sister brand Volvo—both owned by China's Geely—setting a dangerous precedent for picking winners and losers in the auto industry. The thread zeroed in on the inconsistency: if the excuse is Chinese ownership or surveillance, Volvo is just as Chinese and its EVs run similar software, so the decision looks either politically motivated or simply clueless. Several people pointed out that Polestar is a new, unknown brand while Volvo has decades of American loyalty, making a Volvo ban politically untenable for the administration, and one commenter bluntly called it a donor-driven shakedown to protect legacy automakers. A tangent emerged comparing this to the government's unchecked authority over AI models and encryption in the 90s, with some arguing the real problem is granting the executive branch discretionary powers that can be wielded arbitrarily. A few voices dismissed the terror entirely, saying the market will sort it out or that blocking Chinese EVs is just common sense national security.
How Many Elementary Particles Are There, Really? [comments]
129 points · 111 comments · www.quantamagazine.org · 19h ago
Quanta Magazine’s article runs through the many ways to count elementary particles, from the familiar poster-model tally of 17 all the way up to 995.5 — a fractional number that arises from a 2011 theorem about how many “degrees of freedom” a quantum field theory can have. The thread immediately zeroed in on the definitional problem: “elementary” isn’t self-evident, so the count depends entirely on whether you include antiparticles, color charges, chirality, or treat particles as ripples in fields rather than independent objects. Several people who actually work in QFT pushed back hard on the higher numbers, arguing that counting gluon color variants or left- vs right-handed fermions as separate particles is misleading — those are just different excitation states of the same field, not distinct fundamental entities. A sizable split emerged between those who think the 17-field picture is the honest answer (you’re counting field types, not particle permutations) and those who find the 995.5 result genuinely interesting because it reveals something deep about how the Standard Model’s degrees of freedom are constrained, even if it feels like a parlor trick. A smaller tangent spun up around whether “everything is a wave” dissolves the counting problem entirely, which got met with a sharp reminder that quantization is real and standing waves are still discrete.
Beer CSS – Build material design in record time [comments]
128 points · 59 comments · www.beercss.com · 22h ago
The linked article wasn't available to this summarizer; from the discussion, Beer CSS is a library for building Material Design 3 interfaces quickly. The thread was overwhelmingly negative, with most people arguing the project looks terrible on mobile and desktop—buttons misaligned, animations painfully slow, and the landing page itself a broken mess that requires JavaScript to render anything. A recurring split emerged: many hate Material Design itself, calling it visually unappealing and dated, while a few defenders said the library works fine for simple projects and has good DX, though even they acknowledged LLMs struggle to use it. Several commenters pointed out that Google has already abandoned three different Material Design web implementations, making this feel like a low-effort AI-generated project that nobody should adopt.
AMD Strix Halo RDMA Cluster Setup Guide [comments]
117 points · 31 comments · github.com · 7h ago
This is a guide for setting up a two-node cluster of AMD Strix Halo APUs (like the Framework Desktop mainboard) over 100GbE RDMA to run distributed vLLM inference, using Tensor Parallelism across the unified memory. HN is impressed by the technical depth and the sheer capability—it finally lets you run big models on prosumer hardware—but the main tension is the cost: multiple commenters point out that Strix Halo prices have nearly doubled in the last year (one user saw a 128GB laptop go from €2,800 to €7,899), making the whole setup feel like a rich-hobbyist move right now. Others push back on the hardware choices, noting the PCIe x4 slot bottlenecks the 100GbE NIC, and that Thunderbolt 4 (which the guide also covers) lacks RDMA, so latency is much higher than the ~5µs promised. A lively tangent argues whether laptops are even sensible for this kind of compute—some insist you overpay for thermal throttling, others say a single MacBook Pro with M5 Max and 128GB is still the better value despite the price hikes. The consensus: the tech is brilliant and a glimpse of the future, but the current market makes it hard to justify unless you’re already all-in on Strix Halo.
Show HN: Decomp Academy – Learn to decompile GameCube games into matching C [comments]
109 points · 36 comments · decomp-academy.dev · 6h ago
Decomp Academy is an interactive site that teaches you to reverse-engineer GameCube games by writing C that compiles to byte-for-byte matching PowerPC assembly, using a live Metrowerks CodeWarrior compiler running on AWS Lambda. The HN crowd immediately hit the typical angles: several people wanted a way to contribute to real decomp projects, and the creator pointed them to decomp.me instead of reinventing the wheel. A heated split emerged around using AI for decompilation — one side argued LLMs get you 95% of the way there but stall on the tricky tail, while another warned that AI-generated decomp code may not qualify as creative transformation under US copyright law, putting projects in legal jeopardy that human-written decompilation avoids. The backend itself was a talking point: the compiler is proprietary and tough to shoehorn into a Lambda, and a burst of popularity promptly caused a "no space left on device" error that the author had to hotfix.
Show HN: Adrafinil – keep a lid-closed Mac awake only while agents work [comments]
108 points · 71 comments · github.com · 11h ago
The submission is a macOS menu-bar app called Adrafinil that keeps a MacBook awake with the lid closed only while AI coding agents (Claude Code, Codex, etc.) are actively working, then lets it sleep immediately when they’re done — a targeted alternative to always-on tools like `caffeinate` or Amphetamine. The HN thread quickly split: a bunch of people argued that Amphetamine already handles this via app-triggers or AppleScript, but the author pushed back hard, noting that Amphetamine can’t override lid-closed sleep without external power or a dummy plug, and that forgetting to disable it drains your bag-battery — Adrafinil hooks directly into agent turn-cycles and re-enables sleep the moment work stops. Several commenters pointed out the app is “just a fancy `sudo pmset disablesleep 1` wrapper” and called the 15k LOC bloat, while others defended the safety features (thermal cutoff, battery monitoring, reference-counted assertions) as necessary to prevent a closed-lid Mac from cooking itself inside a backpack. There was also a side conversation about the naming trend (Caffeine → Amphetamine → Modafinil → Adrafinil) and whether ARM MacBooks overheat less than Intel ones with the lid shut, with the dev confirming the app has a configurable thermal cutoff that force-releases sleep before damage.
A Farmer Arrested for Going 5 Seconds over His Time Limit at Data Center Meeting [comments]
105 points · 54 comments · www.gadgetreview.com · 11h ago
The article covers a Claremore, Oklahoma farmer who was arrested at a city council meeting for speaking roughly five seconds past a three-minute public comment limit during a debate over a proposed 300-acre data center. HN immediately split into two camps: one side saw blatant authoritarian overreach and a chilling effect on civic participation, while the other pointed out that bodycam footage shows he refused multiple orders to leave and kept shuffling papers while the crowd grew chaotic, making the “five seconds” headline misleading. A significant thread argued the farmer is a known local activist with previous run-ins, and that this story is being leveraged for Streisand effect. Beneath the arrest itself, the comments veered into a broader resentment of data center buildouts—especially for AI—with many insisting no community actually wants them, counterbalanced by defenders noting that people want the services data centers provide and that towns can negotiate real benefits if they don’t cave on tax incentives.
Screen time can damage under-twos' development, landmark study suggests [comments]
87 points · 32 comments · www.theguardian.com · 15h ago
The Guardian reports on a study claiming screen time before age two “can damage” development, though the paper itself explicitly didn’t establish causality. The thread immediately seized on that mismatch—several people pointed out that “using a screen as a babysitter” correlates heavily with stressed, overworked parents and less family support, making causation hard to untangle from poverty or lack of alloparenting. Others pushed back on the blanket condemnation, arguing that video calls with distant relatives or watching educational programming with a parent are qualitatively different from handing a toddler a smartphone to quiet them. A side debate emerged about whether modern child-rearing is artificially hard because nuclear families lack the village of extended kin that evolution presumed, with one parent lamenting that they had a kid at 49 and now have zero help, while another cited research that hunter-gatherer parents only handle their own kids half the time.
Doctors suspected man had brain cancer. He had worms [comments]
69 points · 29 comments · arstechnica.com · 15h ago
A 60-year-old man in Spain with worsening headaches and behavioral changes was initially thought to have metastatic brain cancer, but an MRI revealed tapeworm larvae encysted in his brain — a condition called neurocysticercosis likely contracted from a coworker, despite the man never traveling abroad. The thread immediately split between people calling it a real-life House episode and others diving into the gritty transmission mechanics: the worms spread through fecal-oral contamination, and several people pointed out that norovirus spreads the same way and kills hundreds of thousands annually, so this isn't some exotic freak event. A handful of participants debated prophylactic deworming — one person admitted to taking albendazole yearly, bought over-the-counter in India, while others noted that US doctors won't prescribe antiparasitics without active infection, and a few pushed back arguing that parasitic infections are far more common than the medical establishment admits, especially with pet-to-human transmission. The more skeptical corner warned that even after successful drug treatment, the decaying larvae can trigger a Herxheimer reaction and leave permanent cognitive damage, though others cited studies showing most cases are asymptomatic and the brain can recover remarkably well.
Enhancing X11 Application Security with LXC (2025) [comments]
63 points · 31 comments · dobrowolski.dev · 10h ago
This article walks through using LXC containers to sandbox untrusted GUI apps like browsers, mapping UIDs and forwarding X11 and PulseAudio sockets to keep the blast radius away from the home directory. The thread quickly zeroed in on the elephant in the room: passing through the X11 socket is a giant sandbox escape, since any app can sniff keystrokes, grab windows, or control a root terminal running on the same display. People pointed to Firejail or SELinux sandbox as simpler drop-in alternatives, and there was a lively split over whether nested X servers like Xephyr are the only real fix versus the argument that LXC still shrinks the attack surface meaningfully. A deeper tangent erupted over the state of X11 vs Wayland—some argued Wayland’s security model is the only sane path, while others pushed back hard on Wayland dropping features (input handling, network transparency) and noted that X11Libre and new X servers in Zig are actively maintained and might outlast Wayland’s hype.
America's Large Homebuilders Shift the Cost of Shoddy Construction to Buyers [comments]
60 points · 37 comments · hntrbrk.com · 7h ago
The article details how America's largest homebuilders—D.R. Horton and Lennar—systematically evade responsibility for shoddy construction through one-sided contracts, defective warranties, and forced arbitration, leaving homeowners with massive repair bills. The thread quickly turned historical, with people arguing over whether this is truly unprecedented or just a return to Gilded Age levels of corruption, citing Levittown's infamous septic failures and Grant's scandal-ridden presidency. Several commenters shared first-hand nightmares—a Lennar buyer described a project manager actively discouraging inspections and a yard overrun with voles—while others pointed to YouTube inspector Cy as essential viewing for anyone considering a new build. A significant split emerged around the publisher's business model: Hunterbrook discloses it partners with litigation firms and an affiliated investment fund, which led many to dismiss the piece as PR dressed as short-seller reportage, even as they agreed the underlying complaints are real. The strongest consensus was that production-built homes from any large builder are garbage, and the only sane path is either a pre-1995 house (trading old-growth wood for lead paint and asbestos) or an extremely expensive custom build.
Generated 2026-06-28 08:26 UTC
Generated by Sauron from Hacker News discussions and linked articles.