HN Brief: 2026-05-27
Today’s HN was a referendum on whether big tech has already jumped the shark. AI spending took a beating from multiple angles: Uber’s president admitted their burn rate makes no sense, Xiaomi slashed API prices by 99% to squeeze US labs, and a dev argued outsourcing plus local models will soon beat paying for frontier APIs. The labor counter-punch was just as loud—Wikipedia editors are threatening the first editorial strike ever after the foundation fired union organizers, Massachusetts ride-share drivers formed the first US gig-economy union, and a thread debating the real cost of homeownership turned into a surprisingly sharp argument about whether renting is actually the sane move.
Threads worth clicking into: “Big tech's anti-labor playbook has come for Wikipedia” for the raw tension between a $296M non-profit and the volunteers who actually make it work; “Spain blocks prediction markets Polymarket, Kalshi over lack of gambling licence” because the line between speculation, insurance, and gambling gets pulled apart in genuinely useful ways; “That Methyl Methacrylate Tank” for a Derek Lowe chemistry post that somehow turns into a debate about whether consumers want cheap Plexiglas more than they want refineries that don’t explode; “Uber president says AI spending is getting 'harder to justify'” because the list of actual shipped features is a masterclass in why engineers hate product management; and “The real cost of owning a home” for the brutal takedown of condos as the worst of both worlds and the forced-savings argument getting eviscerated by basic math.
Spain blocks prediction markets Polymarket, Kalshi over lack of gambling licence [comments]
894 points · 412 comments · www.reuters.com · 18h ago
Spain's consumer ministry has temporarily blocked Polymarket and Kalshi, ruling that prediction markets count as gambling under local law because they lack a license. The Hacker News thread largely agreed with the regulatory call—many called prediction markets exactly what they are, a casino—but the real fight broke out over where to draw the line between speculation and gambling. A vocal group argued these markets provide genuinely useful signals (election odds, geopolitical risk) and that insider trading actually improves prediction quality, while others countered that >90% of wagers are useless noise like celebrity word-mouthing. The insurance analogy got pulled apart: you can’t take out a policy on a stranger’s house because you lack an insurable interest, but Polymarket lets you bet on someone else’s death or disaster, creating perverse incentives that could lead to real-world harm. A few pushed back that this is just a classic moral-hazard problem, already solved by existing fraud and manipulation laws, and that bannings like this are an overreaction to a tool that’s only accidentally gambling.
Netherlands blocks US takeover of vital digital supplier [comments]
553 points · 219 comments · www.politico.eu · 20h ago
The Dutch government blocked a US company, Kyndryl, from acquiring Solvinity, the IT supplier that runs the backend for DigiD, the Netherlands' national digital identity system used for everything from doctor's appointments to tax filings. HN immediately zeroed in on the US CLOUD Act as the core reason—any US-owned company can be compelled to hand over data to American authorities, making this a clear sovereignty play. The thread dug into Dutch domestic politics, noting parliament had already voted to end Solvinity's contract but the government dragged its feet, leaving a blocked takeover as the last line of defense. Commenters also debated whether Solvinity could even access DigiD data directly (some argued Logius, a government agency, owned the stack and merely contracted Solvinity for ops), which led into a broader argument about vendor lock-in and the painful reality that migrating away will take five-plus years.
Big tech's anti-labor playbook has come for Wikipedia [comments]
481 points · 288 comments · medium.com · 11h ago
The article details how the Wikimedia Foundation recently fired the longtime lead developer of MediaWiki and disbanded the Community Tech team — the group responsible for building features requested by volunteer editors — with most of those terminated being union organizers, prompting editors to threaten an editorial strike for the first time. The discussion immediately split over how to label the conflict: one faction insisted this is standard private equity behavior, pointing to the new CEO’s Wall Street and State Department background, while others pushed back hard, noting the Foundation is a non-profit with no shareholders, making the PE framing incoherent. A significant chunk of the thread revolved around whether the union’s demands — transparency, planning input, mental health support — are genuinely modest or just vague enough to be impossible to satisfy, with one side arguing the Foundation’s $296 million in reserves means it can easily afford six engineers, and the other countering that 17 months of operating expenses is actually prudent for a long-term preservation project. A few commenters steered the conversation into darker territory, suggesting the real motive behind the shakeup is not labor cost-cutting but a Beltway-driven effort to exert editorial control over the world’s most consulted reference work, though others shot that down by pointing out the fired engineers had no influence over article content. The underlying tension throughout the thread is between those who see this as a straightforward labor solidarity story where a rich mission-driven organization is betraying its values, and those who worry that a union capturing a charity’s resources could eventually undermine the mission itself.
The real cost of owning a home [comments]
358 points · 733 comments · ericturner.dev · 15h ago
The article is a detailed personal breakdown of the actual costs of homeownership, arguing that renting isn't necessarily throwing money away once you account for loan fees, maintenance, taxes, insurance, and the expensive process of selling. The thread quickly pivoted from the author’s specific numbers to a debate about opportunity cost—several people pushed back hard that the author didn't factor in the value of a high-income tech worker's time spent on DIY maintenance, with one side arguing it's often cheaper and better to just hire out. Another major thread questioned whether condos solve the problem, drawing fierce corrections: condos introduce HOA politics, special assessments that can run into six figures, and the fact that you don't own the appreciating land, making them potentially the worst of both worlds. A split emerged on the forced-savings argument of a mortgage, where some argued it beats undisciplined spending while others countered that conservative investing can match real estate returns without the illiquidity and risk of a forced divorce sale. The psychological benefits of customization came up, but were countered by a sharp observation that many owners are ironically just as afraid to modify their home as renters, paralyzed by worry about resale value.
Dropbox CEO Drew Houston to step down [comments]
340 points · 360 comments · www.cnbc.com · 18h ago
Drew Houston is stepping down as Dropbox CEO after 19 years, handing the reins to product chief Ashraf Alkarmi in a co-CEO transition before Alkarmi takes over solo. The HN thread mostly ignored the article's business analysis and turned into a long, nostalgic wake for Dropbox's early days — specifically circling back to that legendary 2007 comment where someone told Houston his idea was pointless because people could just use FTP and rsync. A faction of the thread spent serious time arguing that comment has been unfairly meme'd into "dumbest prediction ever" when it was actually thoughtful feedback on Houston's YC application, not the product itself, and that sync was genuinely seen as a non-problem for normal people back then. Others reminisced about the original public folder feature that let you host static sites directly from Dropbox, lamenting how the company killed that power-user functionality and how the free plan was deliberately crippled just as Apple and Microsoft started pushing their own competing cloud services.
That Methyl Methacrylate Tank [comments]
321 points · 124 comments · www.science.org · 12h ago
Derek Lowe’s blog post walks through the chemistry of the methyl methacrylate tank that nearly blew up in Garden Grove, explaining why the monomer wants to spontaneously polymerize and how the heat from that reaction feeds on itself. HN mostly ran with two things: first, playful speculation about whether the tank would end up as a giant block of Plexiglas—someone pointed out that given the overheating and decomposition, it would more likely be a cracked, bubbly mess, like a failed NileRed experiment. Second, a much sharper split erupted over whether the lack of passive safety systems was a regulatory failure or just the cost of doing business. One side argued the chemical industry should have had neutralizer injection or deluge systems as a default, calling it corporate negligence; the other pushed back hard, saying consumers want cheap Plexiglas and battery materials without paying for the safety infrastructure, and that the real problem is a “you” problem of demanding nice things without accepting industrial risk.
Outsourcing plus local AI will soon become more economical vs. frontier labs [comments]
280 points · 299 comments · www.signalbloom.ai · 19h ago
The article argues that pairing a cheap overseas engineer with a local or open-source AI model like DeepSeek will soon be more cost-effective than paying for expensive frontier API calls from OpenAI or Anthropic, projecting a crossover within a year. The thread largely agrees that frontier labs have painted themselves into a corner with massive price hikes—GPT-5.5 costs triple what GPT-5 did eight months ago—while DeepSeek offers a 30x cheaper alternative that's already good enough for most coding work. A major split emerged over whether pure energy costs will determine the winner, with one camp arguing the US can't compete with China's cheap electricity (and its willingness to burn coal) for inference, while another countered that the US has far more data centers and higher energy per capita, and that nuclear regulation is a solvable political problem. Several people pushed back hard on the idea that AI pricing works like old software, because inference has a real, non-zero marginal cost—unlike a SaaS product—and that the current prices from both frontier and open labs are distorted by investment cash and loss-leader strategies rather than actual economics. The Bitcoin analogy got thrashed out: some saw both as energy-heavy first-gen tech that will get cheaper, but others pointed out that Bitcoin is *designed* to waste energy no matter what, while AI actually gets more capable with more compute, making it a fundamentally different cost-benefit equation.
Uber president says AI spending is getting 'harder to justify' [comments]
280 points · 137 comments · www.theverge.com · 22h ago
Uber’s president admitted the company blew through its entire annual AI budget in just four months and can’t draw a clear line between massive token consumption and actual useful features shipping to customers. The thread immediately latched onto that disconnect, with several people pointing out that as an end user, the Uber experience hasn’t changed at all, so what exactly did all those billions of tokens buy? Someone listed the new features Uber built with AI—in-app hotel bookings, travel suggestions, a service where your driver picks up takeout for you to eat in the car—and the consensus was brutal: those are worthless, actively terrible ideas that make the app worse, with the food-in-the-car concept drawing particular disgust for the smell and hassle it creates for drivers. A long, cynical comment compared it to the inevitable decay of every successful tech company, where leadership can’t stop bolting useless crap onto a working product instead of finding the next real thing. Others dug into the management absurdity, noting that the same execs who cheap out on essential tools like monitoring or configuration management are suddenly giving everyone multiple AI subscriptions under pressure from investors who never pushed blockchain or NFTs this hard.
Uber, Lyft drivers in Massachusetts form first US ride-share union [comments]
275 points · 193 comments · www.reuters.com · 16h ago
Massachusetts ride-share drivers just secured what the state and union leaders say is the first officially recognized gig-economy union in the country, covering nearly 70,000 drivers after a 2024 ballot measure created a state-level framework that sidesteps federal labor law. The thread quickly turned into a much bigger fight about what happens when driving as a whole collapses as a profession—commenters zeroed in on trucking automation as the real ticking clock, arguing that neither labor nor government can pull off a soft landing for millions of workers. One camp went straight to threat-modeling: if autonomous trucks hit the roads en masse, the math of physical sabotage (burning them, puncturing tires, spoofing sensors) looks a lot easier than the math of enforcement, given only a million soldiers versus a hundred million workers. Others pushed back hard, saying destroying technology has never worked for workers in history, and the cargo-container revolution actually shows a path where longshoremen got buy-in through a just transition, but policymakers have to mean it. A separate, scrappier debate flared up about whether Uber drivers are really exploited or just doing a voluntary gig they like for flexible cash, with one side arguing drivers aren't stupid and the math on car depreciation is obvious, and the other side pointing out that if the drivers themselves just voted to unionize, something clearly needs changing.
Stripe is friendly to “friendly fraud” [comments]
262 points · 164 comments · www.gingerlime.com · 7h ago
The post is a merchant's detailed account of a customer pulling off "friendly fraud" twice in a row — buying cigar glue, getting it delivered, gloating about the scam in writing, and then winning both chargebacks — with the merchant arguing Stripe refuses to use the clear evidence of abuse (including screenshots of the customer admitting it) to flag the fraudster for any other merchant on its network. HN largely agreed this is a real, painful problem for small sellers, but the discussion split hard on who to blame: a lot of people pointed out that Stripe is stuck between the card networks and the customer's bank, and that any system designed to block repeat friendly-fraudsters would inevitably nail innocent customers who had their cards stolen, creating a massive PR and legal liability nightmare. Others pushed back that Stripe literally sells *Radar* as a network-level fraud detection tool and takes a cut of every transaction, so their refusal to use merchant-provided evidence of confirmed fraud feels less like a technical limitation and more like a deliberate choice to outsource the cost of fraud to small businesses. A few commenters dug into the logistics, noting Stripe does support Visa's Compelling Evidence 3.0 but that it only works for recurring disputes over 4–12 months, not this kind of single-shot fraud, and the thread also wandered briefly into whether Monero or crypto could sidestep the whole chargeback system (with the counter that grandma won't buy cigar glue that way).
The worst job interview I ever had [comments]
237 points · 202 comments · www.oliverio.dev · 11h ago
The article is a first-person account from an engineer who went through an interview at a mental health startup that turned into what felt like an unsolicited, trauma-baiting therapy session, asking about the hardest day of their life and biggest personal struggles, only to be rejected with a one-line email. HN broadly agreed that the author dodged a bullet, with many calling the interview abusive and a major red flag, while a few pushed back that this sounds like a standard behavioral interview aimed at assessing how someone handles adversity, and that the author may have misread the room. The discussion quickly spun into a sharp debate about the legality and ethics of "culture fit" interviews, with multiple people pointing out that asking about hobbies or personal life opens the door to discrimination lawsuits based on protected class. A deep etymological tangent broke out over the author's use of the word "cracked" to mean "highly skilled," tracing it from early 20th-century idioms through Apex Legends gaming culture, with several people genuinely confused whether it meant "great" or "damaged." The consensus was clear: trauma-dumping as a screening tactic is invasive and ineffective, and the mental health startup should have known better.
Erin Brockovich made a map to track data centers around the country [comments]
215 points · 198 comments · www.brockovichdatacenter.com · 7h ago
Erin Brockovich released a crowdsourced map tracking AI data centers across the U.S., inviting people to report local impacts. The thread immediately split: some questioned the map's accuracy, noting that many "community reported" sites look like duplicate or mistaken entries—people seem to be flagging any big new building or small telco closet as a data center, while genuine hyperscale locations are missing or misplaced. Others zeroed in on the irony that the site itself appears AI-generated, with several people saying they're highly confident Claude built it, which undercuts the anti-AI framing. A recurring argument emerged between those who say the environmental and local costs are real and deserve grassroots organizing, and a counter-crowd dismissing the whole effort as performative populist brainrot driven by people who don't understand how the internet works. The conversation also veered into whether Brockovich's map adds anything over existing commercial data center databases, with the defense being that her name brings visibility and a political angle, even if the data is sloppier.
Cloudflare Flagship [comments]
201 points · 96 comments · developers.cloudflare.com · 8h ago
Cloudflare launched Flagship, a feature flag service that lets you toggle features without redeploying, with native Workers bindings and OpenFeature compatibility. The HN thread quickly turned into a broader referendum on Cloudflare’s current trajectory—people are enthusiastic about the product breadth but frustrated that fine-grained permissions and spending limits still don’t exist, with multiple voices saying “I’d use them for real work if they’d just fix access controls.” A solid chunk of the discussion dismissed feature flags as “booleans as a service” you can implement in a database yourself, while others pushed back saying real-world needs like per-customer rules, audit logging, and streaming updates get gnarly fast. There’s also genuine appreciation for OpenFeature as a standard, and a warning that the client SDK’s API token isn’t scoped per app, so you’d be exposing all your flags to the browser—something the team says is “WIP.”
A sleep-like consolidation mechanism for LLMs [comments]
198 points · 132 comments · arxiv.org · 16h ago
An arxiv paper proposes giving LLMs a “sleep” phase where they run offline training passes over recent context to update internal “fast weights,” then clear the KV cache, aiming to handle long contexts without scaling attention. The HN thread immediately split over whether this is just a dressed-up version of context pruning and compaction tricks, with some arguing the method genuinely updates model parameters based on context rather than just discarding tokens, while others insisted it only touches the SSM state vectors, not actual weight matrices. The biggest debate, however, wasn’t about the mechanism—it was about the name “sleep”: a loud faction complained the analogy is pure anthropomorphization that clouds objective analysis, while others defended it as a legitimate bionic parallel to hippocampal replay and memory consolidation, noting that biological sleep emerged independently across lineages, suggesting it solves a fundamental information-processing constraint that artificial systems might also hit. A few commenters pointed out that the paper’s own framing of sleep as a period with no external input and a clear cognitive benefit after the fact is a fair mechanistic analogy, but the counter-argument that no LLM actually *dies* without this process—unlike animals—kept the tension alive.
Is "colorectal cancer" rising in "young people"? [comments]
196 points · 209 comments · dynomight.net · 16h ago
The article argues that colorectal cancer is indeed rising in younger generations, driven by a cohort effect where people born after 1960 face higher risk at every age, though it finds no single compelling cause and notes that several other cancers are rising similarly. The HN discussion quickly zeroed in on the article’s admission that this isn’t unique to colorectal cancer, with some pushing back that age-adjusted cancer rates overall are dropping and the article cherry-picked a handful of cancers that are rising. A significant split emerged over screening: several commenters shared horror stories of botched colonoscopies and perforated intestines, while others with direct experience—including one who is currently undergoing treatment for colon cancer—argued bluntly that the complications from screening are nothing compared to the disease itself. The conversation also dove into practical details about bowel prep methods, sedation options, and the relative merits of FIT stool tests versus colonoscopy, with a general consensus that prep is the worst part and that anecdotal fear of complications should be weighed against statistical safety. No one really disputed the headline claim, but the thread redirected the focus from “is this real” to “what should I do about it, and is the screening worth the risk.”
AWS Fired the One Employee Who Gave a Damn [comments]
194 points · 107 comments · www.seuros.com · 18h ago
The article recounts how AWS deleted the author’s decade-old account, and how Tarus Balog—a 20-year open-source veteran on AWS’s Open Source Strategy team—escalated it to a CEO-level Severity 2 ticket, restored the account, and was then laid off within ten months as part of AWS’s broader pivot toward GenAI at the expense of core services like S3 and EC2. The HN thread largely split into two camps: one group was furious at the substance, arguing that Balog’s firing perfectly illustrates how Amazon systematically removes anyone who breaks procedure to actually help customers, even when that help gets CEO attention; the other group couldn’t get past the article’s obviously AI-generated prose, sloppy design, and dramatic sentence fragments (“Not X, not Y, not Z—A!”), calling it Claudeslop that devalues a genuinely important story. Several commenters who had their own AWS nightmare accounts confirmed that Balog personally reached out to them too, making the loss feel concrete, while others pushed back hard on the article’s framing—pointing out that former AWS engineers retiring to open bakeries isn’t a “trauma response” but simply what happens when you get rich enough to walk away. The biggest split was meta: is this a damning case study of AWS’s culture, or an ironically AI-generated screed that undermines its own argument by flooding the zone with synthetic outrage?
Where does next-token prediction leave us? [comments]
176 points · 131 comments · pop.rdi.sh · 6h ago
The essay argues that AI maximalism is a class issue, where proponents—cushioned by safety nets—gleefully declare industries "solved" while stripping workers of their only bargaining chip: labor. The thread largely agreed with the premise but pushed back on the article's assumption that only economically secure people support AI, pointing to survey data showing developing countries like China are far more positive about AI than the West. A major split emerged over why: some argued China’s growing economy and state-led plans for retraining make AI feel like collective uplift, while the West’s oligarchic messaging—CEOs bragging about cutting labor—makes it feel like a threat. Others went further, arguing AI enthusiasm correlates with corruption and cultures that value quick results over truth, while skepticism aligns with societies that invest in long-term correctness. A side debate questioned whether working at AI companies is morally comparable to working at a gun manufacturer, with many unconvinced by the analogy given the explicit corporate goal of making human labor obsolete.
Stop Advertising in Your Commits [comments]
175 points · 150 comments · akselmo.dev · 13h ago
The post argues that AI tool attribution lines like "co-authored-by: Claude" in git commits are just free advertising for companies that don't pay open-source contributors, and that disclosure belongs in merge requests, not in the permanent commit history. The HN thread split hard on whether this is advertising or useful transparency — some people insisted it's valuable signal that code was AI-generated and should be reviewed accordingly, while others pointed out that "Sent from my iPhone" email footers were similarly justified as useful context but mostly served as marketing. A significant thread dug into the theory that Anthropic is deliberately adding these tags to scrape GitHub for RLHF training data, correlating commits against their Claude Code sessions to see how humans refine the output. The "co-authored-by" framing specifically drew fire from people who noted that no one attributes commits to `gcc` or `alembic` for generating code, and that calling an LLM a "co-author" is an anthropomorphizing growth hack, not a technical necessity.
Stack Overflow’s forum is dead but the company’s still kicking [comments]
161 points · 235 comments · sherwood.news · 14h ago
The article charts Stack Overflow’s pivot from a dying Q&A forum—traffic cratering as LLMs like ChatGPT absorb all the simple questions—to a profitable AI data-licensing and enterprise-search business, with revenue doubling to $115M even as the public face of the site wilts. The thread largely side-stepped the business story to re-litigate the site’s long, bitter decline, arguing that LLMs were just the final blow after years of draconian moderation drove away askers. A deep split emerged: one camp insists SO was always strictly moderated and that was its feature, while the other points to an inflection point around 2014 where moderation curdled from curating quality into performative pedantry—closing valid questions as duplicates of unrelated old posts, or lecturing users about XY problems instead of answering. Several people pushed back on the “it was always that way” narrative with data, linking to a post showing SO’s engagement peaked in 2017 and had been bleeding users for half a decade before ChatGPT. The most damning takeaway, repeated from multiple angles, was that the site’s culture taught people to solve problems alone rather than risk posting, which ironically made them better developers but hollowed out the community.
The just-say-no engineer was a ZIRP phenomenon [comments]
136 points · 108 comments · www.seangoedecke.com · 4h ago
The piece argues that the “just-say-no” engineer—the senior gatekeeper who blocks complexity and slow-walks changes—was a creature of the zero-interest-rate era, when tech companies hired massively without needing productivity, and that the shift to a profit-hungry, AI-driven post-ZIRP world has made that role toxic. The thread split sharply: some called the article cynical revisionism, pointing out that hiring during ZIRP was often about denying talent to competitors or hitting growth targets, not about creating busywork for gatekeepers to manage. Others pushed back hard on the idea that the end of ZIRP makes gatekeepers obsolete—arguing that expensive capital should make companies *more* cautious, not less, and that the real shift is AI hype forcing leaders to accept shippable-but-sloppy code. A few commenters from non-startup domains (finance, infrastructure) said the piece only applies to web/app product work, where stability matters less, and that regulated or mission-critical environments still need the “no” engineer. The top sentiment was less consensus than a sharp debate over whether the article was a useful diagnosis of structural change or just a well-written rationalization for a nothing-burger.
What color is your function? (2015) [comments]
118 points · 150 comments · journal.stuffwithstuff.com · 16h ago
Bob Nystrom’s 2015 essay lays out the “function coloring” problem—how async and sync code in most languages act as two incompatible “colors” that force a painful, viral split through your entire call stack, and argues that languages with user-space threads (Go, Lua, Ruby) avoid it entirely. The thread largely agrees with the diagnosis but gets into a knuckle-down fight over the cure: a vocal camp insists you should just ban async in Python and throw hardware at the problem via threads and processes, calling async a “solution in search of a problem” that adds developer overhead for minimal gain. A counter-camp pushes back hard, pointing out that concurrency helpers like `asyncio.gather` are far simpler than managing thread pools or message queues, and that the entire frontend world runs on async without complaint. A deeper dig into the comments clarifies that the real poison of “color” isn’t just inconvenience—it’s that it’s non-optional and non-encapsulable, forcing a change ten layers up the stack with no way to hide it, unlike error types or context objects which can be locally handled or faked. Someone also notes that Go’s implicit approach just trades one color for a different set of subtle pitfalls, like channel blocking being invisible until runtime.
Don't Subscribe So Casually [comments]
117 points · 84 comments · thebestworstcase.substack.com · 17h ago
The essay argues that subscriptions, unlike one-time purchases, reshape your habits and identity over time, with companies deliberately optimizing for retention and behavioral change rather than product value. The thread largely pivoted away from the article's philosophical warning to a tactical skirmish over the "cancel immediately" strategy—a commenter insisted you should subscribe, cancel right away, and still use the service for the paid month, which sparked a debate about whether this works. Several people pushed back hard, pointing out that Uber One and Adobe kill the benefits instantly on cancellation without prorating, calling it "blatant theft" or "Scrooge McDuck" behavior. Others defended the practice, arguing that if you signal you're not a future customer, why should the company keep giving you free trial benefits, though the counter was that cancellation doesn't signal disinterest so much as a desire to avoid accidental billing. The thread also meandered into practical workarounds like burner credit cards, catch-all email domains, and the observation that most people pay for subscriptions because the convenience of not re-setting everything up is worth more to them than the money.
From Rust to Ruby [comments]
111 points · 82 comments · xlii.space · 10h ago
A developer wrote about converting his personal 15,000-line Rust web app into Ruby on Rails using a local LLM, cutting the codebase by 77% in about 30 minutes — except he hasn't actually run the converted code yet. The HN crowd immediately zeroed in on that: people are baffled he wrote a blog post declaring victory over a codebase he never executed, with some calling it vibe-coded vaporware that "defies belief." The LLM hate is strong here — a lot of people dismissed the whole exercise as "I told my underling to do it" and argued the comparison is worthless without knowing whether the Rails app even compiles, let alone works correctly. That kicked off a broader brawl about whether Ruby on Rails actually delivers on developer happiness, with one camp arguing it's the most joyful framework available and another countering that Rails magic makes debugging a nightmare of invisible methods and painful upgrades. On the language-choice front, several people pointed out a blunt take: if you can switch from Rust to Ruby and it makes sense, then Rust was the wrong choice from the start.
Xiaomi MiMo-v2.5 Series API Permanent Price Reduction Up to 99% [comments]
109 points · 120 comments · platform.xiaomimimo.com · 14h ago
Xiaomi announced a permanent price cut of up to 99% on its MiMo-v2.5 API pricing, driven by technical optimizations like Sliding Window Attention that slashed KV Cache data transfer to about a seventh of previous levels. Hacker News mostly treated this as China's AI sector pulling the same aggressive play it did with EVs—state-backed loss leaders (or at least heavily subsidized inference) designed to force prices to zero and capture market share, with several people noting the timing perfectly aligns US West Coast business hours with Beijing off-peak rates. A lot of the pushback zeroed in on the fine print: the 99% cut applies to cached inputs, not raw generation, though that still matches DeepSeek's pricing floor, and users who logged in found their token allocations had ballooned 5–10x. The thread split cleanly between those who see this as a deliberate strategy to burst the American AI valuation bubble and those arguing US companies like Anthropic and OpenAI can't possibly match these prices without bleeding even more cash—and will instead lobby for a ban on Chinese models to protect their margins.
Sonny Rollins, jazz saxophonist, has died [comments]
106 points · 14 comments · www.rollingstone.com · 7h ago
The article reports the death of jazz saxophonist Sonny Rollins at age 95, a colossus of the instrument whose career spanned from bebop with Charlie Parker to a solo on a Rolling Stones track. The thread immediately zeroes in on his status as the literal last of a generation—one person noted he was the final living legend from the *A Great Day in Harlem* photograph, prompting a quick correction that bassist Ron Carter and Herbie Hancock are still alive and touring. Another commenter confessed they'd assumed Rollins was already dead, which kicked off a list of current musicians who might become legends (Wynton and Branford Marsalis, Joshua Redman, Brad Mehldau) versus those who are phenomenal but less visible. The rest of the conversation is pure live-memory storytelling: a fan who saw him in 1998 got 36 solo choruses on "St. Thomas," and the best anecdote has Rollins hearing "Waiting on a Friend" in a grocery store and thinking it was finally a Stones song he liked, before remembering he played the sax solo on it.
The Ballad of TIGIT [comments]
102 points · 20 comments · www.owlposting.com · 16h ago
The linked article tells the story of TIGIT, a promising cancer immunotherapy target that pharmaceutical giants like Roche, Merck, and Gilead bet billions on, only for clinical trial after clinical trial to fail. The Hacker News thread zeroed in on the concept of "herding" the article describes—where companies pile onto a target out of competitive FOMO, spending over $3 billion and enrolling 49,000 patients chasing the same failed molecule. Several commenters defended the collective overinvestment as rational: when a single success like Keytruda is worth hundreds of billions, a 10% shot at the next one justifies the spending, even if hindsight makes it look foolish. The discussion broadened to compare this to tech venture capital, where missing the next big thing is far more costly to your career than losing money on a wrong bet, and to contrast TIGIT's flameout with the genuine progress finally being made on previously "undruggable" targets like KRAS. One side thread got into the weeds on why the field never bothers to explain what "TIGIT" stands for, since gene names are often accidental historical artifacts rather than meaningful descriptors.
Modern Blu-ray drives can now rip GameCube, Wii, and Xbox 360 games to PC [comments]
100 points · 29 comments · github.com · 13h ago
Modern Blu-ray drives can now rip GameCube, Wii, Xbox 360, and Dreamcast games to PC using a new third-party firmware called OmniDrive, which bypasses the need to modify the original consoles. The Hacker News thread quickly pointed out that the article buries the actual GitHub repository — several people just posted the direct link with blunt frustration, calling out the unnecessary detour through a video and a Tom's Hardware article. A chunk of the discussion zeroed in on how OmniDrive relates to the existing LibreDrive for MakeMKV: the key difference is that LibreDrive exploits a firmware bug for Blu-ray decryption, while OmniDrive is a patched firmware that enables reading the proprietary disc formats for old console games, with some overlap in supported hardware. There was also a small but sharp side-argument about the GitHub readme using green X marks for "not supported," with people noting that’s a colorblind-hostile choice and that a red X or stop sign would be clearer. Overall, the consensus is that this is a major usability win for dumping physical retro games, but you have to carefully check the chipset compatibility first because flashing the wrong drive will brick it.
Flatpak Will Depend on Systemd [comments]
97 points · 122 comments · www.osnews.com · 22h ago
Flatpak is planning a major 2.0 rewrite that will introduce a hard dependency on systemd, moving permission management into a new service called systemd-appd. The HN thread zeroed in on the toxic fallout from that announcement: after a talk, some non-systemd distribution users got harassed by a bystander, leading Flatpak developers to publicly declare they're done accommodating alternative init systems. This reignited the familiar systemd flamewar, with people arguing that systemd has become an unavoidable part of the Linux desktop stack and others pushing back on its monolithic, Red Hat-driven scope. A parallel debate broke out over whether AppImage is a viable cross-distro alternative, with most people concluding it solves a different, less portable problem. The whole thread carries a weary sense that this is another step in systemd absorbing everything in userspace, and that distributions like Void and Alpine are getting squeezed out of the desktop app ecosystem with no real recourse.
Incident with Actions and Pages [comments]
95 points · 375 comments · www.githubstatus.com · 20h ago
GitHub suffered another multi-hour outage yesterday that took down both Actions and Pages, with the post-mortem noting that some Issues, PRs, Comments, and Discussions got incorrectly marked as hidden during the incident. The thread quickly turned skeptical about GitHub's reliability, with people pointing out that the official status page shows roughly two nines of uptime over 90 days and that the instability problems trace back to the Microsoft acquisition era, not just the recent AI load surge. A lot of people running self-hosted runners were furious that they still couldn't dispatch jobs—the control plane was down, so their own hardware was useless—and there was a real split between those arguing you should never fully depend on a third party for deployments versus the pragmatic reality that GitHub has become so embedded that it's treated as synonymous with git itself. The most energetic debate was about whether LLM-generated traffic is actually the cause, with one side pointing to 1000x increases in repo clones from agentic coding and Microsoft's boast that 30% of their code is now AI-written, while the other side countered that GitHub's competitors haven't seen the same instability despite facing the same industry shift.
Launch HN: Minicor (YC P26) – Windows desktop automations at scale [comments]
87 points · 54 comments · www.minicor.com · 17h ago
Minicor is a YC-backed platform that uses AI agents to build and maintain Python scripts for automating clicks and keystrokes in legacy Windows desktop apps that have no API—think medical record systems or old ERPs. A lot of the thread was people asking what "RPA" means (the founders clarified it stands for Robotic Process Automation), but the real debate was about whether the hybrid approach—deterministic code with AI for recovery and maintenance—is smarter than relying on pure computer-use agents, with the founders arguing that their method cuts the 30% failure rates common in pure agentic systems. Security and privacy came up heavily: the founders confirmed that screenshots sent to the LLM for verification contain no PHI, and all logs stay on the customer’s own infrastructure. There was also pushback on the landing page's "on-premise" phrasing, a dead trust center link, and complaints that the customer logos were nearly invisible, but the overall vibe was that legacy system users are desperate enough for automation that they’ll pay well for this kind of integration layer.
Generated 2026-05-27 08:19 UTC
Generated by Sauron from Hacker News discussions and linked articles.