HN Brief: 2026-05-28
Today's HN was split between pessimism about what AI means for work and a quieter defiance of it. A thread on being tired of AI conversations turned into a deep structural critique of bullshit jobs and the impossibility of competing with infinite slop, while a separate discussion about asking for a four-day week if AI truly 10xes productivity got shot down by the grim consensus that under capitalism, gains don't flow to workers. The counterpoint came from DuckDuckGo seeing a 28% traffic spike after Google touted its AI mode, and from Last.fm going independent, which sparked a fierce comparison of old collaborative filtering to modern streaming payola.
The threads most worth clicking into: "I'm Tired of Talking to AI" — the real fight is over whether AI detectors are a scam, with domain experts tearing them apart. "DuckDuckGo search saw 28% more visits after Google said people love AI mode" — the meat is the debate over Google's ad monopoly being mafia-style protection money. "Last.fm is now independent" — the nostalgia is a pretext for a brutal critique of how Spotify's recommendations degraded into payola. "I think Anthropic and OpenAI have found product-market fit" — the split over whether the trillion-dollar buildout is a WorldCom-style collapse in waiting. "Google employee charged with $1M Polymarket insider trading" — the argument over whether prediction markets are designed for this, plus the sheer stupidity of risking a Staff-level salary.
I'm Tired of Talking to AI [comments]
1902 points · 909 comments · orchidfiles.com · 21h ago
The article is a personal rant about the increasing frustration of interacting with AI replacements for human conversation, illustrated with anecdotes about malware reports getting AI copy-paste replies and a Reddit user unknowingly chatting with a bot. The HN thread broadly agreed with the sentiment but immediately dug into the structural roots of the problem, arguing that the real issue isn't AI itself but a society built on "bullshit jobs" that AI is perfectly suited to simulate. A major split emerged over AI detectors: one side cited a study claiming nearly half of online articles are now AI-generated, while others with domain experience tore into those detectors as completely unreliable, pointing out that the tools flag human writing from 2020 as "100% AI" and that the entire detection methodology is an SEO company's flimsy marketing. The discussion pivoted to the idea that the value of actual human thinking has skyrocketed precisely because AI has made fake thinking effortless and cheap, though a pessimistic camp countered that you can't sell that thinking if the market has already been flooded with indistinguishable slop.
Can we have the day off? [comments]
986 points · 587 comments · mlsu.io · 7h ago
The piece is a tongue-in-cheek argument that if AI truly 10xes productivity, workers should get the gains as a four-day week—or even a three-day one. Hacker News immediately shot that down: the overwhelming consensus was that under capitalism, productivity gains go to shareholders and executives, not to workers, so the realistic outcome is layoffs and more output expected, not time off. Several commenters pointed to Ted Chiang’s framing that fear of AI is really fear of capitalism, and the thread spiraled into a debate about whether post-scarcity is a political or cultural problem—people arguing we already produce enough to work less, but the system creates artificial scarcity and consumerism to keep us working. A few shared firsthand experiences of compressed workweeks (three 12-hour shifts, four days one week and three the next) and said the free time is transformative, but the dominant takeaway was grim: don’t expect a day off unless you’re ready to be your own owner or escape wage labor entirely.
DuckDuckGo search saw 28% more visits after Google said people love AI mode [comments]
835 points · 390 comments · www.pcgamer.com · 15h ago
PC Gamer reports that DuckDuckGo's AI-free search page saw nearly a 28% spike in traffic right after Google's CEO claimed users love AI mode. The HN discussion largely zeroed in on Google's real business model, arguing the company isn't protecting search as a product but rather its ad monopoly, and that shoving AI into results is a desperate move to keep users inside Google's sponsored walled garden. A strong split emerged between people who see AI overviews as a useful summarization tool and those who find them restrictive, with several people noting Gemini outright refuses to answer questions that Claude or DuckDuckGo handle fine. One thread turned into a deep debate about whether Google's ad business survives if search quality collapses, with multiple people pointing out that companies now have to pay to appear for their own brand name—essentially mafia-style protection money. Outside the business logic, there were real-world anecdotes of non-technical friends suddenly switching to DuckDuckGo out of sheer frustration, alongside the counterpoint that many people in real life are either neutral or actively enthusiastic about AI, making the backlash appear louder online than it is at the dinner table.
I think Anthropic and OpenAI have found product-market fit [comments]
831 points · 958 comments · simonwillison.net · 15h ago
Simon Willison argues that Anthropic and OpenAI have reached genuine product-market fit, based on evidence like rumored profitability at Anthropic, exploding enterprise API token bills, and hiring surges in enterprise sales. The thread largely buys the premise but splits hard on sustainability: one faction sees these tools as transformative for software engineering velocity, while another warns that the trillion-dollar infrastructure buildout demands a future where developers are 5x–10x more productive, not just 20–40%, and that current token spend as a percentage of salary doesn't yet justify the investment. A big recurring debate is whether Chinese open-weight models like DeepSeek and GLM are real substitutes—plenty of developers report dropping $200/month subscriptions after switching to them for code tasks, but others push back hard on data-security concerns, the difficulty of self-hosting frontier models at scale, and the political reality that most North American and EU enterprises won't route core infrastructure through Chinese providers. The article's note that enterprise pricing is now API-based rather than flat per-seat gets a lot of follow-up: people point out that the $200/month consumer plans are an incredible deal compared to API costs, and predict those subsidies won't last, while others remain skeptical that any of these companies are running sustainably, likening the current situation to the WorldCom collapse.
YouTube to automatically label AI-generated videos [comments]
826 points · 478 comments · blog.youtube · 12h ago
YouTube is rolling out more prominent labels for AI-generated content and will begin automatically detecting and tagging videos that look photorealistically AI-altered even when creators don't disclose it. The thread immediately zeroed in on whether automatic detection is even technically viable, pointing out that whatever system YouTube uses—SynthID, Sight Engine, or something else—will inevitably produce false positives and false negatives, and that mistaken labeling could harm legitimate creators who depend on the platform for income. A large chunk of the discussion shifted from the labeling policy itself to the broader complaint that YouTube's recommendations are already flooded with AI slop, with many arguing that a label doesn't help if the algorithm is still shoving AI-generated content into everyone's feed. Others pushed back hard on that, saying people who see nothing but AI crap are either logged out, not training their recommendations, or watching the wrong stuff, and that their own feeds are full of high-quality human-made content. There was also a cynical undercurrent that YouTube will use this dual role—detecting AI while also selling AI tools like Veo—to train detection systems that keep their own AI output undetectable, leaving independent creators to bear the brunt of enforcement mistakes.
Last.fm is now independent [comments]
734 points · 194 comments · support.last.fm · 16h ago
Last.fm announced it's now an independent company after being owned by CBS/Paramount since 2007, with no changes to user data, scrobbles, or subscriptions. The Hacker News thread turned into a massive nostalgia trip, with longtime users recounting 15–20 years of listening history and sharing how they met friends through the platform, but the real meat was a fierce critique of modern streaming recommendation algorithms—many argued Last.fm's old collaborative filtering was far superior to Spotify's degradated, payola-influenced engine that keeps surfacing the same mainstream tracks. Several people noted that CBS had "completely ignored" the service once they bought it, so there's cautious optimism that independence could let the team focus on actual music discovery rather than corporate meddling, though some pointed out that Last.fm's own similarity algorithm isn't all that great either.
Tech CEOs are apparently suffering from AI psychosis [comments]
639 points · 319 comments · techcrunch.com · 16h ago
The article argues that tech CEOs are deluding themselves about AI’s actual capabilities, laying off thousands based on prototypes that can’t handle the “last mile” of real work. Most of the thread immediately zeroed in on the word “psychosis,” with a large faction calling it cheap clickbait that waters down a real medical term, while others countered that the term accurately describes a collective loss of contact with reality when you’re firing engineers for agentic systems that hallucinate library calls. That debate quickly metastasized into a sprawling argument about whether “being wrong” is psychosis or just stupidity, and whether the term is only being criticized now because it’s aimed at CEOs rather than ordinary people. A significant chunk of the discussion broke away entirely to argue about rent, housing, and survival pressure—some insisting that the real psychosis is a system that makes people fear homelessness, others defending constant pressure as natural and evolutionarily healthy.
Canada to order military plane fleet from Sweden in shift from US suppliers [comments]
525 points · 364 comments · www.theguardian.com · 15h ago
Canada is buying a fleet of Saab GlobalEye early warning planes instead of Boeing’s E-7 Wedgetail, part of a broader push to reduce reliance on U.S. defense firms after Trump-era tariffs and threats. The thread largely accepts this as a sound technical choice—Saab’s system is proven, the airframe is built in Canada by Bombardier, and Boeing’s offering has been a mess of delays and cancellations—but the real fight is over why this pivot happened at all. Most of the discussion argues it’s a direct consequence of the U.S. political climate: Canada’s aircraft industry was already gutted by Trump’s 2017 tariffs, and the current administration’s behavior has killed any presumption of reliable partnership. A smaller but vocal set insists this was coming anyway due to cost and mission fit, but the prevailing view is that the U.S. defense industry now has to compete on merit for the first time in decades. The thread also branches into whether Canada should dump the F-35 for Gripen fighters, with some pointing out that dispersed, road-base-capable jets like Saab’s are better suited to modern drone warfare than the USAF’s giant-base model.
Private equity bought America's essential services [comments]
508 points · 516 comments · rubbishtalk.com · 20h ago
The article argues that private equity has systematically consolidated essential services like fire truck manufacturing, nursing homes, and housing, using leveraged buyouts and debt-loading to extract profits while cutting quality and safety — illustrated by a Chicago fire truck failure that killed four people. The HN discussion immediately zeroed in on antitrust policy, with a strong consensus that pre-1980s enforcement needs to return, but a vocal minority pushed back by calling those old standards vague and unactionable, pointing to failed enforcement and the complexity of modern acquisitions like Y Combinator startups. That triggered a deeper split: one side insisted the guidelines were perfectly actionable and that the real failure is bought-off politicians, while the other side kept arguing that without clear, predictable thresholds, regulators will keep losing. A notable tangent was a few people defending startup acquisition exits as a necessary part of innovation, but they got firmly told that sensible antitrust is worth the cost of breaking that pathway.
All of human cooking compressed into 2 megabytes [comments]
406 points · 162 comments · arxiv.org · 23h ago
A research paper from a London-based startup presents Epicure, a set of ingredient embeddings trained on millions of multilingual recipes, aiming to map culinary flavor and chemistry relationships into a compact model. The HN thread immediately pushed back on the grandiose title—“All of human cooking compressed into 2 megabytes”—arguing it’s really about ingredients and their co-occurrence, not actual cooking techniques, proportions, or preparation methods; several people pointed out that the flavor geometry is interesting but useless without the nuance of technique and ratios that make a dish work. The top comment completely derailed into a separate project for recipe schematics as dependency graphs, which sparked a long and lively tangent about flowchart-style recipe cards, their readability for cooks, and comparisons to tools like Cooking for Engineers and Object Process Methodology. The split was clear: some loved the compressed visualization as a cognitive aid, while others found it overengineered or missing critical details like heat levels and ingredient amounts.
SimCity 3k in 4k (2025) [comments]
354 points · 130 comments · www.thran.uk · 14h ago
A blog post details the six-step process to get SimCity 3000 running in true 4K resolution on modern Windows, including fixes for widescreen support, mouse acceleration, D3D wrappers, and a 4GB RAM patch. The HN thread was a lovefest for SC3K, with many declaring it the best in the series and pushing back against SimCity 4's complexity and 3D advisors. A significant split emerged over the era's 2D isometric art versus later 3D games, with several people pointedly correcting the misconception that SC3K or SC4 were 3D titles. The conversation also veered into a lengthy debate about fire safety in arcologies and high-rise buildings after a user connected the architecture theme to the Grenfell Tower tragedy, while the author's inclusion of a 9/11 joke in a photo caption and his other blog posts about road rage sparked some sharp criticism of his personality.
FBI Arrests CIA Official with $40M in Gold Bars in His Home [comments]
301 points · 192 comments · www.nytimes.com · 8h ago
The FBI arrested a senior CIA official after finding 303 gold bars worth $40 million stashed in his Virginia home, along with luxury watches and cash, though he's only been charged with inflating his credentials and fraudulent military leave pay. The thread largely dismisses the idea this was simple embezzlement—most people assume the gold came from off-the-books covert operations, and that Rush is either being hung out to dry by higher-ups who authorized the spending or caught in some internal CIA power struggle. A strong contingent points out that the agency's internal review of where the gold was stored suggests it wasn't personal compensation but operational funds, and the arrest itself was referred by the CIA director, making it look like a deliberate burn rather than a routine bust. Others dig into the CIA's long history of running slush funds through gold, drugs, and foreign currency, comparing it to post-WWII schemes or modern shadow finance, and a few wonder why the agency couldn't verify his military discharge status before hiring him in the first place.
Incident with Pull Requests, Issues, Git Operations and API Requests [comments]
296 points · 194 comments · www.githubstatus.com · 19h ago
The linked article wasn't available to this summarizer; from the discussion, GitHub suffered another significant outage affecting pull requests, issues, git operations, and API requests. The HN thread immediately turned into a broader reckoning with GitHub’s reliability, with many declaring they’re done waiting for things to stabilize after a particularly bad month of incidents. A fierce debate broke out between GitLab and GitHub defenders, where one experienced DevOps user argued GitLab’s power is its Ops-centric depth while its UI is a labyrinth of third-option discovery, and another countered that GitHub’s developer-centric surface hides a brittle, insecure Actions model that you must minimize your use of. The most substantive argument centered on whether Microsoft’s AI push is to blame: some insisted the real culprit is 14x commit growth from automated agents overwhelming a write-heavy system that can’t be cached, while others rejected that as a convenient excuse, pointing out GitHub knew about the "billion developers" scale years ago and the real failure is that they moved to Azure without properly preparing.
What Apple and Google are doing to push notifications [comments]
263 points · 264 comments · www.jacquescorbytuech.com · 12h ago
The article argues that Apple and Google have turned push notifications into a fully intermediated channel, much like email—they now throttle, summarize, deprioritize, and even rewrite notifications on-device, and senders have no way to tell what happened. HN was sharply split: a loud group of readers cheered the platform controls as long-overdue spam defense, with one calling it “world’s tiniest violin” for marketers upset their promotional notifications get quietly binned, while others pushed back that Apple and Google themselves are not trustworthy (citing Senator Wyden’s concerns about government access to push infrastructure) and that the real problem is apps like Uber and eBay abusing transactional channels for marketing. A former Android engineer at Google chimed in to say push notifications have always sucked for users and the article’s hand-wringing reads as marketer grievance, especially since the author works in revenue operations. A few commenters pointed out that the real missing feature is OS-level enforcement of separate marketing vs. transactional channels, because app developers have no incentive to categorize honestly, and some noted that on-device summarization is genuinely bad for users when it mangles critical alerts.
Valve raises Steam Deck prices [comments]
257 points · 248 comments · www.theverge.com · 14h ago
Valve raised Steam Deck prices by $200–$300, blaming rising memory and storage costs, and the HN thread largely turned that into a wider lament about how AI-driven demand for components is inflating consumer electronics across the board. Many argued this is pushing people toward "anti-tech extremism" because gadgets no longer get cheaper, the job market is rotten, and the average person sees little upside from AI while shouldering the costs—though a vocal minority insisted free chatbots are genuinely useful and worth the tradeoff. The conversation also spilled into skepticism about Valve's upcoming Steam Machine: if the Deck now costs $949 for the 1TB OLED, how can the company possibly hit a sub-$1,000 console price point without subsidizing hardware it can't lock down? A few commenters countered that housing and food inflation is a bigger problem, but others pointed out that tech was one of the few bright spots getting cheaper—and now that's gone too, leaving everything expensive at once.
Mini Micro Fantasy Computer [comments]
250 points · 80 comments · miniscript.org · 22h ago
The linked article wasn't available to this summarizer; from the discussion, Mini Micro is a fantasy computer running a BASIC-like language called MiniScript, designed for easy tinkering. A big chunk of the thread argued about whether indentation-based scoping helps or hurts beginners, with the teacher crowd insisting it forces good habits and avoids bracket errors, while experienced programmers countered that conflating visual layout with semantics creates hidden bugs. The language actually uses `end if` keywords, not indentation, so most of that debate was about Python-style languages in general, not Mini Micro itself. Others pointed out the software is free but not fully open source (the language is MIT, the main system disk has no license), and some argued real hardware like an Arduino is more rewarding, but most pushed back that a virtual machine with sprites is far more accessible than wiring up LEDs.
Go: Support for Generic Methods [comments]
242 points · 202 comments · github.com · 23h ago
The linked article wasn't available to this summarizer; from the discussion, it's about a new Go proposal to add generic methods to the language, building on the generics that were added a few years late. The thread immediately split into camps: one side sees this as Go slowly admitting it was wrong to reject generics for so long, while the other argues that adding features over time is normal growth, not a design failure—pointing out that languages like Rust and Zig evolve constantly too. A loud faction pushed back hard, calling generics, iota, and the `GOPATH` mess "fundamental" mistakes that Go should have gotten right from the start, dismissing its success as a product of Google's brand and the team's pedigree rather than good design. The conversation then veered into a sprawling referendum on which language "got it right" out of the gate—OCaml, Common Lisp, and Elixir were nominated, but each got picked apart for missing static types, missing multithreading, or being an unportable potpourri of implementations.
Training our own AI models [comments]
198 points · 138 comments · posthog.com · 15h ago
PostHog, an open-source analytics platform, announced it will begin training its own AI models on user data, with US cloud users opted in by default and EU users opted out. The HN thread overwhelmingly rejected the company’s framing of “opt-in by default,” calling it an oxymoron that amounts to an unethical opt-out scheme, with several people noting the company is only respecting EU law because it has to. A vocal group of users said they’ll immediately migrate off PostHog, accusing the company of heading straight toward enshittification after they’d adopted it precisely for privacy-respecting analytics. The conversation veered into a heated side debate comparing default opt-in for AI training to default organ donation, where some argued the comparison is nonsensical since one saves lives and the other profits from customers, while others defended defaults as necessary for progress. A few commenters also dug into GDPR complications, questioning whether PostHog’s promised anonymization meets the legal bar and whether EU data is truly protected when processed on US servers.
I found a second vote.gov – and it's registered to the White House [comments]
193 points · 13 comments · thedreydossier.substack.com · 13h ago
The linked article wasn't available to this summarizer; from the discussion, the piece reports that a second, unauthorized staging site for vote.gov was discovered registered to the White House, alongside similar prototype domains for passport services. The thread zeroes in on the legal implications: Congress passed a law after Florida 2000 specifically barring voter registration from operating inside the White House, and the DOJ recently told a federal court that this infrastructure doesn't exist. Several people argue that either the Department of Justice lied to a judge, or the National Design Studio—run by Airbnb cofounder Joe Gebbia under a 2025 executive order—is secretly building a replacement for the country's voter registration system without any of the required privacy disclosures or federal contracts, violating laws dating back to Watergate. A vocal split emerges between those who see this as a straightforward, alarming power grab to centralize all citizen-facing services under the President's office and those who dismiss it as standard prototyping, though the former camp points out that even if it's just a staging environment, the lack of any oversight paperwork is itself a pattern of illegality.
I'm Getting into Mesh Networks (Meshtastic, MeshCore, and Reticulum) [comments]
181 points · 58 comments · www.jonaharagon.com · 12h ago
The linked article wasn't available to this summarizer; from the discussion, it's a comparison of three mesh networking projects—Meshtastic, MeshCore, and Reticulum—with the author favoring Reticulum as the more serious option. The HN crowd largely agreed that these mesh technologies are currently fun toys for nerds, great for local meetups, orienteering games, or boat-to-boat chat at an atoll, but ran into a split on whether they can ever scale beyond that. Optimists pointed to real-world MeshCore networks spanning hundreds of miles (Toronto to Buffalo, across southern California), while skeptics hammered on fundamental limitations: the 900 MHz radios have the range and data rate of a late-90s cordless phone in hilly terrain, Meshtastic's flood routing is a known dead-end, and the whole space feels like a return to packet radio or BBS culture rather than a serious internet alternative. A recurring thread argued that the very weakness of these networks—slow, text-only, unmoderated—is actually the feature, keeping it free of spam and corporate influence, though others countered that far-right spammers have already found the public channels on MeshCore and Meshtastic, proving that no unmoderated space stays pure.
Hallucinate – Massively Multiplayer Online Rave [comments]
164 points · 65 comments · hallucinate.site · 4h ago
The linked article wasn’t available to this summarizer; from the discussion, it’s a browser-based massively multiplayer rave where you drop in as a low-poly avatar and dance to music alongside NPCs and other real people. The thread mostly loved the vibe, calling it a genuine return to anonymous, camera-free nightlife and a better metaverse than anything Meta built, but the server kept crashing under the HN hug of death and the dev was scrambling to fix it live, refusing to push updates because it would restart everything and drop connections. The big, ugly split came from racism in the chat—people were spamming slurs, the dev was manually blocking IPs, and the thread immediately argued about whether to add a filter, a moderation panel, or just let people shoot each other, which the dev shot down as “only love.” Someone suggested a real player count, another wanted jumping to see over the crowd, and a few people dismissed the whole thing as a starbucks-dweller event or pointed out that the real reason people don’t go out anymore is Ticketmaster, not phone cameras.
Google employee charged with $1M Polymarket insider trading bet on search term [comments]
155 points · 79 comments · www.cnbc.com · 7h ago
A Google staff security engineer allegedly used internal data to place $1.2 million in Polymarket bets on which singer and TV show would top Google's Year in Search results, and now faces federal fraud charges. The thread immediately split over whether this is a straightforward crime or just prediction markets working as designed—several people argued the whole point of these markets is to surface non-public information through prices, and that insiders are inevitable. Others shot back that this logic would also justify stock market insider trading, and that the supposed "public benefit" here is knowing which celebrity was searched most, which is worthless. A loud contingent focused on the sheer stupidity of risking a Staff-level Google salary—easily half a million a year—for a $1 million score, calling it a catastrophic negative-EV gamble even before factoring in prison time and a ruined career. A few noted that Forbes had flagged the account back in December, and that federal prosecutors moving within months is actually unusually fast.
Rust (and Slint) on a Jailbroken Kindle [comments]
149 points · 19 comments · sverre.me · 12h ago
The article walks through jailbreaking a 7th-gen Kindle Paperwhite to run Rust and Slint GUI apps, using Zig's cross-compilation toolchain and a custom backend that writes grayscale frames directly to the e-ink framebuffer. The thread largely celebrates the project as a fun, practical hack, with multiple people sharing their own experiences cross-compiling Rust for old Kindles and other constrained ARM Linux devices. Several commenters pivot to recommending Kobo devices instead, arguing they're not locked down at all and run standard Linux, making them easier targets for custom software. A split emerges around GUI frameworks: some praise Slint's embedded/no-std capabilities and company backing, while others push iced as the best Rust GUI library, noting that Druid is abandoned and its successor Xilem hasn't gained traction. The biggest practical concern raised is the reliability of Kindle jailbreaks, with detailed advice on blocking OTA updates by renaming binary files to prevent accidental auto-updates bricking the jailbreak.
Corporations can vote in some Delaware elections, judge says [comments]
143 points · 205 comments · news.bloomberglaw.com · 16h ago
A Delaware Superior Court judge ruled that corporations, LLCs, and trusts that own property in Fenwick Island have the right to vote in local elections, rejecting an ACLU challenge that argued entity voting dilutes the power of actual people. The HN thread immediately zeroed in on the practical implications: with a population of about 400 and Delaware LLC setup costs as low as $109, commenters quickly calculated that anyone with a few thousand dollars could flood the town with corporate owners and swing any election. A faction pushed back, noting the town’s charter caps voting at one per person even if they own multiple properties through multiple entities, but others pointed out that a determined actor could spread tiny slivers of land across dozens of shell LLCs, each getting its own vote. The discussion split between those who see this as a straightforward threat to democratic control and those who argued that property-owner voting already exists in many contexts (like HOAs and special tax districts), with some calling out the City of London and Hong Kong as precedents where corporate votes have long been a feature. The deeper tension was over whether the ruling is a logical extension of corporate personhood or an absurd loophole that should force a national reckoning with *Citizens United*.
Lombardy increases charges for the construction of data centres in green areas [comments]
139 points · 193 comments · en.ilsole24ore.com · 18h ago
Lombardy passed a law jacking up construction fees for data centres in green and agricultural areas by 100-200%, pushing developers toward disused industrial sites instead. The thread immediately split: some called it empty populism, arguing Italy's high energy costs mean nobody was going to build data centres there anyway, while others pointed out that Lombardy already hosts 33 active data centres in the Milan metro area plus 33 more under construction or evaluation, making it Italy's data centre hub. A strong contingent argued data centres are uniquely bad neighbours — they consume huge energy and water yet create almost no local jobs or tax revenue, unlike factories or warehouses, so taxing them captures value they'd otherwise drain. Others countered that blocking data centre construction is anti-innovation and that Europe's green politics will just push investment to Scandinavia or the southern sunbelt where solar is getting cheap. A quieter undercurrent mocked the idea that AI data centres are tiny, and called out the difference between crypto's flexible energy footprint and LLMs' apparently insatiable appetite for hardware.
Stress disrupts hippocampal integration of overlapping events, memory inference [comments]
117 points · 20 comments · www.science.org · 15h ago
This study provides neuroscientific evidence that stress specifically impairs the hippocampus's ability to link overlapping memories, which is the mechanism needed for making inferences and applying past learning to new situations. People in education and memory research pointed out that the basic finding—stress hampers learning—has been known for decades, but the real novelty here is the precise neural mechanism identified for how stress disrupts the relational binding that underpins flexible, inferential thinking. The thread quickly pivoted to a sharp debate about universities: some argued that elite schools deliberately create stressful environments because they are in the credentialing business, selecting for resilience rather than genuine learning, while others pushed back that grade inflation shows institutions are actually trying to reduce pressure, not increase it. Several commenters extended the logic to academic publishing, suggesting the "publish or perish" culture creates exactly the kind of chronic stress that this study shows is counterproductive to deep thinking and scientific progress.
Gemini, Gophers, and Fingers. Oh My Alternative Internets Beyond HTTPS [comments]
116 points · 52 comments · brennan.day · 14h ago
The linked article explores the Finger, Gopher, and Gemini protocols as alternatives to the modern HTTPS web, framing them as lighter, privacy-respecting technologies aligned with solarpunk philosophy. The Hacker News thread largely debated whether Gemini’s deliberate limitations—no inline images, no cookies, no JavaScript—are a feature or a flaw. A major split emerged: some argued the restrictions are the only known way to prevent adtech and tracking from creeping in, while others countered that simple HTML over HTTPS already accomplishes the same goal without abandoning the mainstream web. The finger protocol sparked nostalgia, with people recalling John Carmack’s detailed “.plan” files, but the conversation quickly derailed into juvenile humor about the name “finger” and whether a TLS-enabled version should be called “glove” or “fingercot.” A separate critique noted the article’s claim that Chrome is the only viable browser engine, pointing out that Safari and WebKit were unfairly excluded from the argument.
Show HN: Open-source Workspace (mail,docs,spreadsheet,drive) web/iOS [comments]
88 points · 33 comments · tinycld.org · 17h ago
Nathan posted his open-source TinyCld suite—mail, calendar, contacts, drive, text, and spreadsheets—built to replace the Google Apps account he lost after twenty years. The thread quickly zeroed in on how email delivery actually works here: several people assumed it meant running their own mail server, but Nathan clarified that TinyCld relies on transactional services like Postmark (and soon Mailgun and SES), sparking a wider debate about leaning on third-party providers versus pure SMTP. A lot of the pushback centered on the spreadsheet and document compatibility with Microsoft Office, with one person linking the classic xkcd about edge cases, while others dug into the actual libraries (HyperFormula, excelize, y-crdt) that make it tick. There was genuine praise for the multi-org feature and the modern, lightweight feel compared to Nextcloud, but also healthy skepticism: one reader stressed that a greenfield project is always simpler than a mature one, and a quick stress test of the spreadsheet undo functionality revealed it’s not quite there yet. Nathan was upfront about leaning heavily on AI to write the ~200k lines, and while that made some people nervous about supply chain and dependency bloat, the overall vibe was “this is worth a weekend spin, especially if you’ve been burned by older self-hosted suites.”
I analysed 20 years of my chats [comments]
87 points · 24 comments · drobinin.com · 8h ago
The article is a deep dive by someone who fed a decade of chat logs through LLMs to build a personal CRM, surfacing everything from vocabulary plateauing in his early twenties to the exact moments his friendships cooled off. The thread was split between admiration for the technical feat and a creeping unease about consent: multiple people pointed out that archiving chats where you’ve agreed to disappearing messages is a trust violation, and one person who used to save MSN logs said the cringe was so overwhelming he deleted everything anyway. Others dug into the methodology—someone suggested TF-IDF instead of his manual noise-filtering, and there was a whole aside about losing old WhatsApp backups and whether a CVE could crack a forgotten BitLocker password on an encrypted drive. A few people pushed back on the numbers, noting that 15 close friends feels like a lot and wondering whether “active contact” means just online chat or actually grabbing a beer, while the privacy-minded side worried that even private messages from ten years ago aren’t safe from being harvested by AI.
On Labubu and the Hyperreal [comments]
79 points · 87 comments · 2earth.github.io · 12h ago
The article uses Baudrillard's concept of hyperreality to analyze Labubu, a toy that exploded in popularity partly through K-pop and social media, arguing these plushies function as portable avatars for a lonely, socially isolated generation seeking connection. HN immediately pushed back hard: several people told the author he was seven months late and totally missed that Labubus were simply a fashion accessory trend tied to “performative male” aesthetics aimed at attracting women, not some deep philosophical statement. Others dismissed the entire Baudrillard framing as bogus, noting that fashion trends, status symbols, and symbolic consumption have existed for millennia and aren't some new digital phenomenon. A significant faction took a more pragmatic angle, arguing the discussion itself was overblown moralizing about what's essentially just plastic garbage that people enjoy collecting, with one person pointing out the real story is Pop Mart exporting Asian marketing strategies (blind boxes, gambling mechanics) to the West. The author showed up in the comments to admit he's out of touch, doesn't use social media, and just wanted an excuse to read Baudrillard, which the thread treated as confirmation that the whole analysis was disconnected from the actual cultural moment.
Generated 2026-05-28 08:21 UTC
Generated by Sauron from Hacker News discussions and linked articles.