HN Brief: 2026-05-17
Today’s HN was dominated by a deep skepticism of both surveillance and AI hype. The biggest threads—Mozilla’s plea to UK regulators, the Flock camera destruction spree, and the Bitlocker backdoor disclosure—all circled the same nerve: the gap between what governments and companies claim they’re doing (protecting kids, fighting crime, securing data) and what critics see as mass surveillance, lock-in, or backdoor access. Meanwhile, two AI essays struck a similar chord: one argued AI won’t speed up organizational processes because the real bottleneck is upstream bureaucracy, not coding speed, and another warned AI subscriptions are loss leaders that will hammer enterprises with price hikes once the labs need real margins. A few threads drifted into nostalgia—VoIP-powered pay phones, Amiga versus Atari flame wars—but the mood was distinctly anti-authority and anti-easy-answer.
The threads most worth clicking: “Security researcher says Microsoft built a Bitlocker backdoor, releases exploit” because it forces a rare choice between encryption sovereignty and the creeping helplessness of users locked out of their own machines. “At least 25 Flock cameras have been destroyed in five states since April 2025” because it’s a live, messy confrontation between surveillance infrastructure and public anger, with no clean moral lines. “I don't think AI will make your processes go faster” because it cuts through the productivity hype to name the real bottleneck: vague specs and pointless meetings. “Mozilla to UK regulators: VPNs are essential privacy and security tools” because it exposes the chasm between child-protection rhetoric and what commenters call mass surveillance by another name. And “AI subscriptions are a ticking time bomb for enterprise” because it pokes at the question nobody in the AI industry wants to answer: what happens when the VC subsidies dry up.
Mozilla to UK regulators: VPNs are essential privacy and security tools [comments]
648 points · 271 comments · blog.mozilla.org · 19h ago
Mozilla formally told UK regulators that VPNs are essential privacy and security tools for everyone, including young people, and urged them not to age-gate VPNs under the Online Safety Act. Hacker News was deeply skeptical: the top sentiment was that the UK isn't a free society and this is just another step toward mass surveillance disguised as child protection, with one comment calling age verification “not age verification, it’s mass surveillance under a fake name.” A significant split emerged split around the public’s appetite for this kind of control—several people pointed out that outside the HN bubble, “normies” actually support lockdowns and porn blocks, and that HN itself called for social media regulation until it meant ID checks. The thread also spent serious energy debating whether invoking *1984* was trite or apt, with a long subthread arguing that Orwell was naive compared to *Brave New World’s* more insidious trade-off of comfort for freedom, and that the real threat is ordinary, everyday erosion rather than dramatic totalitarianism.
Security researcher says Microsoft built a Bitlocker backdoor, releases exploit [comments]
550 points · 239 comments · www.techspot.com · 12h ago
A security researcher released an exploit called YellowKey that allegedly bypasses BitLocker encryption on Windows 11 and Server 2022/2025 by injecting a malicious folder into the Windows Recovery Environment, with a second exploit called GreenPlasma for privilege escalation also disclosed. The HN thread immediately split into two camps: one group argued this is a good thing because Microsoft aggressively pushes device encryption without user consent, locking people out of their own machines until they jump through cloud account hoops—several people recounted parents or non-technical users being trapped by blue screens demanding recovery keys they never knew existed. The other camp pushed back hard that BitLocker is essential for protecting laptops from theft, and that encryption plus proper backups is standard practice on Linux with LUKS, where users can print recovery keys or store them on USB drives without being locked to a motherboard. A significant tangent emerged questioning whether all big tech companies are in the same boat, with someone pointing out that Apple’s iCloud backups aren’t E2E encrypted by default either, giving law enforcement access to chats and browser history—so this isn’t uniquely a Microsoft problem, just the latest example of companies prioritizing government access over user sovereignty.
I don't think AI will make your processes go faster [comments]
497 points · 352 comments · frederickvanbrabant.com · 14h ago
The article argues that AI won't speed up organizational processes because the real bottleneck is upstream—vague requirements and incomplete specs—not the speed of writing code. The HN crowd largely agreed, with one top take bluntly suggesting that canceling pointless meetings with no agenda would do more for productivity than any AI workshop. A strong split emerged: several people reported that AI gives them a real but modest 10-20% speed boost on solo work, while others insisted that in larger teams with coordination overhead, AI-generated code becomes low-context noise that takes longer to review and reconcile than writing it yourself. The thread’s most skeptical voices pointed out that management is socially and financially incentivized to chase AI regardless, and that the real lesson—fix the bureaucracy first—keeps getting rediscovered every time a new technology comes along.
At least 25 Flock cameras have been destroyed in five states since April 2025 [comments]
415 points · 311 comments · stateofsurveillance.org · 9h ago
The article reports that at least 25 Flock Safety surveillance cameras have been destroyed across five states since April 2025, with incidents tied to public anger over the company's documented ICE connections—including over 4,000 immigration-related lookups by local police. The Hacker News thread split sharply: one camp openly celebrated the destruction as morally coherent resistance to a surveillance state they compared to a secret police force, while others dismissed the article as an attempt to normalize lawbreaking, with some accusing Reddit of astroturfing or suggesting Russian destabilization efforts. A significant chunk of the discussion veered into a broader debate about speed cameras and "safetyism," with arguments that surveillance creep turns every public safety measure into a vehicle for ICE enforcement and unconstitutional overreach. A minority pushed back with practical alternatives, like dismantling and returning cameras whole to avoid theft charges, or hacking them to feed bad data, though the dominant sentiment was that these cameras represent an intolerable erosion of Fourth Amendment protections.
Native all the way, until you need text [comments]
390 points · 261 comments · justsitandgrin.im · 14h ago
A veteran Apple developer describes spending months trying to build a Markdown chat app using SwiftUI, NSTextView, and TextKit 2—only to hit CPU spikes, selection bugs, and cell blinking—before discovering that a thrown-together Electron project handled streaming text, Markdown rendering, and system integrations perfectly out of the box. The thread split sharply: plenty of people nodded along, sharing their own horror stories with iOS text rendering or pointing out that WebKit is itself a native macOS framework (so embedding a web view for Markdown is totally fine, not a surrender to Electron). Others pushed back hard, arguing that the author’s specific UI demands (selecting an entire Markdown document) are unusual and that NSTextView or attributed strings have worked for decades if you know the older Cocoa stack. A smaller camp insisted the real problem is Apple’s neglect of AppKit and the half-baked SwiftUI, and pointed to Qt or Rust-based alternatives as saner cross-platform paths, though the memory-versus-lag tradeoff of Electron got its own heated sub-debate.
AI subscriptions are a ticking time bomb for enterprise [comments]
375 points · 374 comments · www.thestateofbrand.com · 14h ago
The article argues that AI subscriptions are an unsustainable loss leader and enterprise customers will face massive price hikes once OpenAI and Anthropic go public and need real margins. The HN thread immediately split on whether tokens are actually sold at a loss—some pointed to Brad Gerstner’s claim that inference is profitable, while others countered that he’s an interested party and that ignoring training, R&D, and capital expenditure is delusional. Several people pushed back by noting that open-source and Chinese providers (like GLM and Kimi) serve comparable models at a fraction of the price and presumably turn a profit, suggesting the big labs could be profitable on inference alone if they stopped the arms race in training. A smaller faction argued the real ticking bomb isn’t for enterprises but for AI investors—if the bubble bursts, the models and infrastructure still exist, and customers can just switch to cheaper alternatives or local models.
AI is a technology not a product [comments]
335 points · 136 comments · daringfireball.net · 13h ago
John Gruber’s piece argues that AI is pervasive infrastructure like wireless networking, not a product Apple needs to “ship,” and that the phone form factor isn’t going anywhere. Most of the thread backs him up, leaning hard on the “work backwards from the customer experience” Jobs/Amazon philosophy — one long comment chain debates what that actually looked like at Amazon, with presenter grilling and 6-page writeups instead of PowerPoint. But Siri gets dragged out repeatedly as the counterexample: Apple bought by Jobs himself that Apple let rot, and people note Apple’s early rightness turned into strategic atrophy. A split emerges over whether LLMs have materially improved a random non-coder’s life yet: the skeptics point to misinformation and cheap templates that existed before, while defenders rattle off translation, small-business websites, and a rational expert in your pocket. The broader consensus is that Apple is fine unless the phone stops being the main conduit, which nobody sees happening soon.
Apple Silicon costs more than OpenRouter [comments]
298 points · 254 comments · www.williamangel.net · 14h ago
A developer calculated the total cost of owning a $4,299 M5 Max MacBook Pro for local LLM inference and found it’s roughly 3x more expensive per million tokens than buying the same model on OpenRouter—plus 3–7x slower token speeds. The thread immediately jumped on the fact that you’re pricing the whole laptop as a token burner when most people already own a computer, and that buying a machine for inference then reselling it later means hardware depreciation isn’t a pure loss (Mac Studios with large RAM are reselling near MSRP because of supply shortages). Several engineers pointed out the analysis ignored input token costs, which dominate agentic workloads and are effectively free locally, and that cloud providers are burning VC money below cost—so the comparison flips once subsidies dry up. The real split wasn’t about the math but about whether speed and SOTA models outweigh privacy, sovereignty, and the ability to run uncensored models offline without rug-pulls or unpredictable bills.
WHO declares Ebola outbreak a global health emergency [comments]
277 points · 176 comments · apnews.com · 12h ago
The WHO declared the ongoing Ebola outbreak in Congo and Uganda a global health emergency, driven by the Bundibugyo strain which has no approved vaccine or therapeutics, and cases have reached capital cities. The discussion quickly zeroed in on how US aid cuts and the withdrawal from the WHO are hampering containment—some argued this is directly compounding the crisis, while others pushed back, saying it’s on regional governments to take responsibility, though that line got met with sharp retorts about teaching a neighbor to dig a latrine rather than crapping in your water supply. A big tangent spun out over whether a less deadly strain might actually spread further, with people diving into the virulence-transmission trade-off, using HIV as the go‑to counterexample of a high‑mortality virus that spreads because it takes years to kill you. There was also skepticism about the WHO’s insistence this isn’t a pandemic emergency, with speculation that the agency is trying to avoid perverse incentives for countries to hide outbreaks. The top‑voted link was a non‑paywalled AP version of the story.
I turned a $80 RK3562 Android tablet into a Debian Linux workstation [comments]
257 points · 120 comments · github.com · 12h ago
The project reverse-engineers an $80 Doogee U10 tablet to boot a full Debian Linux from SD card without touching the internal Android install, relying on extracted DTBs and AI-assisted debugging. The HN crowd was genuinely impressed by the technical achievement—especially the SD-card-first boot trick and what works (Wi‑Fi, touch, NPU)—but the thread quickly got hijacked by a heated meta‑argument about whether the author's top‑level comment was AI‑generated slop. Several people defended the project, pointing out that the AI use was disclosed, the replies were human, and the actual reverse‑engineering work was real; others argued that any AI‑written submission poisons the well, citing past fake projects that used LLMs for marketing copy. The underlying split is less about this specific tablet and more about whether a project should be judged solely on technical merit or punished for its presentation format.
Meta deletes popular 1M follower account after Kuwaiti request [comments]
188 points · 126 comments · twitter.com · 9h ago
Meta deleted a popular 1M-follower Instagram account after Kuwait requested it — but the thread quickly moved beyond that single event. The discussion split on whether this is just a corporation following local laws to protect market access, or a cover for Meta’s own political alignment (several people pointed out that the account promoted the Muslim Brotherhood, which many countries designate a terrorist organization). Others pushed back hard on the idea that sovereign governments like Kuwait have no leverage over tech giants, arguing that access to a market of millions and the ability to jail local employees gives them real power that Elon or Bezos can’t match with just money. A parallel thread zeroed in on Meta’s habit of citing vague “Community Standards” instead of admitting a legal order came from Kuwait — the criticism being that this secrecy lets the company dodge accountability while still doing the censorship.
Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep [comments]
178 points · 44 comments · github.com · 10h ago
The article presents Semble, an open-source code search tool that combines static embeddings with BM25 to dramatically cut token usage for AI agents searching codebases, claiming 98% fewer tokens than grep+read while maintaining near-parity with a larger transformer model. The HN thread quickly zeroed in on a crucial gap: the benchmarks only measure retrieval accuracy (NDCG), not end-to-end agent performance, and several people pointed out that token savings don't matter if the agent doesn't trust the results and burns tokens retrying or taking extra turns—one person even ran evals and found higher overall costs despite lower quality despite fewer tokens. The submitter acknowledged that end-to-end evaluation is on the roadmap, and anecdotally claimed Anthropic models seem to trust the tool. Others debated whether tools like LSP or RTK already solve this, and a few commenters pushed back on the Python implementation and the probabilistic nature of semantic search, noting that deterministic grep can't miss results but a tiny model can.
Tesla Solar Roof is on life support as it pivot to panels [comments]
168 points · 171 comments · electrek.co · 22h ago
The article reports that Tesla is effectively abandoning its Solar Roof tile product and pivoting back to conventional solar panels. The HN thread largely treated this as inevitable, with a strong consensus that the Solar Roof was never a serious product—many argued it was introduced in 2016 specifically to pump Tesla’s stock and bail out Musk’s cousins at SolarCity, pointing to a famously faked demo at a Hollywood backlot. A vocal minority pushed back against the cynicism, saying the company genuinely tried but failed to make the economics work: the tiles were too expensive, installation too complex, and the payback period too long for most homeowners, even though some owners genuinely loved the look. The conversation also took a sharp detour into whether Electrek’s editor Fred Lambert is a biased Tesla critic or a credible journalist, with several commenters accusing him of twisting facts to fit a negative narrative.
Ten Signs of Fascism. America has all of them [comments]
167 points · 84 comments · rutgerbregman.substack.com · 18h ago
The linked article wasn't available to this summarizer; from the discussion, the piece argues that the U.S. now meets all ten of a historian's criteria for fascism, drawing direct parallels to Nazi Germany's rise. The thread quickly split between those who accept the diagnosis as urgent and those who push back on the label, with several people pointing out that America's press is still critical, courts still push back, and there's no civil war-level violence yet, unlike Hitler's first month. A major line of argument rejects simple hope for elections, arguing that structural damage—gerrymandering, captured courts, media silos, and a GOP that stopped playing by democratic rules—means a Democratic win wouldn't fix the underlying rot, and that grassroots cultural change, not party loyalty, is what's needed. Others countered that the comparison is overblown and that democracies naturally generate doom-saying, but this was met with sharp replies about actual deportations and crackdowns on speech already underway, and a consensus emerged that Hungary under Orbán—not 1930s Germany—is the more precise model for the slow-motion, legalistic erosion of democratic norms happening now.
EU weighs restricting use of US cloud platforms to process government data [comments]
138 points · 66 comments · www.osnews.com · 8h ago
The EU is reportedly considering blocking US cloud providers from handling sensitive government data, pushing for European alternatives. The thread quickly split between those who argue on-premise or EU-based providers like Hetzner and Scaleway are perfectly capable, and a more skeptical camp pointing out that most European government IT is still deeply embedded in Microsoft and Google—one commenter noted their own municipal clients demand EU hosting while running Entra and Office 365 themselves. A recurring pushback came from people who actually run infrastructure, warning that on-prem is far from trivial and that the real bottleneck isn't cloud vs. not-cloud but the lack of a genuinely competitive EU cloud platform at AWS/Azure scale. Others pivoted to a darker angle: the US has already proven it will spy on allies, and the EU’s real problem isn't technical but political, with decades of regulatory paralysis and lobbyist capture making decisive action unlikely. One thread devolved into a practical nightmare about email deliverability, where leaving Gmail or Outlook can get your servers blacklisted with no recourse, suggesting that genuine digital sovereignty might require China-style bans rather than polite regulation.
CUDA Books [comments]
126 points · 26 comments · github.com · 13h ago
A GitHub user curated a massive list of CUDA books, from beginner guides like "CUDA by Example" to 2026 releases on debugging and C++26 optimization. The thread immediately pivoted to the tension between reading books and the corporate push to boost productivity with LLMs—several people lamented that companies want prompters, not coders, and one person joked about living on the edge by sneaking in an hour of reading. Others debated whether you should even write your own CUDA kernels anymore, with some citing Nvidia insiders who say don’t bother unless it’s your full-time job, while others pushed back hard, arguing that fusing kernels and low-level control still matter and that avoiding raw CUDA is like saying avoid C because Python exists. There was also concrete pushback on the list itself: a link was dead, "Programming Massively Parallel Processors" was called out for having confusing sentences and incorrect code examples despite being the most up-to-date, and one reader who teaches the subject announced he’s writing his own book next year to fix the gap.
VoIP brings back old-fashioned pay phones to rural Vermont (2025) [comments]
115 points · 34 comments · spectrum.ieee.org · 6h ago
A Vermont engineer is restoring old pay phones and running them over VoIP as free public phones, driven by the state’s upcoming smartphone ban in schools and spotty cell coverage. The HN thread immediately splits over semantics—several people insist “free-to-use pay phone” is an oxymoron and want the term “public phone” instead, while others push back that “pay phone” describes the hardware, not the payment model. A deeper argument breaks out over whether the FCC’s proposed Know-Your-Customer rules for robocallers could eventually require pay phone users to register ID; one commenter who works in telecoms is worried, but others clarify the rules target service providers and wouldn’t apply to a person walking up to a handset. Beyond policy, the thread drifts into nostalgia for memorized phone numbers and a sidebar praising Telstra for making Australia’s pay phone network free nationwide, with someone calling it a lifeline for abuse victims leaving their mobiles behind.
GenCAD [comments]
114 points · 23 comments · gencad.github.io · 4h ago
GenCAD is an MIT research project that turns images into editable parametric CAD command histories rather than just mesh files. The thread immediately got stuck on what the output actually is—turns out it’s a JSON payload based on DeepCAD/Onshape operations, not something native to a specific CAD program, which prompted pushback that “it doesn’t matter” because it’s the operational history, countered by people pointing out that storing mesh operation history is already possible in Blender. Several commenters pivoted to their own workflows: one argued OpenSCAD plus an LLM already does this painlessly, another pointed to zoo.dev’s KCL-based engine, and someone plugged a Rust BRep kernel wired up to Claude. The examples on the page were dismissed as too basic, and the auto-playing video got called out as annoying.
Playing Atari ST Music on the Amiga with Zero CPU [comments]
101 points · 37 comments · arnaud-carre.github.io · 18h ago
The article details a clever hack that uses the Amiga’s Paula chip in an “attached voice” mode to emulate the Atari ST’s YM2149 sound chip without any CPU overhead, even pulling off a role-swapping trick to get higher-quality output. The HN crowd mostly ignored the technical meat to launch into a nostalgic slugfest over whether the Amiga was genuinely ahead of its time or if PCs were just embarrassingly behind — people argued that AmigaOS had preemptive multitasking years before Windows 95, while others pointed out that the Amiga lacked memory protection and that OS/2 and Windows/386 already multitasked DOS apps. The thread also veered into a side debate about the Apple IIgs, which used an Ensoniq synth chip designed by the same guy behind the SID, and whether the Amiga’s custom chipset was a brilliant game machine or a dead end that made it hard to upgrade. The consistent takeaway: no one actually argued about the emulation technique; they just wanted to relitigate the 16-bit home computer wars.
Two EA-18 fighter jets collide at Mountain Home airshow, pilots ejected safely [comments]
98 points · 82 comments · idahonews.com · 4h ago
Two EA-18G Growlers collided during an airshow in Idaho, and all four crew members ejected safely — the core fact of the article. The thread immediately focused on how remarkable it was that the pilots got out alive given the tumbling orientation and low altitude, though several people pointed out that ejection almost always ends a flying career, with one citing a former Harrier pilot who came out two centimeters shorter. A major split emerged over whether these airshows are worth the risk and expense, especially for specialized electronic-warfare planes that cost 60% more than a standard F-18; defenders argued they're critical for recruitment and morale (and just plain cool), while critics called them propaganda spectacles that risk expensive assets and highly trained personnel for little more than a thrill. There was also technical speculation about the aircraft becoming stuck together by aerodynamic forces like paper on a windshield, which might explain why the pilots waited to eject instead of blowing up immediately.
Mistral's CEO: Europe has 2 years to stop becoming America's AI 'vassal state' [comments]
95 points · 155 comments · www.businessinsider.com · 10h ago
Mistral’s CEO told French lawmakers that Europe has about two years to build its own AI infrastructure or become a permanent “vassal state” to US tech giants, warning that control over chips, energy, and compute capacity will decide who wins. The HN thread largely split into two camps: one side argued that Europe is already hopelessly behind thanks to over-regulation, fragmented stock markets, and talent draining to the US, with several people noting that even Mistral’s models haven’t proven uniquely useful. Others pushed back hard, pointing out that US broadband and mobile markets are just as captured by regulatory cartels, and that Europe’s higher taxes and slower salaries are a trade-off for things like universal healthcare and not going bankrupt from cancer. A recurring counterpoint was that China’s DeepSeek models are open-weight and MIT-licensed, so the bottleneck isn’t AI models but the physical infrastructure—GPUs, cheap power, and data centers—where Europe’s fragmented energy grid and high electricity costs (outside France and the Nordics) put it at a real disadvantage. The conversation kept returning to the same uncomfortable question: even if Mistral succeeds, what’s stopping their best engineers from taking American stock options instead?
The occasional ECONNRESET [comments]
94 points · 21 comments · movq.de · 9h ago
The article walks through a real debugging session where a server closes a TCP socket with unread client data still in the buffer, which triggers a TCP RST instead of a clean FIN — and the HN thread's main takeaway is that this is absolutely not a mystery, it's been standard TCP behavior for over thirty years, codified in RFC 2525. Someone brings up `shutdown()` as a fix, but part 2 of the article apparently tested it and it didn't help, which kicks off a deeper discussion about the semantic difference between RST and FIN: RST doesn't just terminate the connection, it can discard hundreds of kilobytes of data still sitting in the client's receive buffer, meaning the client never gets the server's HTTP response. The thread pivots to how this bites HTTP servers like gunicorn that short-circuit request processing — if you return a 404 without reading the POST body, you've just nuked your own response — and why `Expect: 100-continue` exists specifically to prevent this scenario. A Go developer chimes in with a related story about connection reuse breaking when responses are discarded without being read, tying the whole thing back to how hard it is to get TCP's close semantics right in practice.
Canada’s Bill C-22 would weaken protections on private messages [comments]
93 points · 27 comments · dontsurveil.me · 9h ago
The article details Canada’s Bill C-22, which would force tech companies to build surveillance capabilities into encrypted messaging services, mandate bulk metadata retention, and include gag orders preventing disclosure. HN mostly agreed the bill is dangerous, but the discussion got tangled in meta-criticism of the advocacy site itself—multiple people called it a sloppily “vibecoded” AI-generated page with bloated formatting, arguing the poor presentation undermines the message. A few commenters pushed back on the absolutist “only you hold the key” framing, pointing out that lawful intercept was used effectively by Canada and the US to track the killers in the Sikh assassination case, suggesting there’s a legitimate use case that this narrative ignores. Others steered the thread toward broader examples of surveillance abuse, like NSA employees using “LOVEINT” to spy on romantic interests and Gulf states using WhatsApp data to arrest foreigners over private photos—making the case that once the infrastructure exists, it gets abused, not reserved for serious crimes.
Grafana Labs internal source code accessed [comments]
82 points · 26 comments · x.com · 22h ago
Grafana Labs revealed that an attacker used a stolen GitHub token to download their internal codebase and then tried to blackmail them—they chose not to pay. The HN crowd quickly split: some shrugged off the leak, arguing that Grafana's value is in its hosted service and support, not proprietary code (since most features like SSO are already open-source), while others pointed out that enterprise extensions and internal infrastructure could still expose vulnerabilities or be repurposed. A separate thread derailed into a heated debate about Grafana's job listings demanding “AI-native” experience, with multiple people calling out the absurdity of requiring months of expertise in tools that have only existed for a year. The discussion also got meta on security jargon—“threat actor” and “attack vector” drew eye-rolls, and one commenter dryly compared the blackmail refusal to the Dane-geld poem, warning that paying just invites more attackers.
Trials on veterans suggest ibogaine could provide a new treatment for PTSD [comments]
81 points · 89 comments · www.bbc.com · 14h ago
The BBC piece covers a Stanford study where special forces veterans with PTSD took ibogaine—a banned hallucinogen—in a Tijuana clinic and saw significant improvements in symptoms after a single session, though the drug carries cardiac risks and remains illegal in the US. On HN, the main split was between people who think ibogaine’s heart danger is manageable with proper screening and those who argue it’s reckless when safer psychedelics like salvia or DMT exist and might work via the same kappa-opioid mechanism. A lot of pushback focused on why the research and funding (Texas and Trump’s FDA executive order) target veterans specifically—some called it political posturing using a hallowed group to launder an agenda, while others pointed out that veterans have a dedicated healthcare system and high suicide rates, making them a practical and politically safe test population. A few commenters went on tangents about ECT being overlooked despite good results, and one long anecdote described a traveling ibogaine healer whose patients either recovered from heroin addiction or went insane—underscoring the split between those who see the intense “life review” as essential for change and those who think the effects are just temporary dissociation.
The AI water issue is fake [comments]
78 points · 56 comments · blog.andymasley.com · 7h ago
The article argues that fears over AI data centers' water consumption are overblown, citing tiny percentages of total U.S. freshwater use and favorable tax-revenue comparisons with golf courses. HN largely rejected that framing, pushing back hard that the piece glosses over local impacts, regulatory evasion, and environmental injustice—commenters pointed to specific cases where data centers violated water regulations with impunity, got approvals over local protests, and disproportionately affected disadvantaged communities. Several called the article's approach reminiscent of tobacco-industry spin, funded by effective altruism sources, and noted that the real issue isn't global water totals but strained local aquifers in arid regions where these centers are built. The submitter themselves acknowledged the article felt biased and asked for the community's take, which overwhelmingly shifted the discussion toward unregulated industry power and concrete local damage rather than aggregate arithmetic.
An AI Hate Wave Is Here [comments]
73 points · 91 comments · www.axios.com · 4h ago
The Axios piece reports on polling showing that over 70% of Americans now think AI is advancing too quickly, with negative views rising from 34% to 50% in three years, and data center cancellations spiking amid community resistance. Hacker News immediately split into two camps: one arguing the backlash is really about a lousy economy and housing unaffordability, not AI itself—calling AI a convenient scapegoat for systemic failures—while the other pushed back by citing data that Gen Z actually has higher inflation-adjusted incomes and homeownership rates than previous generations at the same age. A parallel thread debated whether AI’s job threat is contained to tech or is eating into law, accounting, and education, with one commenter pointing out that senior lawyers will protect their own licenses while junior associates and self-represented litigants get squeezed. Others veered into blaming tech CEOs’ bizarre public personas for stoking the hate, arguing the substance of their views is genuinely repellent, not just poorly marketed.
XS: A programming language. Anywhere, anytime, by anyone [comments]
60 points · 38 comments · xslang.org · 11h ago
The submission pitches XS as a programming language packed into a single 2.9MB binary with an absurdly broad toolkit — compiler, debugger, formatter, JIT, transpiler to C and JS, all running on everything from ESP32 to iOS. HN quickly zeroed in on the fact that the entire website and documentation read like AI slop, with no identifiable human author or provenance, which killed any trust for most readers. A deep-dive commenter tore into the codebase, finding two parallel type inference systems, a VM cribbed from *Crafting Interpreters*, and telltale LLM context spew (like an in-code note that a section was deleted "because benchmarks showed tier-2 dominated"). Others split hard on whether the lack of a human behind it matters if the language actually works, but the consensus was that the GIL in 2026 and the half-baked type system make it uninteresting regardless. A separate tangent emerged when Perl developers pointed out the name collision with Perl's XS extension system, causing confusion for anyone who clicked thinking it was about Perl bindings.
Prolog Coding Horror [comments]
59 points · 21 comments · www.metalevel.at · 4h ago
The article is a guide from Prolog expert Markus Triska on common anti-patterns—like using cuts, assert/retract, and raw arithmetic—and replacing them with pure monotonic constructs like dif/2 and CLP(FD) constraints for full generality. The thread quickly pivoted to the perennial “who actually uses Prolog in the real world?” question, with replies pointing to Datalog (as used in Datomic, the Polonius borrow checker in rustc, and graph DBs) and drawing a sharp semantic line between Prolog and Datalog’s declarative guarantees. One person pushed back, arguing the article’s first “horror” (accepting some wrong answers for generality) is overblown, while another defended it by noting that in practice (e.g., sales forecasting returning zero vs. an approximation) pragmatic imperfection can be more useful than a hard failure. A side tangent celebrated the four-port model as essential to understanding Prolog’s solution-space navigation, which devolved into a joke about smartphones losing ports—but the serious takeaway was a reminder that Triska isn’t a beginner discovering Prolog’s quirks; he’s the author of its core constraint libraries, so the advice carries weight.
Design posters showcasing your country's electrical grid [comments]
57 points · 15 comments · github.com · 3h ago
The submission is an open-source Python tool that turns OpenStreetMap transmission line data into print-ready posters of any country's electrical grid. HN had fun with the phrasing, with one person admitting they'd hoped for actual propaganda posters about the grid, but the real discussion quickly turned to whether falling solar and battery costs might eventually let people disconnect from it entirely—a back-and-forth where one side argued utility connections still offer flexibility for maintenance and repairs, while the other pointed out that rising utility rates could push more households to go off-grid, leaving remaining customers with higher costs. Several people dug into the Africa example poster, pointing out that Africa isn't a country, though others noted the tool handles continents and predefined regions just fine. Someone also linked OpenInfraMap, leading to a comment about tracing a major power line cutting through a New Jersey neighborhood and finding the line specs fascinating.
Generated 2026-05-18 02:46 UTC
Generated by Sauron from Hacker News discussions and linked articles.