HN Brief: 2026-05-18
Today's HN was dominated by a rolling argument about AI trust—or rather, the lack of it. An Axios poll showed only 18% of young people feeling hopeful; a commencement speaker got booed for telling graduates they'll "work for AI"; and most Americans say they don't trust the technology or the people building it. Meanwhile, the security front delivered two firestorms: a researcher released a Bitlocker exploit that bypasses full-disk encryption, and Flock cameras are being destroyed across five states over ICE ties. A scattered third theme was the growing tension between Europe and US tech giants, from cloud sovereignty rules to Mistral's warning about becoming a "vassal state."
Start with "Security researcher says Microsoft built a Bitlocker backdoor, releases exploit"—the inevitable flamewar over whether it's a gift or a vulnerability is worth jumping in for the encryption-versus-convenience debate alone. "At least 25 Flock cameras have been destroyed in five states since April 2025" splits the room cleanly on civil disobedience versus rule of law. "Native all the way, until you need text" nails every Apple developer's nightmare with SwiftUI and TextKit for streaming UIs. "AI subscriptions are a ticking time bomb for enterprise" offers the clearest look yet at whether the VC-subsidized token party ends soon. And "TWO EA-18 fighter jets collide at Mountain Home airshow, pilots ejected safely" is the human-meets-hardware story of the day—ejection seats that work in impossible orientations, and the crowd split on whether airshows are worth it.
I don't think AI will make your processes go faster [comments]
566 points · 391 comments · frederickvanbrabant.com · 19h ago
The post argues that throwing AI at organizational processes won't speed things up because the real bottlenecks are upstream — vague requirements, coordination overhead, and unclear inputs — and that AI-generated code just forces domain experts to handhold through endless specification loops. The thread largely agreed with the premise, with many noting that bureaucracy, not typing speed, is what slows everything down, and that AI often just produces low-context output that needs careful review and reconciliation. There was a sharp split, though: some insisted AI gives real 10-50% speedups for individual contributors, especially in small teams or solo projects, while others countered that those gains vanish in larger orgs where code reading and team coordination dominate, and that "vibe coding" produces unmaintainable commits that make cooperation impossible. A few people pushed back hard against the article’s claim that AI code is rarely correct, arguing that when you know your codebase and use a language with strong idioms like Rust, the output is almost always right — just not placed where you'd like it. And a resigned undercurrent ran through it all: these posts keep appearing, but leadership is too socially and financially incentivized to pretend AI is magic, so the real action is waiting for the mania to burn out.
Security researcher says Microsoft built a Bitlocker backdoor, releases exploit [comments]
565 points · 257 comments · www.techspot.com · 18h ago
A security researcher claims to have found a backdoor in BitLocker on Windows 11 (and Server 2022/2025) called YellowKey that lets an attacker bypass full-disk encryption by booting into the Windows Recovery Environment with a specially crafted USB folder; third parties have confirmed it works. The HN thread immediately split into two camps: one side saw it as a gift from God, pointing out that Microsoft has been tricking users into encrypting their drives without consent via dark patterns during account setup, effectively holding their data hostage unless they find the recovery key. The other side pushed back hard, arguing that this is just a vulnerability, not a deliberate backdoor, and that BitLocker’s recovery mechanisms have been well-documented for nearly two decades—users who don’t save their recovery key are just negligent. A big chunk of the discussion veered into personal threat models, with some arguing they don’t want encryption at all because they want to yank the drive and read it after a hardware failure, while others countered that proper backup strategy makes encryption free and that dm-crypt on Linux handles this more sanely. The thread also got into whether security professionals should refuse to work on Microsoft products given recurring “greed and tech debt,” with a reminder that Apple and Google have their own surveillance-friendly defaults.
At least 25 Flock cameras have been destroyed in five states since April 2025 [comments]
426 points · 313 comments · stateofsurveillance.org · 14h ago
The article reports that at least 25 Flock Safety cameras have been destroyed across five states since April 2025, driven by public anger over the company's documented ties to ICE and the Fourth Amendment concerns of one Virginia man who took down 13 of them himself. HN immediately split on whether this is justified civil disobedience or dangerous vigilantism: a vocal contingent argued the cameras are a tool of an unaccountable secret police and destroying them is morally coherent, while others pushed back hard, calling the article an attempt to normalize lawbreaking and warning that property rights and rule of law erode when people take matters into their own hands. A few commenters derided the article's reliance on Reddit sentiment as proof of popular support, triggering a meta-debate about astroturfing, bot accounts, and whether HN itself is sliding into political shilling. The thread also veered into a broader fight about speed cameras—some defended them as a minor evil for road safety, others insisted they're a Trojan horse for mass surveillance that gets abused by ICE and local governments, with one lengthy takedown arguing pedestrian deaths are a street design problem, not a speed problem.
Native all the way, until you need text [comments]
418 points · 276 comments · justsitandgrin.im · 20h ago
A veteran Apple developer detailed his Sisyphean attempt to build a Markdown-rendering chat app purely with SwiftUI, AppKit, and TextKit, only to conclude that even after dropping down to raw TextKit 2, streaming text and basic selection either broke or required months to match baseline macOS behaviors—while an Electron prototype “just worked” out of the box. The thread largely validated his pain: several people who built AI chat apps chimed in to confirm that every native text component they tried on iOS and macOS was buggy, laggy, or incompatible with streaming. The pushback wasn’t denial of the problem, but rather a split on the solution—some argued that WebKit is itself a native framework and using it for Markdown is perfectly appropriate, while others insisted the author’s specific demand (selecting an entire Markdown document) is an edge case that doesn’t justify abandoning native primitives. A few skeptics questioned his code quality, but they were quickly countered by commenters recalling how even trivial tasks like rendering a clickable link in a paragraph have been a nightmare on Apple platforms for years. The consensus: Apple’s toolchain for rich, streaming text is genuinely broken for modern chat UIs, and the community is still searching for a lightweight cross-platform C or Rust library that can match what a web view gives you for free.
AI is a technology not a product [comments]
402 points · 175 comments · daringfireball.net · 18h ago
John Gruber’s piece pushes back on Steven Levy’s Wired article saying Apple’s next CEO needs a “killer AI product,” arguing instead that AI is just a technology—like wireless networking—that should be invisible inside great products, not something to ship as its own device. The thread largely agrees with Gruber, with many pointing to Steve Jobs’ old “work backwards from the customer experience” talk as the definitive counter to the hype, though a few commenters note Apple botched Siri, which was supposed to be that invisible AI layer and instead got worse. There’s a strong split on whether AI will actually improve the average person’s life: one side says LLMs already give small-town businesses websites and break down gatekeeping, the other insists the technology hasn’t delivered anything non-crying-worthy and risks flooding the world with misinformation. A tangent that surprised me was a long, detailed digression on Amazon’s “no PowerPoint, 1-6 page document, intellectual bloodbath” meeting culture—prompted by someone noting Amazon also uses “work backwards”—with people debating whether that actually works outside a low-politics environment. Overall the thread treats Gruber’s take as mostly correct, but the real tension is whether Apple can keep the phone as the primary conduit for AI without someone else inventing a new form factor that makes the phone feel obsolete.
AI subscriptions are a ticking time bomb for enterprise [comments]
393 points · 384 comments · www.thestateofbrand.com · 20h ago
The article argues that AI labs are running a massive loss-leader scheme, subsidizing enterprise subscription pricing is nowhere near covering real inference costs, and a reckoning is coming as IPOs force price hikes or usage caps. HN immediately split on the core premise: some pointed to Brad Gerstner’s claim that tokens are already sold at a profit, but others shot that down as interested-party spin, noting that even if inference makes money, the costs of training, R&D, and debt-backed infrastructure are astronomical and have to be recouped eventually. A strong contingent pushed back with the open-source counterargument — models like GLM, Kimi, and Qwen run on OpenRouter for a fraction of the price, so if those providers can make money at those rates, the idea that Claude or GPT is unprofitable to serve starts looking shakier by the minute. The thread also wrestled with whether this is actually an enterprise time bomb or an investor time bomb: if AI companies stop training and just run inference, they’d be profitable, and the real risk is that the bubble pops and the subsidies vanish, leaving companies that hard-wired cheap AI into their workflows.
I turned a $80 RK3562 Android tablet into a Debian Linux workstation [comments]
340 points · 153 comments · github.com · 18h ago
A developer reverse-engineered a $80 Doogee U10 tablet to run full Debian 12 from an SD card without unlocking the bootloader or modifying the internal Android storage. The thread quickly split into two camps: one side accused the submitter of using AI-generated comments and readme text, sparking a wider debate about whether LLM-assisted projects should be judged by technical merit or dismissed as slop. Others pushed back, pointing out that the AI use was disclosed upfront, the submitter's replies were clearly human-written, and the actual hardware hacking—UART debugging, contacting chip factories in Chinese, iterating on DTBs—was genuine. Technical discussion centered on why the RK3562 boots from SD first (BootROM priority, no SPI flash) and whether the same approach could target internal eMMC, while a separate typographic argument erupted over whether em dashes should have spaces around them.
Apple Silicon costs more than OpenRouter [comments]
312 points · 269 comments · www.williamangel.net · 19h ago
The article crunches the numbers on running LLM inference locally on an M5 MacBook Pro versus paying for OpenRouter, concluding that depreciation on the hardware makes local cost about 3x higher per million tokens. The thread pushed back hard: several people pointed out that the author only compared output token costs, but ignored input tokens, which dominate agentic workloads and are essentially free locally due to fast prompt processing and caching. Others argued the analysis is misleading because it treats a $4,300 laptop as a dedicated inference server rather than a laptop you’d already own—if you need the machine anyway the marginal cost of inference drops drastically, and the resale value on these Macs is strong due to the RAM shortage. A recurring theme was privacy, control, and resilience: even if cloud is cheaper and faster, many find peace of mind in owning their entire stack and not being subject to VC-subsidized pricing that could vanish. The consensus split cleanly between cost accountants who said the math is correct for a dedicated rig and pragmatists who said the real-world economics favor local for anyone who already has a capable machine or cares about data sovereignty.
Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep [comments]
302 points · 104 comments · github.com · 16h ago
The submission is Semble, an open-source code search tool for AI agents that claims 98% fewer tokens than grep+read by combining static embeddings with BM25 on CPU. The thread immediately questioned whether those token savings hold up in an actual agent loop—several people reported that their agents, heavily trained on grep patterns, often ignore or retry alternative tools like Semble, burning through tokens anyway. The authors agreed that full agent benchmarks are the missing piece and said they’re working on them, though a few who tried Semble directly with Claude Code said it worked well right away. Other tangents pitted Semble against LSP-based approaches and tools like RTK or ck, with one user running a concrete test where Semble indexed the rust-lang/rust repo in 26 seconds while ck would take days. The split came down to short-term retrieval metrics looking good versus the harder question of whether agents actually trust the results enough to end-to-end cheaper.
WHO declares Ebola outbreak a global health emergency [comments]
298 points · 184 comments · www.nytimes.com · 18h ago
The World Health Organization just declared an Ebola outbreak in Congo and Uganda a global health emergency, driven by a strain of the virus (Bundibugyo) that has no approved vaccine or therapeutics. The thread quickly zeroed in on the lack of a vaccine for this specific strain and argued about whether the existence of a vaccine for the Zaire strain makes developing one for Bundibugyo easier—with some pushing back that the real bottleneck is medical distrust in Africa fueled by a history of colonial experimentation. Several people pointed out that this outbreak appears to have been detected late, and they directly blamed the Trump administration’s dismantling of USAID and withdrawal from the WHO for crippling surveillance and response capacity. A recurring debate emerged over the trade-off between virulence and transmissibility, with commenters using HIV as a counterexample to the idea that less deadly necessarily means more spread, while others dug up case fatality data showing the current 33% rate is actually within historic range for this strain. There were also pointed criticisms of the WHO for hesitating to declare a pandemic emergency, with some arguing that economic and political pressure from member states routinely delays such declarations.
GenCAD [comments]
278 points · 66 comments · gencad.github.io · 10h ago
GenCAD is an MIT research project that generates full parametric CAD command sequences—not just meshes—from input images, preserving the editable history of operations. The HN thread quickly zeroed in on the practical question of which CAD program it actually outputs to, with domain experts explaining that portability is fundamentally limited by different geometry kernels and tolerance handling, which is why professional CAD relies on neutral formats like STEP that discard command history. Several commenters pushed back by pointing to existing LLM-driven workflows with OpenSCAD and tools like vibe-modeling, arguing those already solve the "generate from description" problem more transparently, though others countered that OpenSCAD’s CSG approach can't match the B-rep modeling needed for real engineering. The consensus was that the demos look trivial—basic cubes and extrusions—and that the hard part of CAD isn't generating shapes but getting correct dimensions, constraints, and tolerances, making this feel like a solution in search of a problem unless it can handle real-world complexity.
Meta deletes popular 1M follower account after Kuwaiti request [comments]
194 points · 138 comments · x.com · 14h ago
The linked article wasn’t available to this summarizer; from the discussion, Meta deleted a popular account with over a million followers at the request of Kuwait, with the company citing vague "Community Standards" instead of the actual legal order. The thread immediately split: some argue Meta is just following local laws to keep access to a market, no different from how any corporation operates, while others insist this is selective censorship — pointing out that Meta doesn’t ban accounts for homosexuality in Kuwait, so the real reason is the account’s pro-Muslim Brotherhood stance, which aligns with Meta's own political biases. A loud faction defends the deletion by labeling the Muslim Brotherhood a terrorist organization and fascist, dragging the conversation into a heated back-and-forth about Zionism, ethnic cleansing, and whether criticizing Israel is antisemitic. Meanwhile, a few commenters note that Elon Musk and Jeff Bezos have far more power to influence Meta than a small Gulf state, but that misses the point: sovereign governments can threaten your local employees and block your entire market, which is a lever no billionaire has.
Two EA-18 fighter jets collide at Mountain Home airshow, pilots ejected safely [comments]
187 points · 173 comments · idahonews.com · 10h ago
Two Navy EA-18G Growlers collided during an airshow at Mountain Home Air Force Base, and all four crew members ejected safely—which the thread treats as borderline miraculous given the spinning, entangled way the aircraft in the video. Most of the discussion zooms in on the mechanics: how ejection seats can function in almost any orientation, why the pilots might have delayed ejecting (hoping to break free), and the brutal physical toll even a successful ejection takes, with commenters noting spinal compression fractures and career-ending injuries. A separate, heated split emerges over whether airshows are worth the risk and expense, especially when two expensive electronic-warfare platforms go down for a display—some argue it’s necessary for recruitment and PR, others call it an expensive spectacle with a non-zero fatality rate that keeps happening for no great military reason. The thread largely agrees that these pilots were incredibly lucky, and the escape systems, performed remarkably, but it's less consensus on whether the show itself was a good idea.
CUDA Books [comments]
183 points · 38 comments · github.com · 19h ago
The submission is a curated GitHub list of CUDA programming books spanning from 2010 beginner titles to 2026 releases. The thread quickly turned skeptical: several veteran commenters pointed out that the most popular book on the list, "Programming Massively Parallel Processors" is riddled with errors and confusing prose, and that many of the older titles (like "CUDA by Example") overabstract too much over modern architecture to be useful now. A heated split emerged over whether it’s even worth writing raw CUDA kernels anymore — one camp argued that Nvidia’s own architects recommend reaching for higher-level libraries unless kernel hacking is your day job, while others countered that abandoning CUDA optimizations just hands more lock-in to Nvidia and misses the whole point of learning to fuse and tune. The most useful takeaway wasn’t a book recommendation at all: a CUDA architect’s highly condensed video walkthrough was promoted as the single resource that covers everything you actually need.
VoIP brings back old-fashioned pay phones to rural Vermont (2025) [comments]
148 points · 44 comments · spectrum.ieee.org · 12h ago
A Vermont engineer is converting old pay phones to free VoIP phones, installing them in rural schools and town halls where both cell service and smartphones are scarce. HN spent most of the thread arguing about the definition: several people insisted "free-to-use pay phone" is an oxymoron and pushed for "public phone," while others pointed out that "pay phone" just names the hardware form factor, not the payment model. The biggest substantive debate was over an FCC proposed rule—some worried it would require payphone operators to collect ID from every caller, but a commenter in telecom pushed back, clarifying that the rule targets robocallers signing up for service, not casual passersby using a public handset. A separate thread praised Telstra’s free payphone network in Australia as a lifeline for abuse victims and a source of public WiFi, with one person noting the US would likely “fuck it up” or “ring every penny from the stone” instead.
EU weighs restricting use of US cloud platforms to process government data [comments]
144 points · 68 comments · www.osnews.com · 14h ago
The EU is reportedly considering rules to stop its member governments from using US cloud platforms like AWS, Azure, and GCP for sensitive data, with the OSNews author arguing the US has forfeited trust and the EU needs to act. The thread quickly split into two main camps: one side insisted that European providers like Hetzner, Scaleway, and OVH are already production-ready and that on-premises government data centers are simpler and cheaper than the hype suggests, while the other side pushed back hard that running real-scale government IT is nothing like a hobbyist Debian box in a basement, pointing to skills shortages, disaster recovery failures (like South Korea losing years of data), and the genuine flexibility cloud offers for unpredictable workloads. Several people noted that even if EU datacenters host the data, US companies still control the software and remain subject to American extraterritorial laws like the CLOUD Act, making any sovereignty claim hollow unless the entire stack is replaced. The discussion also highlighted an irony: the EU itself requires citizens to have Google or Apple accounts for upcoming digital ID and age-verification apps, which would lock out anyone using degoogled Android or dumbphones, and the thread’s bottom line was that the real barrier isn’t technical capability but political will and the deep entanglement of US services in every layer of government operations, from email to Entra ID to DNS.
Prolog Coding Horror [comments]
121 points · 49 comments · www.metalevel.at · 10h ago
The article catalogs common Prolog pitfalls—like using cuts, asserts, and impure arithmetic—and advocates for declarative alternatives such as `dif/2` and CLP(FD) constraints. The thread quickly diverged into a debate about whether Prolog is actually used in production, with several people pointing to Datalog (used in Datomic, Polonius, and graph databases) as the more practical logic-programming cousin, and pushing back hard on the idea that Datalog is just a subset of Prolog—it's a fundamentally different computational model. One side called the article’s horrors “overblown,” while another defended the author (Markus Triska, a Prolog heavyweight) and argued that accepting logically impure answers is a pragmatic trade-off when P ≠ NP. A separate tangent compared Erlang to Prolog, but the consensus was that they share only surface syntax—Erlang is functional, not logic-driven.
The occasional ECONNRESET [comments]
108 points · 24 comments · movq.de · 14h ago
The post digs into a maddening ECONNRESET bug between two services talking over TCP on localhost, showing how the server closing a socket while the client still had unread data in the kernel buffer triggers an immediate RST — not a clean FIN. HN immediately recognized this as a well-documented TCP gotcha that’s been around since the 90s, with people pointing straight to RFC 2525 and 1122 to confirm the server is supposed to send a RST when data is lost. The discussion quickly got into the weeds about why RST is preferable to a hung connection (the old FIN behavior could permanently stall an FTP server if the client vanished), and whether `shutdown(SHUT_WR)` plus draining the socket is the real fix — the author’s part 2 apparently tried that and found it didn’t help, which sparked more debate over lingering closes and event-loop architectures. There was also a side note about Go’s HTTP client refusing to reuse connections when the response body isn’t fully read, since the same RST behavior kills the connection pool.
WriteUp: 16 Bytes of x86 that turn Matrix rain into sound [comments]
107 points · 23 comments · hellmood.111mb.de · 8h ago
A 16-byte MS-DOS demo generates a Sierpinski fractal from video memory and plays it directly through the PC speaker, producing surprisingly coherent music. The thread is full of people baffled that such a tiny loop of x86 instructions — just an XOR, a few pointer moves, and a port write — can produce something that sounds like actual rhythm and melody rather than static. The author jumps in to explain that the sound’s character partly comes from BIOS memory initialization quirks, which vary by machine and emulator, turning each run into a unique audiovisual fingerprint. A few people note the “Matrix rain” in the title is misleading — the visual is the fractal sheared diagonally, not falling characters — but the author clarifies the sound came first and the screen just shows what you’re hearing. Comparisons to other tiny demos like “A Mind Is Born” and “Rainy 32b” pop up, but the consensus is that this is mind-bendingly dense code that still manages to be musical.
Mistral's CEO: Europe has 2 years to stop becoming America's AI 'vassal state' [comments]
100 points · 178 comments · www.businessinsider.com · 15h ago
The Mistral CEO told French lawmakers Europe has two years to build its own AI infrastructure or become permanently dependent on US tech giants, framing it as a sovereignty crisis where chips, energy, and computing capacity are all controlled by American companies. The thread largely agreed Europe is already in third place behind the US and China, with many arguing that fragmented regulations, high power costs, and a talent drain to American salaries make catch-up nearly impossible. Some pushed back on the tired “America innovates, China replicates, Europe regulates” line, pointing out that US regulatory capture is just as bad and that European consumer protections have real value. Others zeroed in on infrastructure: even if you run open-weight models like DeepSeek, Europe still needs GPUs, cheap electricity, and coherent capital markets, and the comments were split on whether Mistral’s models are genuinely competitive or just riding a political wave to raise more money. The sovereignty pitch resonated with some who prefer democratic tech, but the practical hurdles—stock option complexity across member states, energy prices, and the sheer scale of US investment—dominated the skepticism.
The History of ThinkPad: From IBM’s Bento Box to Lenovo’s AI Workstations [comments]
95 points · 45 comments · www.jdhodges.com · 10h ago
The article traces the ThinkPad’s evolution from IBM’s 1992 “bento box” 700C through Lenovo’s 2026 AI workstations. Hacker News readers dove into personal war stories—veterans still running P50s as Proxmox servers, fans mourning the dropped 7-row keyboard and dedicated TrackPoint buttons, and one self-described “button man” whose entire comment was a digression into Ted Selker’s obsessive spare-nub bag. A chunk of the thread paused to accuse the article itself of being AI-generated slop, citing length and generic phrasing, which prompted a debate about whether HN is overrun with LinkedIn-style bots. The main pushback from the crowd: new ThinkPads are overpriced for the specs, but the real value is in buying gently used models on eBay for a fraction of retail—a formula that still delivers a rugged Linux machine with a legendary keyboard.
Canada’s Bill C-22 would weaken protections on private messages [comments]
94 points · 27 comments · dontsurveil.me · 15h ago
The linked article warns that Canada’s Bill C-22 would require every messaging app to build a way around end-to-end encryption, giving the government a copy of the key. The thread quickly split into two camps: people who agreed with the substance but trashed the site itself as a “vibecoded” AI-generated mess with bloated copy and distracting design, and a longer debate about whether surveillance powers ever actually target serious crime—someone cited UK police making 12,000 arrests a year for online speech, though others pointed out that includes threats and harassment, not just “mean tweets.” A few commenters noted that Signal has already said it would rather leave Canada than comply, and raised the old TSA master-key leak as a cautionary tale for mandated backdoors. The thread also veered into a fight over wage theft versus retail theft numbers and whether locking up repeat offenders actually fixes anything, but the core takeaway was that the bill is moving fast through committee and nobody on the public record except the government itself is defending it.
Trials on veterans suggest ibogaine could provide a new treatment for PTSD [comments]
93 points · 97 comments · www.bbc.com · 20h ago
The article reports on a Stanford study where U.S. special forces veterans with PTSD received ibogaine—a banned hallucinogen with cardiac risks—in a Tijuana clinic and saw significant symptom improvements after a single session. Hacker News immediately split on why ibogaine was chosen over safer alternatives like psilocybin or LSD, with some pointing to Texas funding and others arguing that ibogaine’s action on kappa-opioid receptors and neuroplasticity is uniquely suited for traumatic brain injury, not just PTSD. A long-running side debate questioned why the research focuses on veterans instead of the much larger population of assault survivors, with several people noting it’s a political move to get bipartisan support—though others countered that the VA simply provides an easier recruitment pipeline. An old anecdote about a traveling ibogaine healer who saw half his patients cured and the other half go insane added a stark real-world note, while a separate thread critiqued the hype around psychedelics compared to ECT, which several argued is safer and better studied despite its stigma.
Most Americans don't trust AI – or the people in charge of it (2025) [comments]
86 points · 73 comments · www.theverge.com · 6h ago
The Verge reports on Pew and Gallup data showing most Americans don’t trust AI or the people running it, with a huge optimism gap where 75% of experts think AI will benefit them personally but only 25% of the public agrees. On HN, the thread immediately questioned whether the distrust is really about AI or just a broader collapse of trust in institutions, government, and big tech companies that have been making products worse for a decade. A significant split emerged: some people rattled off genuinely useful everyday applications (identifying birds, drafting recipes, museum tours), while others insisted that outside of programming, no normal person has seen AI do anything but hallucinate and ruin their dinner. The real pushback came from those pointing out that AI companies are burning their social license by building resource-hogging data centers that guzzle water, drive up power prices, and dump GHG emissions on communities that see none of the profit. A few commenters also challenged the polling itself, arguing the questions were loaded and that their own non-tech friends are using AI constantly, but others retorted that you won't see that outside of hip coffee shops in big metro areas.
An AI Hate Wave Is Here [comments]
83 points · 105 comments · www.axios.com · 10h ago
The Axios piece reports that AI sentiment has turned sharply negative, with polls showing only 18% of young people feeling hopeful and over 70% of Americans thinking AI is advancing too fast, punctuated by a Florida commencement speaker getting booed for calling AI "the next Industrial Revolution." Hacker News largely sidestepped the polling to argue over *why* people hate AI: a strong faction insists it's not really about AI but a misdirected anger at a stalled economy, housing unaffordability, and corporate behavior—AI is just the scapegoat being trotted out by execs to explain why life sucks. Others pushed back hard against that economic pessimism, citing Economist data that Gen Z has higher inflation-adjusted incomes and homeownership rates than millennials at the same age, calling the doom narrative "folk wisdom" fed by TikTok. A separate vein of the thread dug into job displacement, comparing AI to the Industrial Revolution's looms and noting that while lawyers and accountants might be protected by licensing laws, junior associates and knowledge workers are already being automated out of the pyramid. The strongest consensus was that the real friction isn't the technology itself but the corporations' clumsy, tone-deaf insistence that it's inevitable—execs saying "we don't really see that" backlash while communities cancel data centers and investors lose confidence.
The AI water issue is fake [comments]
81 points · 58 comments · blog.andymasley.com · 13h ago
The article argues that AI data-center water use is a tiny fraction of overall US freshwater consumption—0.008% for AI in 2023—and that alarmist headlines ignore how most reported “water use” comes from electricity generation rather than on-site cooling, making it a non-issue blown out of proportion. HN immediately pushed back hard, calling the piece misleading and likening it to tobacco-industry PR, with several people pointing to specific cases where data centers have circumvented environmental regulations, drained stressed local aquifers, and been built in disadvantaged communities without meaningful oversight. A sharp split emerged: some commenters insisted the real problem is not the raw volume but the lack of regulation and proper water pricing, while others countered that national aggregates obscure severe local scarcity in arid regions like Arizona and Georgia. The thread also veered into a broader debate about whether the entire AI boom is being granted a regulatory free pass in the name of national competitiveness, with several people arguing that comparing water use to golf courses misses the point that data centers often get tax breaks and don't generate proportional local employment or revenue.
Design posters showcasing your country's electrical grid [comments]
80 points · 18 comments · github.com · 9h ago
Grid2Poster is an open-source Python tool that turns OpenStreetMap power-line data into print-ready posters of any country's or continent's electrical grid. The thread immediately veered into a joke about wishing these were actual propaganda posters for or against your country’s grid, then got sidetracked by someone pointing out that Africa is not a country—a point the project’s README already addresses by explicitly supporting continents. Several people shared their own generated posters, including one from Japan who noted the tool left in southern islands that aren’t part of the actual grid and had to edit the image, then dropped the interesting detail that Japan runs on 50 Hz in the east and 60 Hz in the west. Others linked to OpenInfraMap for a more interactive look, and a short debate emerged about whether falling solar-and-battery costs might ever make utility grids obsolete, with the counterpoint that a grid connection still offers flexibility for maintenance and repairs.
Don’t Outsource the Learning [comments]
68 points · 34 comments · addyosmani.com · 15h ago
Addy Osmani’s piece argues that leaning on AI defaults quietly erodes your engineering skills because the tools optimize for shipping, not teaching — and he backs it with studies showing comprehension drops and a phenomenon he calls “cognitive debt.” The thread took a sharp turn into hypocrisy, though, because several people dug up Osmani’s own AI-generated bio, which is clumsily boastful and reads like a press release, making his warning feel insulting. Others pushed back on the premise itself: if AI keeps getting better, why bother learning skills that will be irrelevant tomorrow? Some compared it to the calculator panic, arguing we’ve always outsourced thinking to tools. A few commenters struggled with the practical advice — constantly prompting for explanations still leaves your mental map fuzzy — and wondered if the real fix is tolerating that anxiety rather than chasing full technical grasp.
XS: A programming language. Anywhere, anytime, by anyone [comments]
60 points · 38 comments · xslang.org · 17h ago
XS is a programming language packed into a single 2.9MB static binary that includes a compiler, debugger, LSP, formatter, and more, touting cross-platform portability from desktop to ESP32. The HN thread quickly split: some people were intrigued by the "busybox for languages" idea and modern features like algebraic effects and actors, but a much louder contingent tore into the project as AI-generated slop, pointing to LLM-isms in the code comments and docs, and questioning the type system's coherence (e.g., two parallel inference engines, an undecidable mix of Hindley-Milner and subtyping, and a broken `generalize` that leaves polymorphic types as monotypes). Others dismissed the whole thing on principle—if there's no human author to vet, they won't invest time, and the "anywhere, anytime, by anyone" tagline was called naive. A few commenters noted the unfortunate name collision with Perl's XS extension system, prompting a sidebar about Perl's continued use in production.
Eric Schmidt booed at University of Arizona after praising AI [comments]
54 points · 43 comments · bsky.app · 12h ago
The linked article wasn't available to this summarizer; from the discussion, former Google CEO Eric Schmidt got booed offstage at a University of Arizona commencement after telling graduates “you will work for AI” and that if offered a ride on the rocket ship, “you don't ask questions you just get on.” The thread tore apart that metaphor and the smirk, with most seeing it as a contemptuous power move from someone who stands to profit while young people lose their footing. A long debate kicked off about whether the backlash is really against AI itself or against its capture by a few massive companies, with one camp arguing the real anger is about wealth concentration, data center water bills, and the systematic devaluation of human talent, while another camp flatly hates the AI output itself and the relentless hype that drowns out actual human communication. Some older commenters compared it to the dotcom boom and argued this feels fundamentally different: back then, online shopping slowly eroded stores, but today’s AI threatens a permanent underclass by eliminating entire tiers of white-collar entry jobs at once. The consensus was that Schmidt’s speech crystallized the disconnect—the youth see through the bullshit, and the tech elite just smirk.
Generated 2026-05-18 08:41 UTC
Generated by Sauron from Hacker News discussions and linked articles.