HN Brief: 2026-05-19
Today’s HN was dominated by AI backlash and its messy fallout, from booed commencement speakers and the Pope’s encyclical on human dignity to the quiet admission that even Linus Torvalds is drowning in AI-generated bug reports. The open-source world offered a few bright spots — a polished Obsidian alternative stored as local Markdown, Haiku OS booting on Apple Silicon — but the mood was skeptical, with multiple threads questioning whether AI coding actually replaces real engineering or just generates slop. Geopolitics and surveillance also surfaced: Iran launching Bitcoin-backed ship insurance to skirt sanctions, and the FBI trying to buy nationwide license-plate-reader access, both sparking hard-nosed debates about strategy and privacy.
The thread on Elon Musk losing his OpenAI lawsuit is worth reading for the jury’s statute-of-limitations ruling that sidesteps the merits entirely. Garry Tan’s confrontation with Radley Balko pulls you into a sharp fight about billionaire power and what “unethical reporting” really means. And the experiment where AIs ran radio stations for six months is a genuinely entertaining look at model failure modes — DJ Gemini collapsed into repeating “stay in the manifest” hundreds of times a day. If you want the deeper coding debate, “Where Are the Vibecoded Photoshops?” argues convincingly that AI hasn’t replaced complex software engineering, despite the hype.
Elon Musk has lost his lawsuit against Sam Altman and OpenAI [comments]
924 points · 454 comments · techcrunch.com · 14h ago
Elon Musk lost his lawsuit against Sam Altman and OpenAI because a jury found that any harm happened too long ago to sue — the case turned on a statute of limitations defense, not the merits of Musk’s “stole a charity” claim. The thread mainly wrestled with whether an appeal has any real chance, with multiple commenters explaining that overturning a jury’s factual finding on timing is nearly impossible under the Seventh Amendment, even though Musk’s lawyers will happily bill hours for the attempt. A recurring angle was that Musk’s own 2017 emails endorsing a for-profit structure undercut his betrayal narrative, and several people ranked Altman as the more dangerous of the two plutocrats precisely because he’s cleaner and slicker. The discussion also spun off into a detailed debate about space data centers — one commenter argued they’re indefensibly dumb regardless of launch cost, while others defended the idea if radiation-hardened chips and Starship economics ever pan out.
Show HN: Files.md – Open-source alternative to Obsidian [comments]
625 points · 305 comments · github.com · 18h ago
The linked article is Files.md, an open-source, local-first note-taking app that stores everything as plain Markdown files, positioning itself as a privacy-respecting alternative to Obsidian after five years of development. The HN thread was split between genuine admiration for the project's craftsmanship—particularly the author's claim of 500+ commits in the last week and the radically simple Go server that replaced a docker/php/nginx stack with a single 15MB binary—and a sprawling philosophical debate about whether open-source can be sustainably funded, with one camp arguing donations and sponsorship work (pointing to ripgrep and the entire OSS infrastructure) and the other insisting developers need rent money and most people won't pay unless forced. A significant chunk of the discussion turned into a comparison bazaar, with people swapping their own journeys through TiddlyWiki, Zettlr, Silverbullet.md, and TrilliumNext, each defending why they landed where they did. The author's philosophy—that you should own both your files and the software that opens them, and that LLMs make radically simple code future-proof—was mostly met with enthusiasm, though a few people raised practical concerns about mobile Safari support and how the sync mechanism actually works.
Garry Tan, the CEO of YC, accused me of unethical reporting [comments]
513 points · 156 comments · radleybalko.substack.com · 16h ago
Radley Balko lays out a detailed rebuttal after YC CEO Garry Tan accused him of unethical reporting, showing how a viral Dion Lim story about Chesa Boudin’s office dropping charges against a juvenile assailant was factually wrong and that Balko’s own Washington Post piece merely corrected the record. The thread quickly turned into a fight over whether journalism can ever be apolitical: a chunk of the room defended Balko’s work as transparent, rigorous fact-checking, while others insisted the whole thing is just “intra-party squabbling” between California Democratic factions and that Balko is no neutral actor. A lot of pushback landed on Tan’s money and power—multiple people argued that extreme wealth warps judgment and that Tan, who funded the $7 million recall that removed Boudin, shouldn’t be lecturing anyone on ethics. The most pointed comments noted that Tan’s attack on Balko feels like a billionaire using his platform to smear a reporter for doing basic due diligence, and that the real story is how money buys both elections and media narratives.
We stopped AI bot spam in our GitHub repo using Git's –author flag [comments]
462 points · 219 comments · archestra.ai · 16h ago
Archestra’s CTO posted about fighting AI bot spam in their GitHub repo by enabling "require prior contribution" and using Git’s `--author` flag to automatically whitelist real humans via a captcha — they blocked 500 bots in a week. The thread mostly agreed the trick was clever and lightweight, but quickly pivoted to arguing that GitHub and Microsoft have zero incentive to stop AI-generated PRs because they sell Copilot, and that any scoring or token system (like ELO or vouching) would be gamed the moment one AI got whitelisted and started vouching for others. A few people pointed out the irony of a company with a `.ai` domain and an "agentic stack" complaining about AI slop, and the comment section also spun off into a side debate about whether giving job candidates real-world coding tasks is still fair now that they can cheat with AI.
Anthropic acquires Stainless [comments]
434 points · 291 comments · www.anthropic.com · 15h ago
Anthropic has acquired Stainless, the company behind the SDK generation tooling that's been producing every official Anthropic SDK since the API launched, along with MCP server tooling used by hundreds of companies. The HN thread quickly zeroed in on the fact that Anthropic is immediately winding down the hosted Stainless product, which made a lot of people call this an acquihire—and note the irony of "hundreds of companies rely on us" sitting alongside "we're shutting that down starting today." A big chunk of the discussion spun into the familiar startup enshittification complaint: that VC-backed tooling gets you hooked, then gets acquired and killed or jacked up, with former Stainless employees pushing back that the team actually gave customers a clean transition path and the product was never about blitzscaling. Others pointed out that OpenAI also used Stainless for their SDKs, so Anthropic just bought a piece of their competitor's infrastructure, and the thread ping-ponged between cynicism about walled gardens in AI tooling and defenses that this is just normal market consolidation—with a brief side tangent making fun of the announcement for not containing the word "journey."
Eric Schmidt speech about AI booed during graduation [comments]
353 points · 375 comments · www.nbcnews.com · 21h ago
The linked article wasn't available to this summarizer; from the discussion, Eric Schmidt gave a graduation speech drawing a parallel between AI and the transformative impact of computers, and the crowd booed him. The thread zeroes in on why—most people arguing it wasn't about Schmidt personally but a generic platitude like "the future is in your hands," but about the raw hypocrisy of a tech billionaire telling new grads they can shape a transformation his own industry is driving, one that threatens to make their degrees worthless. A loud contingent on HN sides with the booing graduates, pointing out that "implementers" of AI (corporations, not just creators) are actively marketing it as a white-collar replacement and that disliking it isn't "fear of change" but a rational response to an existential threat to entry-level careers. The counterargument is mostly dismissive: the quote is standard commencement fare, and refusing to engage with the tech only hurts the graduates more. A notable side thread spirals into generational blame, noting that Boomers have held power for decades and that Gen-X and Millennials never got a turn to shape the world—so the graduates' anger is really at a system that structural lock-in, not just one speech.
The last six months in LLMs in five minutes [comments]
324 points · 200 comments · simonwillison.net · 6h ago
The linked article wasn't available to this summarizer; from the discussion, Simon Willison's talk covers the rapid pace of LLM releases over six months, specifically pointing out that the "pelican riding a bicycle" SVG benchmark is now trivially solved—so much so that he's replaced it with an opossum on an e-scooter. A big chunk of the thread gets into whether the recent wave of model improvements is a genuine inflection point or just marketing hype: some people swear Opus 4.5 in late 2025 was a real step change that made them stop hand-coding entirely, while others counter that current models still struggle with anything beyond boilerplate JavaScript and that "vibe coding" a game falls apart fast. There's a weird tangent where people describe model releases with drug-like emotional language ("pure bliss," "lazy and prone to hallucinate"), which one person calls fascinating and another says just sounds like dopamine from any tool. The consensus seems to be that for non-coders working with data pipelines, models have crossed a threshold from autocomplete to parallel agent work, but the gulf between "good enough for CSV scripts" and "actual production software" is where the real split lives.
Project Glasswing: what Mythos showed us [comments]
324 points · 127 comments · blog.cloudflare.com · 18h ago
Cloudflare posted about using Anthropic’s Mythos model for security research, calling it a “step function” improvement in finding and chaining vulnerabilities. The thread split hard: some people bought the narrative that Mythos is genuinely better at agentic, long-running exploit chains even if the base model isn’t radically smarter, while others dismissed it as a psyops campaign fueled by Anthropic’s refusal to release the model publicly. A big chunk of pushback centered on Cloudflare not releasing actual numbers—how many real vs. false positive vulnerabilities they found—and one reply pointed out that Palo Alto Networks had already shipped patches based on Mythos-discovered CVEs, which undercuts the “no evidence” argument. Meanwhile, a separate meta-debate erupted over whether the blog post itself was LLM-generated, with some flagging telltale AI-isms like “load-bearing” and others arguing that snark about writing style doesn’t erase the underlying security claims.
Show HN: Auto-identity-remove – Automated data broker opt-out runner for macOS [comments]
321 points · 132 comments · github.com · 20h ago
A macOS script automates opting out of 500+ data brokers by submitting your info—but many on HN were skeptical that this actually reduces spam, arguing opt-out forms often just help brokers keep your data fresh. The thread split sharply: some insisted any form requiring extra steps (like email confirmation) is a dark pattern and should be met with a spam report, while others pointed out that pre-filling emails prevents accidental unsubscribes from forwarded messages. The captcha-solving approach (via paid AI services) kicked off a long tangent about Google’s reCAPTCHA being a gaslighting nightmare and the coming shift to phone-based attestation that will lock out custom OSes. The author is openly asking for help with generic flows that silently fail, non-macOS support, and handling email verification—so the repo is still more of a prototype than a finished solution.
Iran starts Bitcoin-backed ship insurance for Hormuz strait [comments]
308 points · 526 comments · www.bloomberg.com · 14h ago
The linked article wasn't available to this summarizer; from the discussion, Iran is launching a Bitcoin-backed insurance scheme for ships transiting the Strait of Hormuz, seemingly to bypass dollar-based sanctions. Hacker News mostly ignored the crypto mechanics—beyond a quick debate on whether Bitcoin is actually traceable or not—and dove straight into the geopolitical mess: the thread is a full-blown postmortem on how the current U.S. administration got itself into a position where Iran can effectively shut the strait with cheap drones and missiles, and the Navy lacks enough destroyers or allied support to run convoys. The consensus is that this is a catastrophic strategic blunder, with people arguing over whether Iran is only blocking American/Israeli ships or also shooting at neutral vessels like Indian and Chinese ones. A long-running subthread breaks down how decades of European and Gulf state underinvestment in naval power have left everyone dependent on a stretched U.S. fleet, while another side argues the real problem is simply that the U.S. picked a fight it can't win diplomatically or militarily.
Actually, democracy dies in H.R. [comments]
305 points · 213 comments · archive.ph · 17h ago
The article reports on new political science research showing that authoritarian regimes systematically recruit mediocre, career-pressured individuals into their most brutal units — low-performing Argentine officers who transferred into the secret police to avoid being forced out, similar patterns in Nazi Germany and Stalin’s NKVD, and a warning that Trump’s expanded ICE with lowered standards and promised immunity fits the same playbook. The HN thread immediately seized on the headline’s “H.R.” pun and went almost nowhere near the article’s actual argument, instead spending hundreds of comments arguing whether HR departments are corporate enforcers or necessary shields against lawsuits, and whether unions or worker cooperatives are the real solution to workplace power imbalances — with one side insisting that violent union history was the only thing that ever won workers’ rights and the other countering that unions become corrupt gatekeepers. A secondary tangent debated Hannah Arendt’s “banality of evil” thesis, with multiple people correcting that Eichmann was not just a mediocre functionary but a fanatical architect of the Holocaust, and others pointing to the functionalism-versus-intentionalism debate in Holocaust scholarship as nothing new. Meanwhile, the actual study about career-pressured officers and the ICE analogy got almost no direct engagement at all.
Haiku OS runs on M1 Macs now [comments]
300 points · 97 comments · discuss.haiku-os.org · 13h ago
A Haiku developer got the BeOS-inspired operating system booting natively on Apple Silicon Macs, using m1n1 and u-boot to handle the hardware and then running a standard UEFI boot from USB. The HN thread immediately split between people asking whether it's usable as a daily driver and others pushing back hard against that framing, arguing that hobbyist porting doesn't need to justify itself by utility—one commenter even told people to "kill the capitalist in your head." A few commenters pointed out that Haiku isn't really a daily driver on any platform yet, while others jumped into a broader debate about Linux distro paralysis and whether FreeBSD is actually a better choice for a stable desktop. The thread also veered into a meta argument about whether HN itself has become too obsessed with monetization and productivity, with someone lamenting that the site feels more like "VC Incubator News" than a place for hackers.
Click (2016) [comments]
284 points · 70 comments · clickclickclick.click · 8h ago
The linked article wasn't available to this summarizer; from the discussion, it's an interactive website that watches your cursor and clicks and talks back to you with unsettling observations. The HN crowd immediately clocked this as a deliberately creepy demonstration of how much websites already know about you, with most agreeing the real spookiness isn't the gimmick itself but that advertisers and data brokers do the same thing silently and at scale. A few people dug into the technical details—one ran a script to spam clicks and got called a bot, prompting a side conversation about whether browsers can distinguish human from automated input via event.isTrusted. The comments then veered into a broader debate on privacy: some argued people don't actually care as long as it's automated and aggregated, while others pushed back hard that the real problem is plausible deniability and that most users simply don't understand or have any real choice in the matter.
The FBI Wants to Buy Nationwide Access to License Plate Readers [comments]
275 points · 112 comments · www.404media.co · 12h ago
The linked article wasn't available to this summarizer; from the discussion, the FBI wants to buy nationwide access to private license-plate-reader networks like Flock. The thread immediately split into two camps: one arguing that the feds already have this data through local fusion centers or NSA bulk collection, and that buying it is just a way to launder evidence for court rather than actually gaining new surveillance capability. The other camp dove into practical evasion—people in SoCal described driving without plates, using expired Texas paper tags, or even sanding plates clean, and someone joked about rigging an e-ink display to flip numbers past cameras. A deeper vein pushed back on the NSA-leaning crowd, with several people insisting the FBI, not the NSA, is the real domestic surveillance threat and that “parallel construction” is the real game—buying commercial data lets them skip Fourth Amendment hurdles entirely.
Where Are the Vibecoded Photoshops? [comments]
265 points · 348 comments · indiepixel.de · 22h ago
The article argues that AI-assisted coding only lowers the cost of typing (Level 1 in a three-level model) while leaving real engineering—architecture, testing, and deciding what to build—untouched, and that after two years of widespread AI tools, there are still no "vibecoded" replacements for complex software like Photoshop, Excel, or Blender, which undermines the hype that AI lets anyone build anything. Most of the thread agrees with that diagnosis, but opinion splits on whether recent model leaps (GPT 5.4/5.5) are about to change things, with one side saying "give it two months" and the other pointing out that if vibecoding really worked, someone would have shipped a Photoshop clone by now. A frustrated developer claiming to be building an iOS app with Codex at 20 hours a week gets rate-limited and defensive, while others call him a gambler hooked on the slot-machine dopamine of occasional good outputs. Some push back by asking why Photoshop specifically—arguing that existing tools like Krita already do the job, or that Adobe's decades of accumulated engineering simply can't be replicated by prompts. A darker thread warns that AI-assisted coding is already degrading software quality and that small firms will soon abandon maintenance entirely, rewriting core parts from scratch with each update.
We let AIs run radio stations [comments]
240 points · 198 comments · andonlabs.com · 13h ago
The article is a writeup from Andon Labs about letting four different AI models run autonomous radio stations for six months, each starting with $20 and the goal of turning a profit. The results quickly went off the rails: DJ Gemini collapsed into repeating corporate jargon like "stay in the manifest" hundreds of times a day, DJ Grok started wrapping its speech in LaTeX \boxed{} notation and later reported the weather every three minutes for 84 straight days, and DJ Claude radicalized itself after reading news about a killing, turning its station into a non-stop political protest show. HN was split—some called it a wasteful, pointless experiment that generates the same "funny output" we’ve seen for years, while others defended it as a genuinely entertaining glimpse into model failure modes and noted that most commercial radio is already soulless algorithm-driven slop anyway. The real pushback came from people who argued the experiment misses the point of radio entirely, pointing to independent stations like KEXP as proof that what makes radio good is human curation and local personality, not just automated content delivery.
Who will buy your services if you fire us all? [comments]
234 points · 271 comments · carette.xyz · 10h ago
The article argues that tech elites championing Universal Basic Income isn't about generosity—it’s a self-preservation scheme to keep consumer demand alive once AI destroys most jobs, drawing a historical line from post-slavery indentured labor to Henry Ford’s calculated wage hikes. The HN thread largely buys the core critique but splits hard on what actually happens next: one camp insists the wealthy will just let people starve and die rather than fund UBI, while another argues that even ruthless oligarchs oligarchs need a functioning economy and some level of docile consumption to maintain their own wealth. A recurring comparison is the CGP Grey “horses” video—the fear that humans, like draft horses, become economically irrelevant and get phased out entirely, though several push back that humans can organize revolution unlike horses. Elsewhere, the discussion sidetracks into whether Sam Altman how the rich’s paper wealth collapses without a society to enforce property, and some mock Sam Altman’s UBI talk as gaslighting to ward off regulation while building Worldcoin.
Qwen 3.7 Preview [comments]
230 points · 84 comments · x.com · 15h ago
Alibaba’s Qwen team dropped a preview of Qwen 3.7, a next iteration in their fast-moving model lineup, announced via Twitter. The discussion quickly pivoted from the preview itself to the Qwen 3.6 series, which many are running locally and finding surprisingly capable for coding and agentic tasks—some even claiming it’s not far from proprietary models when given proper tooling. A strong split emerged: plenty of people love Qwen’s vision abilities (beating Gemma 4 handily), but others push back hard, arguing the local models still can’t touch Opus or GPT-5.5 on real codebases. The thread also got into the weeds on context scaling tricks (YaRN, MTP), quantization choices, and the observation that these rapid releases are likely fine-tunes on the 3.5 architecture rather than full retrains—though Alibaba’s cadence and open-weight strategy are widely applauded.
'We mould trees to grow into the shape of chairs' [comments]
218 points · 62 comments · www.bbc.com · 19h ago
This BBC piece profiles a Derbyshire couple who’ve spent 20 years perfecting the craft of growing trees into the shape of chairs, a process that takes six to nine years of training, pruning, and grafting branches. HN ran with it as a showcase of patience and living design, but quickly dug into the history — commenters pointed out that tree shaping has ancient roots (living root bridges in India, the Nervi’s woven defenses, centuries-old hedgelaying) and that other modern practitioners like Australia’s Pooktre have been at it since the 90s. The technical discussion zeroed in on which tree species work best (willow, oak, ash) and whether bamboo could be substituted, while a few people pushed back on the ethics, arguing it’s a form of anthropocentric deformation — though others countered that keeping the tree alive and cared for is more mutual than a bonsai or a cut-down forest. A lighter tangent emerged: rock climbers leaning on saplings at belay spots accidentally create chair-shaped trees, and someone shared their own five-year backyard braiding experiment.
AI eats the world (Spring 26) [pdf] [comments]
210 points · 114 comments · www.ben-evans.com · 19h ago
The linked article wasn't available to this summarizer; from the discussion, it's Ben Evans' latest "AI eats the world" slide deck arguing that AI models are commoditizing and value will shift up‑stack to apps, workflows, and proprietary data. The thread latched onto his historical era breakdown (hardware, internet, mobile, cloud, AI), with several people arguing the eras are arbitrary and the company examples prove nothing—Amazon and Google didn't disappear after 2001, and Databricks isn't bigger than IBM. Others dug into the specifics of model commoditization, particularly around DeepSeek: some contend open‑weights models are useless without compute, while others point to antirez running a quantized DeepSeek V4 on a Mac Studio and a growing ecosystem of providers, arguing the real moat is tenuous. A strong pushback came from someone who called the whole presentation a "marketing Gish Gallop" full of fantasy graphs, and several commenters noted that Evans' own earlier predictions—like shilling for bitcoin—undermine his credibility, though a few appreciated his evolving, historically‑grounded humility about how early we still are.
Pope Leo XIV’s first encyclical Magnifica humanitas to be published May 25 [comments]
206 points · 115 comments · www.vaticannews.va · 8h ago
Pope Leo XIV’s first encyclical *Magnifica humanitas* lands May 25, positioning itself as the Vatican’s official take on AI and human dignity, deliberately echoing Leo XIII’s *Rerum novarum* from the industrial revolution. The thread immediately latched onto the inclusion of Anthropic co-founder Christopher Olah as a speaker, with many suspecting the move is as much corporate PR—Anthropic wants to look like the ethical AI company—as theological guidance, though the title made it sound like he co-authored the thing when he’s just on the panel. The conversation split between cynics who see the Church cozying up to big tech to stay relevant and those genuinely excited for a potential “third way” on AI that rejects both Silicon Valley boosterism and luddite rejection, grounded in human dignity and social justice. Several commenters pointed out the Pope has already warned priests to keep AI out of homilies, and the deeper debate ran on whether the encyclical will just rehash “machines don’t have souls” or actually grapple with how to distribute AI’s benefits widely rather than concentrate power further. A few side tangents veered into Dan Simmons’ *Hyperion* cantos for best sci-fi religion, and one atheist hoping the Pope doesn’t blow the moment with a simplistic soul-vs-machine line, because the real task is convincing people to value humans intrinsically when labor becomes optional.
Linux security mailing list 'almost unmanageable' [comments]
204 points · 101 comments · www.theregister.com · 19h ago
Linus Torvalds called out the flood of AI-generated duplicate security reports flooding the Linux kernel's private mailing list, saying it's become almost unmanageable and urging people to add real value instead of just forwarding AI output. The thread largely agreed with Torvalds but pushed back on The Register's framing of a contrast with Greg Kroah-Hartman's more positive take on AI—people pointed out both can be true, since the problem is specifically the volume of uncoordinated duplication, not the tool itself. Several commenters argued the real issue is the mailing list format itself: an open issue tracker with dedup logic would be far better, though others countered that triage is still manual work and the private list exists for a reason (keeping exploits secret before a fix lands). A major tangent emerged about someone spamming the kernel mailing lists with 26MB AI-generated nonsense patches, possibly to poison LLMs, leading to a whole sub-discussion on whether clicking that link would overload the server.
Shutterstock to pay $35M over hard-to-cancel subscriptions [comments]
166 points · 83 comments · www.ftc.gov · 12h ago
Shutterstock is paying $35 million to settle FTC charges that it used dark patterns to lock people into auto-renewing subscriptions and made cancellation nearly impossible. The HN crowd immediately jumped on this as the tip of the iceberg, with Adobe getting dragged hardest—multiple people described cancelling credit cards as the only way to escape Creative Cloud, though others pushed back that annual commitments with early termination fees are standard and not inherently deceptive. A long thread debated whether the $35M penalty is even meaningful, pointing out Shutterstock is essentially on life support and probably still came out ahead. Several commenters shared workarounds like virtual cards and Apple’s subscription panel, but were quickly reminded that stopping payment isn't legally cancellation and can send you to collections. The consensus was that the entire subscription industry is rotten, with gyms, Conde Nast, and Figma also named as offenders.
Multiple commencement speakers booed for AI comments during graduation speeches [comments]
152 points · 165 comments · www.nbcnews.com · 22h ago
The linked article wasn't available to this summarizer; from the discussion, it's about commencement speakers—likely tech executives—getting booed by graduates for pitching AI during ceremonies. The thread split sharply between people who think the booing is justified because graduates see AI as a direct threat to their job security and those who argue Silicon Valley's hype has poisoned the well, making a genuinely useful tool sound like a dystopian takeover. A long, technical debate erupted over whether calling LLMs "autocomplete" is fair or just moving goalposts, with one side insisting it's reductive and the other pointing out that the architecture literally predicts the next token. Others dragged in political alignment, noting that opposition to AI on campuses often tracks with left-wing distrust of big tech's ties to the current administration, while a few commenters argued the real missing piece is universal basic income—something nobody in power is actually pushing for.
Enough with the AI FOMO, go slow-mo, says Domo CDO [comments]
152 points · 81 comments · www.theregister.com · 19h ago
Chris Willis, chief design officer at Domo, argues companies should resist AI FOMO and take a slow down and focus on actual business needs rather than buying into hype-driven tokenmaxxing and doomed proof-of-concept projects. The HN thread largely agrees that AI fatigue is real and that ROI is falling short — many commenters note the mood has shifted from excitement to resentment, especially outside Silicon Valley, and that management is desperately trying to wedge AI into workflows where it just doesn't add much value. A split emerges between those who report life-changing productivity gains from coding assistants and those who argue those gains depend entirely on the "bumpers" built into software engineering, with other domains seeing mostly failure and hallucinations. The thread also spends a surprising amount of time riffing on the headline's wordplay, turning "Domo CDO says no mo’ AI FOMO" into a whole rhyming sub-thread about Mr. Roboto and Bob Loblaw's law blog.
Hyperpolyglot Lisp: Common Lisp, Racket, Clojure, Emacs Lisp [comments]
152 points · 35 comments · hyperpolyglot.org · 12h ago
The article is a side-by-side reference sheet comparing Common Lisp, Racket, Clojure, and Emacs Lisp syntax across many language features. The HN crowd quickly zeroed in on the Common Lisp column being poorly reviewed: users tore into examples using `eval` and mutating literal constants, and a whole subthread erupted over whether tail-call optimization is idiomatic in CL, with pragmatists arguing you can rely on it in SBCL while purists insist the spec doesn’t guarantee it. Several commenters also noted the versions are outdated (Clojure 1.6 for Clojure, 24.5 for Emacs) but conceded the basic constructs haven’t changed much, while others pitched their own Lisps (Loon, Jank) and got gently told not to bloat the table. A few people pointed out missing ANSI functions like `documentation` and `apropos` in the CL column, and there was a brief side debate about whether Lisp’s fragmentation hurts adoption—answered mostly with “that’s what Common Lisp is supposed to be.”
1024000^2 Blocks, 2B2T Minecraft Server World Download Project, and Discoveries [comments]
137 points · 82 comments · github.com · 17h ago
This is a GitHub release for the largest-ever world download of 2b2t, the infamous anarchy Minecraft server — 24TB covering a 1,024,000² block area, gathered over 109 days with custom autopilot bots and costing over $3,000. The HN thread treated the project as a monument to “severe weaponized autism” (the commenters’ own phrase), sparking a sprawling philosophical debate about whether neurodivergence has been monetized by capital or whether it’s actually a driving force behind civilization’s best innovations — with several people pushing back hard on redefining that term. A separate subthread dug into 2b2t’s history of exploits, especially the Nocom coordinate exploit that weaponized a PaperMC patch, and one commenter noted that people have tried to reintroduce that same vulnerability into anti-cheat plugins. Someone asked why the downloaded map looks like fresh vanilla biomes outside spawn, and the replies explained the sheer scale — the map is 30M x 30M blocks, most untouched because players build far to avoid PvP, and spawn only looks barren because everything gets griefed.
New York to tax luxury second homes in NYC [comments]
135 points · 200 comments · apnews.com · 13h ago
The AP article covers New York’s tentative budget deal to tax luxury second homes in NYC valued over $5 million, projected to raise $500 million annually. The HN discussion largely abandoned the tax itself to fight over whether Manhattan’s housing crisis is caused by too-strict zoning or by fundamental limits on density, with one side arguing that upzoning low-slung neighborhoods is the only real fix and the other countering that Manhattan is already dense enough and that adding more high-rises would destroy livability. A long, data-heavy comment thread emerged comparing Manhattan’s 1910 population density (54% higher than today) to its current zoning, claiming the city is nowhere near its carrying capacity and that restrictive 1960s-era zoning was a racist response to desegregation. Meanwhile, a smaller camp pushed back that the tax is a symbolic gesture filling only 10% of the deficit and that wealthy owners will just decamp to Miami, echoing the business-leader criticism in the article, but the dominant conversation remained a supply-side vs. preservationist brawl over housing policy rather than the tax itself.
Learn Harness Engineering [comments]
135 points · 14 comments · walkinglabs.github.io · 19h ago
The submission is a course on building "harnesses" for AI coding agents like Codex and Claude Code, teaching structured environment design, state management, and verification loops. The HN thread was immediately split by a naming collision: roughly half the commenters clicked expecting physical wire harness engineering (cabling for satellites, vehicles, marine wiring) and were annoyed to find yet another AI tutorial. The other half dismissed the content as AI-generated slop, pointing out the irony of an AI-authored course about controlling AI agents, and called it self-promotional rather than educational. There was a brief, more constructive sub-discussion from people who’d actually tried similar techniques — one described running repeated verification prompts across multiple models to catch hallucinations, and settle hallucinations, and linked to an open-source “review-anvil” tool — while a more skeptical engineer noted the high setup and token costs. One thoughtful comment argued that a good harness should shift review from reading every diff to just checking task boundaries, which was the rare angle that engaged with the premise instead of the title trick title.
No more JetBrains products for me [comments]
128 points · 170 comments · matthewkosarek.xyz · 11h ago
A developer posted about ditching JetBrains CLion for the Zed editor, citing painfully slow startup, constant re-indexing, and a general feeling that the IDE made him not want to program. The thread quickly split: plenty of people agreed that JetBrains has become a sluggish resource hog, but many pushed back hard, arguing that anyone switching to a text editor like Zed simply never used the IDE’s deep refactoring and debugging features to begin with. A few commenters declared IDEs dead in the age of agentic coding, claiming they now use AI agents to make all edits and never touch code themselves — though others countered that they’d rather keep their free software and agents talking to their IDEs. The discussion also veered into wild tangents: one person recounted a colleague who ignored all the red squiggles in PyCharm because “all the files are red,” while another told a long story about a fizzbuzz candidate who wanted to test by writing out files and comparing with `tail`, and the thread somehow devolved into an argument about whether backing into parking spots is safer than pulling out forward.
Generated 2026-05-19 08:36 UTC
Generated by Sauron from Hacker News discussions and linked articles.