HN Brief: 2026-05-31
Today’s HN was dominated by a mood of erosion—software you thought you owned getting remotely bricked, AI-generated slop poisoning consulting reports and core infrastructure, and a collective reckoning with whether domain expertise still counts for anything when the models can fake it. The big pattern was a three-way tension: angry pushback against planned obsolescence and LLM-induced breakage, a stubborn belief that deep knowledge still beats prompt engineering, and a quieter thread about whether anyone can even afford the token burn anymore. A surprising number of threads circled back to the idea that the old quality-assurance contract is broken when the same people shipping AI are the ones who built the safeguards.
The thread “Microsoft Office 2019 and 2021 for Mac view-only conversion” is worth clicking for the raw anger and legal strategizing over Microsoft’s plan to remotely degrade perpetual licenses into view-only mode. “Domain expertise has always been the real moat” delivers the sharpest AI optimism vs. pessimism cage match of the day, with working engineers trading concrete counterexamples. The EY citation hallucination scandal in “EY Canada published a cybersecurity report and most citations were hallucinated” is a depressing case study in AI-launched consulting slop—if you can survive the scrolljacking. The rsync maintainer’s AI-assisted commit flood, documented in “Please Do Not Vibe Fuck Up This Software,” captures the exact moment volunteer maintenance meets LLM entropy. And “Leo's first encyclical attacks technological messianism” offers an unexpected institutional critique of techno-optimism from the Vatican, which HN both mocked and engaged with seriously.
Microsoft Office 2019 and 2021 for Mac view-only conversion [comments]
808 points · 286 comments · consumerrights.wiki · 8h ago
A wiki is documenting Microsoft's plan to remotely degrade perpetually-licensed Office 2019 and 2021 for Mac and iOS into a view-only mode on July 13, 2026, by letting a license-validation certificate expire without renewing it for older build versions. The thread largely split between calling this an intentional, calculated move to force people onto subscriptions and arguing it's just another example of how fragile certificate-based licensing is—though most agreed Microsoft quietly rewriting its 2023 end-of-support page to remove the "continue to function" promise makes intent pretty obvious. Some commenters went straight to "this is theft" and recommended small claims court, but others pointed out Office's EULA requires binding individual arbitration, so good luck with that. A few speculated the sudden hard deadline might be about stopping AI labs from using cheap perpetual licenses on cloned VMs for agent workflows, while the rest saw it as classic Microsoft enshittification that's been going on for decades.
Domain expertise has always been the real moat [comments]
527 points · 311 comments · www.brethorsting.com · 11h ago
The post argues that domain expertise—knowing what “right” looks like in a specific field like payroll or logistics—has always been harder than writing the code, and that AI agents now make coding cheap while leaving that deep knowledge as the only real moat. Hacker News split hard on this: a lot of people nodded along, especially engineers who’ve worked on complex regulated systems, and they offered concrete examples like title insurance where LLMs still hallucinate jurisdictional nuances and can’t replace sitting with domain experts. But a vocal counterargument accused the post of being cope—LLMs already encode tons of domain knowledge, and you can query your way to proficiency faster than learning an industry from scratch, so calling domain expertise a “moat” is just moving the goalposts before AI eats that too. The most intense exchange was someone arguing that all these “coding was never the hard part” articles are self-delusion, pointing out that AI is on a trajectory to dominate every layer, including expertise, and that trying to find a safe niche is itself a predictable pattern.
OpenRouter raises $113M Series B [comments]
412 points · 198 comments · openrouter.ai · 14h ago
OpenRouter raised a $113M Series B led by CapitalG and joined by a slew of enterprise infrastructure VCs, positioning itself as the routing and gateway layer companies need as they move from single-model pilots to multi-model production. The thread largely agreed that OpenRouter is the best way to try new models without wrangling a dozen APIs, but the main tension was over the 5% surcharge — some argued that convenience, consolidated billing, and hard spending caps justify it, while others pointed out that heavy users of expensive models like Claude Opus should go direct to save money, and that the surcharge isn’t worth it at scale. A few commenters pushed back on the need for a proxy at all, suggesting an open-source library plus prepaid virtual cards could handle billing and routing, but defenders countered that even Cloudflare and Vercel offer similar gateways and the real value is in competition among providers and not getting locked into one vendor. Privacy and data-use concerns came up repeatedly: free models on OpenRouter likely feed training data, though the platform does let you filter for zero-data-retention providers, and one technical tangent noted that routing through OpenRouter breaks DeepSeek V4’s caching, undermining a key pricing feature.
Anthropic surpasses OpenAI to become most valuable AI startup [comments]
400 points · 457 comments · qazinform.com · 18h ago
The article reports that Anthropic has overtaken OpenAI in valuation after a massive funding round, driven by Claude’s popularity and revenue growth. The Hacker News thread largely ignored the financial milestone and spiraled into a fierce, highly technical debate about whether Claude or GPT models produce better code. Developers split into camps: some argued Claude Code has more “taste” and handles underspecified prompts elegantly, while others insisted GPT-5.5 Codex is far superior for large, complex refactors and less prone to hallucinating impossible solutions. A few commenters pushed back hard, claiming blind tests prove the output is indistinguishable and that preference for Claude is just marketing and vibes—comparing it to Coke vs. Pepsi or luxury cars. The thread also saw sarcastic one-upmanship about model version numbers (“Opus 6.9 is old news, 6.91 is the real SOTA”) mocking the pace of releases.
Pandoc Templates [comments]
391 points · 50 comments · pandoc-templates.org · 22h ago
The linked article is a curated directory of Pandoc templates for turning Markdown into polished PDFs, HTML, and other formats. The thread quickly turned into a broader debate about whether Markdown + Pandoc is actually better than WYSIWYG tools like Word—some argued Markdown gives programmers control and versioning sanity for complex documents like theses and novels, while others pushed back hard, saying for short simple documents, most people don’t want to wrestle with templates and styles, and a good word processor with paragraph styles is faster and less error-prone. There was also a strong contingent of heavy users sharing real pain points: tables that break, Unicode characters silently dropping, and page-break control being a nightmare in LaTeX pipelines, with several commenters pointing to Typst as a better modern backend for Pandoc output. A few people had no idea this template directory existed and were delighted, while others dropped in with alternatives like Quarto and Metanorma, making the thread as much about the ecosystem around Pandoc as about the templates themselves.
Openrsync: An implementation of rsync, by the OpenBSD team [comments]
380 points · 152 comments · github.com · 21h ago
The linked article is a new, BSD-licensed implementation of rsync by the OpenBSD team, compatible with protocol 27 but initially missing some flags like `--exclude` and `-z`. The thread quickly turned into a licensing war, with people arguing whether BSD or GPL is actually "more open"—BSD partisans saying the GPL is coercive, GPL partisans saying BSD lets companies rug-pull open-source projects (Redis and Valkey came up as a case study). A huge tangent then erupted over AI contamination in open-source: someone noticed Postfix had an AI-assisted commit, which set off a bitter split between people who think AI-generated code is radioactive slop that should be avoided at all costs, and boosters who insist you're wasting your time fighting the future and that manual programming will be as quaint as punch cards in a few years. The technical details of openrsync itself were almost an afterthought, though a few people confirmed it works great if you don't need `--exclude`—which, as of OpenBSD 7.9, now works.
Zig: Build System Reworked [comments]
349 points · 226 comments · ziglang.org · 23h ago
Andrew Kelley landed a big Zig build system rework that splits the configurer and maker into separate processes, cutting `zig build --help` from 150ms to 14ms and slashing instruction counts by 95%. The thread quickly split: some devs cheered the speed numbers, but others pushed back hard, pointing out that right now Zig’s initial compiles and ZLS rebuilds are still painfully slow—one person reported a trivial `zig test` taking multiple seconds on a Ryzen Linux box, while a counter-example on an M4 Mac showed 94ms, suggesting the experience varies massively by platform and workload. A long, tangy debate erupted over whether Zig’s compilation times are already “terrific” compared to Rust and C++, with a side argument about Go’s influence on modern tooling and a nostalgic detour into Turbo Pascal’s compilation speeds. There was also real frustration from users who hit breaking API changes in build scripts without seeing the performance improvements they’d hoped for, making the “non-breaking from an API perspective” claim in the devlog feel hollow to anyone who’s had their build.zig rot in a few months.
EY Canada published a cybersecurity report and most citations were hallucinated [comments]
295 points · 131 comments · gptzero.me · 12h ago
GPTZero’s investigation found that a 2025 Ernst & Young Canada cybersecurity report on loyalty fraud was basically a collage of AI-generated slop — 72% of its citations were hallucinated, URLs led nowhere, and the numbers contradicted themselves between pages. HN immediately got sidetracked by the investigation’s own website, which had such aggressive scrolljacking that multiple people literally couldn’t read the article on mobile or desktop, and someone joked that maybe the site was “vibe-coded” too. Once people fought through the UI, the real meat was a resigned consensus that this is what happens when consultants and professionals stop vetting AI output — the report was never reviewed by anyone who knew the domain, and the authors just laundered a fake McKinsey citation from a random fintech blog into a Big Four publication. The broader take was that the “generate now, review later” culture is poisoning the knowledge pool, and that vetting AI slop often takes longer than just doing the work yourself, so firms will either create dedicated AI-QA roles or keep pumping out garbage that ends up in newspapers, AI search summaries, and future training data.
Voxel Space (2017) [comments]
277 points · 58 comments · s-macke.github.io · 17h ago
The submission is a deep dive into the voxel-based terrain rendering engine behind the 1992 game *Comanche*, which used height and color maps to create textured, shaded landscapes entirely on the CPU. The comments quickly got into the weeds on rendering order: several people pointed out that the article presents back-to-front painting for clarity, but the real engine ran front-to-back with a y-buffer to eliminate overdraw — one person even recalled that VGA's plane-based memory layout made column-at-a-time rendering faster, a detail the author acknowledged. A substantial semantic debate broke out over whether this technique qualifies as "voxels" at all, with one side arguing it's just a height map of prisms (no overhangs or gaps) and the other countering that a height map is still a restricted form of volumetric data. There was also nostalgia from someone who independently re-invented lookup tables on a 286 as a teenager, and a tangential thread about how Westwood's *Blade Runner* used a similar "slice" approach that its own developer, Louis Castle, explicitly called "voxels plus" rather than true voxels.
Accenture to acquire Ookla [comments]
276 points · 141 comments · newsroom.accenture.com · 15h ago
Accenture is buying Ookla—the company behind Speedtest, Downdetector, and Ekahau—for an undisclosed sum that the thread assumes is north of a billion dollars. Most of the discussion dismissed the idea that the technology itself justifies that price; the real value is the brand recognition, the massive dataset of real-world network performance (250 million tests per month), and the fact that every ISP and frontline support worker already treats Speedtest as the de facto benchmark. A lot of pushback centered on Downdetector, which several people pointed out is basically just a guestbook that counts search traffic for “is $X down” rather than actually checking services, meaning it’s often misleading. Others dug into how ISPs game speed tests by prioritizing Ookla endpoints, and the whole thing sidetracked into a comparison of alternatives like Cloudflare’s speed test and fast.com, with a brief mention of Accenture’s past child-safety controversies as a sour note on the buyer.
Shantell Sans (2023) [comments]
212 points · 18 comments · shantellsans.com · 9h ago
The article details the creation of Shantell Sans, a variable font by artist Shantell Martin and type designer Stephen Nixon that mixes axes for weight, italic, informality, and bounce, built as a more refined and functional successor to Comic Sans. The HN crowd was overwhelmingly positive, with several people noting that it felt far more readable than Roboto in side-by-side tests, and one person’s dyslexic daughter gave it a big thumbs up. The variable axes—especially the formality slider and bounce—were called out as genuinely clever engineering, with comparisons to Metafont and a sense that variable fonts are finally delivering on their promise. A recurring desire emerged for a monospaced version to use in terminals, with people sharing alternatives like Recursive Mono Casual but lamenting none matched Shantell Sans's beauty. One side discussion wondered whether corporate brands could adopt it site-wide without looking unprofessional, with suggestions that it could work well for headings or accessibility toggles rather than body text.
Leo's first encyclical attacks technological messianism [comments]
210 points · 255 comments · www.economist.com · 21h ago
The linked article wasn't available to this summarizer; from the discussion, it's about Pope Leo's first encyclical taking aim at the belief that technology alone will save humanity. The thread immediately turned into a fight over who gets to control technology—engineers, elected governments, or now the Church—with one camp insisting democratic elections are the answer and another retorting that politicians don't understand the tech they’re supposed to regulate, leaving voters stuck choosing between two brands of “smooth-brains.” A vocal side argued that AI is actually more centralized than any prior tech, making decentralization a fantasy given Manhattan-sized data centers, while others pushed back pointing to open-weight models as proof the genie can be shared. The conversation also veered into a bitter, multi-level argument about steel tariffs and manufacturing jobs, where someone claimed tariff supporters knew the policy would kill more construction jobs than it saved, only to get accused of bad faith and socratic snark.
Zig ELF Linker Improvements Devlog [comments]
198 points · 65 comments · ziglang.org · 14h ago
Zig's new ELF linker, still opt-in behind `-fnew-linker`, just hit a major milestone: it can now build the self-hosted compiler with LLVM and LLD enabled, and incremental rebuilds on x86_64 Linux go from a 36s first build down to ~250ms. The thread mostly skipped the technical details to debate whether this work is a response to the "Bun drama"—the authors showed up in the comments to firmly say no, it's years of steady progress finally getting attention. A chunk of the conversation drifted into language wars: people argued Zig is the "new C" and Rust is the "new C++," others pushed back that Zig's lack of a 1.0 and the removal of `@cImport` make it a hard sell for gradual porting, and a few commenters held up Go, Nim, and even Turbo Pascal as already having fast incremental compilation. There was also a brief but sharp exchange comparing the Rust community's evangelism and code-of-conduct culture to Zig's more laid-back vibe, with one person calling Rust "bondage and discipline" and Zig "optimized for fun."
Hormuz crisis side effect: a sharp rise in container shipping rates [comments]
184 points · 155 comments · www.lloydslist.com · 13h ago
The article details how the effective closure of the Strait of Hormuz is sending container shipping rates skyrocketing, with spot rates doubling since late February and carriers passing on massive fuel cost increases. The HN thread quickly veered into second- and third-order effects: the most concrete tangent was about empty containers becoming a liability, as it's no longer worth the cost to ship them back, leading to ports forcing empties onto truckers, collapsing container resale prices in places like Fontana, and even creating oddball arbitrage where operators fill cans with cheap commodities like hay just to justify the return trip. A long, heated split emerged over whether economists properly account for externalities—some arguing that the field actually does, pointing to Coase and Pigouvian taxes, while others countered that real-world decision-makers systematically ignore them, using climate change inaction as Exhibit A. The geopolitical debate was even sharper: one camp argued the crisis was a foreseeable, even desired outcome of the US attack on Iran, while the opposing side insisted the administration is simply incompetent, having torn up the nuclear deal only to now fight for the same conditions, and that any talk of "finishing the job" is fantasy given Iran's size and the alienation of allies.
The AV2 Video Standard Has Released (Final v1.0 Specification) [comments]
166 points · 58 comments · av2.aomedia.org · 10h ago
The AV2 video standard just hit its 1.0 spec release from the Alliance for Open Media, promising 20–30% better compression than AV1 plus multi-stream support for VR and live sports. The thread immediately grounded expectations: the reference encoder chugs at around 1fps on good hardware, so real hardware decode won’t show up until ~2028, and widespread streams are a 2030 proposition at best—repeating AV1’s glacial rollout. Some argued software encoding is fine for low-bitrate mobile calls and you don’t actually need hardware for distribution platforms where encode cost is amortized across thousands of streams. The Dolby patent threat came up again, but several people pushed back hard on calling Dolby a “patent troll,” noting they actually invest in R&D, while others pointed out the legal defense fund for baseless claims exists and has held so far. A lighter tangent emerged around naming the inevitable decoder library—dav2d doesn’t roll off the tongue, and the pun attempts (2av2quit, D4vd, “deuvid” in French) got more play than any technical detail did.
Corporate America Is Starting to Ration AI as Cost Skyrockets [comments]
164 points · 154 comments · www.wsj.com · 19h ago
The linked WSJ piece wasn’t available to this summarizer; from the discussion, the article argues that companies are starting to ration AI use as costs balloon, but the thread largely dismissed that framing as missing the real problem. The overwhelming consensus among working developers is that the cost crisis is self-inflicted—teams are lazily feeding entire logs and recurring tasks into frontier models instead of using AI to build proper deterministic tools, creating a perpetual token-burning loop. Several people pointed out that this “vibe slop” approach is producing PRs full of agent-generated entropy that takes more human time to review than it saves, and that the real split isn’t about cost but about whether AI is being used as a crutch for people who don’t understand their own domain. A vocal minority pushed back, arguing that the hype has clearly failed to deliver transformative products or revenue growth that justifies the trillion-dollar capex, and that comparable breakthroughs in solar or EVs are actually reshaping the real world while AI just churns out fancier todo apps.
Memory decline after menopause linked to loss of estrogen production in brain [comments]
148 points · 66 comments · news.northwestern.edu · 17h ago
A new Northwestern study in mice suggests memory decline after menopause stems from estrogen loss in the brain's extracellular matrix—the "mortar" between cells that makes up 20% of brain volume—and proposes targeting that matrix as a potential treatment. HN immediately pushed back on the article's framing that "nearly two-thirds of Alzheimer's patients are women" by pointing out that women simply live longer, so more reach the age where Alzheimer's manifests; one commenter dug up Australian data showing the sex difference shrinks when you control for life expectancy, and another quoted the Alzheimer's Association itself saying most U.S. studies find no incidence difference between men and women at any given age. The thread then veered into a heated debate about the Women's Health Initiative's flawed 2002 study that black-boxed hormone replacement therapy for twenty years, with several people arguing the real institutional failure was scaring women off estrogen that could have protected their brains. A bizarre side conversation argued that men have comparable estrogen levels to non-ovulating women, and another wondered whether baldness is inversely correlated with dementia—though that went nowhere.
AI job grief: A psychological crisis hitting tech workers [comments]
143 points · 136 comments · jackmaguire.org · 17h ago
The essay argues that AI-driven job displacement is creating a genuine grief response—complete with a proposed clinical label, "Artificial Intelligence Replacement Dysfunction"—because the threat strikes at professional identity, not just income. Hacker News immediately pushed back on the article’s claim that manufacturing workers didn’t identify with their labor, calling that narrow and laughable, and several people dismissed the heavy reliance on Reddit quotes and upvote counts as "human slop" that undermines the argument. A strong split emerged: some commentators insisted the grief framing is counterproductive propaganda meant to weaken anger, pointing to students booing Eric Schmidt or a Molotov cocktail thrown at Sam Altman’s home as acts of agency, not grief. Others shifted the focus to economics, arguing the real sadness is losing one’s place in society, which sparked a debate over universal basic income—with most agreeing UBI is politically impossible in the US given the deep resistance to welfare. A tangent about chess players improving with AI and a pelican-drawing meme briefly derailed things, but the core tension was whether naming the feeling "grief" helps or just papers over the need for raw anger and systemic change.
To have a moral stance on AI is to be an outcast, and it sucks [comments]
129 points · 286 comments · musings.martyn.berlin · 15h ago
The article is a personal essay from a technologist who is vehemently anti-AI, describing the social and emotional toll of holding that stance in tech circles. HN pushed back hard on the framing: many argued the author’s moral stance isn’t the only possible moral stance, and that presenting it as self-evident while complaining about being an outcast feels like persecution theater. Others pointed out that the real issue isn’t AI itself but centralized big-tech profit-maxing, and that local or open models could sidestep many of the harms. A significant split emerged around whether most people actually love or hate AI—some cited surveys showing broad public skepticism, while others countered that usage numbers tell a different story, and that many people use AI while hating the prisoner’s dilemma it forces on them. There was also sharp disagreement on the environmental arguments, with some calling water/energy concerns overblown and others insisting the datacenter buildout is already straining grids and is a genuine problem.
Show HN: Helios – what plug-in solar could generate for any address in Britain [comments]
118 points · 40 comments · helios.southlondonscientific.com · 20h ago
Plug-in solar panels just became legal in the UK without an electrician, and Helios is a new web tool that uses government LIDAR data to ray-trace the actual skyline at any address and estimate what a typical balcony kit would generate and save. Hacker News liked the clever use of open data—the UK's LIDAR program got its own appreciative tangent—but several people immediately flagged the tool's geographic blind spots: it falls back to a less accurate synthetic horizon for most of Scotland and Wales, and if it can't find your house in OpenStreetMap it defaults to a postcode centroid, which can produce misleadingly precise shading results for a street with gaps. The financial math split the room: some argued a 7-year payback was fine given 25-year panel warranties and that solar tech isn't a fast-moving consumer good, while others countered that £1,000 net over twenty years is pitiful and that German kits are already half the price, making the UK pricing look inflated. The creator was in the thread taking feature requests (mounting on a shed, picking the best spot in a garden, integrating with house-hunting tools) and said they'd look into Inspire Index polygons as a fallback for building footprints, but didn't commit to open-sourcing the code.
Show HN: 500 years of Joseon court omens as an observability dashboard [comments]
115 points · 19 comments · ajin.im · 12h ago
An observability dashboard reimagines 500 years of Joseon court omen records—droughts, tiger incursions, guest stars—as real-time system telemetry, complete with a Mandate Volatility Index and SLAs for rain rituals. The thread loved this framing, with several people noting how the Joseon historians’ refusal to let even the king censor the record (one king fell off his horse and ordered them not to log it; they logged the order) perfectly matches the ethos of observability. A tangent about UFOs in the archives got reframed as incoming bolides, and someone quipped that the current U.S. president being bitten by a pardoned turkey would be a very bad omen—sparking a short riff on what actionable intelligence the “mandate of heaven” would actually give you. The dominant reaction was pure delight: this is a spectacular use of free will, and one commenter called it the definition of a nerd snipe that merged their two obsessions.
Microcode inside the Intel 8087 floating-point chip: register exchange [comments]
112 points · 18 comments · www.righto.com · 14h ago
Ken Shirriff’s latest reverse-engineering deep dive walks through the 8087’s microcode for the FXCH instruction, showing how its 1648 16-bit micro-instructions handle register exchange with empty-check exceptions and NaN substitution. The thread immediately latched onto why Intel used microcode at all, with the consensus that it saved die space and made bug fixes manageable — high-end IBM System/360s and NEC’s V33 went hard-wired for speed, but that was rare and risky. A commenter floated the idea of exposing the microcode to users like a RISC or VLIW, but others pointed out that fast cache was too expensive in 1980 and the 8087 shared the bus with the 8086, so fetching micro-instructions from RAM would interfere with the main processor. The discussion also surfaced a practical horror story from an Intel validation engineer: over half the bug reports were people getting different results from different instruction orderings and insisting one was “right.” There was a technical split over whether single-cycle RISC fetches could keep up with RAM speed — one person argued they could, another said that only worked during a brief era when RAM was faster than the core.
Please Do Not Vibe Fuck Up This Software [comments]
108 points · 32 comments · github.com · 4h ago
The linked GitHub issue documents community fury over rsync’s maintainer using Claude AI to churn out 26,000 lines of changes in two months, breaking decades of stability. The thread quickly splits: one side reels off a list of specific regressions—build failures on old Linux, broken `--delete-missing-args`, Darwin bustage—and users are already pinning old versions or migrating to OpenBSD’s openrsync; the other side argues the issue itself is obnoxious drive-by hate, that a volunteer maintainer asking for help for years deserves leeway, and that Claude’s co-author line doesn’t always mean the AI touched code. There’s also a meta-shoving match about whether the comments are “just hate” or legitimate alarm over a mature, safety-critical tool being treated as an LLM playground—with one observation that the whole mess underscores how the old quality-assurance practices can’t keep up when the same people pushing AI are the ones who built those practices.
Rsync 3.4.3 has hundreds of Claude commits [comments]
99 points · 62 comments · mastodon.gamedev.place · 23h ago
A Mastodon post reported that the latest rsync update (3.4.3) broke incremental backups using `--compare-dest`, and digging into the repo showed dozens of commits co-authored by Claude alongside original author Andrew Tridgell, with no clear changelog entry for the breakage. The HN thread quickly split between alarm over AI-generated code landing in core infrastructure and pushback that nobody had actually pinpointed which commit caused the regression—some noted the Claude commits were mostly tests, CI, and CVE fixes, not feature slop. Others argued Tridge’s reputation for careful engineering makes the situation *more* worrying (“if he can’t handle LLMs without a slopocalypse, no one can”), while defenders countered that we have no idea how he’s using the tool and that blaming AI for every bug is lazy. A few commenters pointed out that the real issue isn’t AI per se but the temptation to skip proper review when tired or stressed, and that FreeBSD’s informal ban on LLM-generated code highlights the growing divide between corporate AI adoption and open-source quality expectations.
Jef Raskin, the Visionary Behind the Mac (2013) [comments]
99 points · 42 comments · lowendmac.com · 12h ago
A long-form interview with Jef Raskin, Apple employee #31 and the original founder of the Macintosh project, was republished from 2005, covering his vision for an appliance-like computer and his sharp critiques of where the Mac ended up. The thread immediately zeroes in on a major historical dispute: Raskin flatly denies the "false legend" that he wanted a text-based machine, insists he always planned a graphical interface, and claims Andy Hertzfeld's folklore.org is "full of errors." That gets met with a heavy counterpunch, as someone pastes Hertzfeld's own account arguing Raskin *did* oppose the 68000 processor and the mouse—the two things that define the Mac—and that the shipped product had almost nothing to do with his ideas. The conversation then veers into the Canon Cat as the real test of his philosophy, with some arguing it proved his approach was a dead end dedicated to word processing, while others point to incremental search and his book *The Humane Interface* as quietly influential ideas that outlast the debate.
Cheese Paper: a text editor specifically designed for writing [comments]
99 points · 23 comments · brie.gay · 9h ago
Cheese Paper is a new open-source text editor that stores fiction scenes, notes, and worldbuilding metadata in flat Markdown files with TOML headers, designed to play nicely with SyncThing and survive being edited in any other app. Most of the thread pivoted straight to comparisons with Scrivener and Manuskript — the consensus was that Manuskript has more features and a decade of development, but Cheese Paper’s choice to just use a folder of plaintext files (instead of a special database) won genuine appreciation from people who’ve been burned by CherryTree’s immobile format. A few writers pushed back on the compression of the UI, arguing it wastes horizontal space for no ergonomic reason, while others loved the “no telemetry, keep your data to yourself” stance. The oddest tangent was someone dreaming up a paragraph-as-linked-nodes editor, which got pointed toward Bike (Mac-only) and a suggestion to just vibecode the prototype.
Dusklight – GC Twilight Princess Decompiled [comments]
95 points · 11 comments · twilitrealm.dev · 11h ago
The linked article describes Dusklight, a native PC and mobile port of *The Legend of Zelda: Twilight Princess* built on top of the completed ZeldaRET decompilation project, which reimplements the game's engine and requires users to supply their own disc-image assets. On HN, the thread quickly split into two camps: one side was purely technical, impressed by the cross-platform support (including Android and a `brew install` option) and curious about how the port differs from a generalized emulator, while the other side dove straight into a heated debate about whether *Twilight Princess* itself is even a good game. A long, pointed takedown argued the game is a "hodgepodge" of mismatched parts—forced wolf segments, empty world, a story that chickens out with Ganondorf—and accused it of being the last game where the author gaslit themselves into liking something expensive. That critique got pushback from people who still rank it alongside *Ocarina of Time*, with one commenter blaming the fractured design on its rushed conversion to a Wii launch title. Separately, a question about the broader boom in decompilation projects prompted a quick correction: someone claimed AI has made it much easier, but another person shot that down, arguing AI-assisted decompilation isn't clearly transformative under fair use and might not hold up legally.
Parallel Reconstruction of Lawful TLS Wiretapping [comments]
90 points · 40 comments · remyhax.xyz · 12h ago
The article reconstructs a 2023 attack on Russia's largest XMPP service where an unknown actor exploited an acme.sh remote code execution vulnerability to get fraudulent TLS certificates issued for interception, with the whole scheme only uncovered because the attackers forgot to renew one. The HN discussion immediately zeroed in on the fact that Certificate Transparency worked exactly as designed here—the certs were publicly logged—but domain owners simply weren't monitoring the logs, and the real vulnerability is that ACME client software runs with excessive privileges and handles validation poorly, not the CA infrastructure itself. Several people pushed back hard on the post's use of "parallel reconstruction," pointing out that term has a specific meaning in law enforcement (fabricating a lawful justification for evidence obtained unlawfully) and this is just reverse engineering, though the author chimed in to say the title was intentionally a play on words about the nuance. The debate quickly shifted to whether CAA records with DNSSEC actually prevent this—turns out the extended CAA attributes for account URI and validation method aren't mandatory yet, and even if they were, if the attacker controls routing at the VPS level (which Hetzner presumably did here), they can just pass ACME challenges normally and get valid certs regardless. A recurring argument was that the whole Web PKI model is fundamentally broken because any CA can issue for any domain, and the only real fix would be to tie certificate issuance to domain registrars as the sole authority, but nobody has the clout to push that through while governments and criminals quietly benefit from the status quo.
Thiel moves family to Milei's libertarian Argentina [comments]
88 points · 50 comments · www.ft.com · 12h ago
The linked article wasn't available to this summarizer; from the discussion, Peter Thiel has moved his family to Argentina under Javier Milei's libertarian regime, buying property in Buenos Aires and Uruguay. The thread treats this less as a lifestyle choice and more as a getaway: the dominant take is that Thiel is fleeing the consequences of the instability he helped create in the US, with several people drawing direct parallels to Nazi officials who fled to Argentina after WWII. There's sharp pushback on the internal logic—Palantir's main revenue comes from government contracts (i.e., taxes), so Thiel's libertarian anti-tax rhetoric gets called performative hypocrisy, and the idea that Argentina (with its history of military coups and economic collapse) offers stability is widely mocked as absurd. A side argument breaks out over whether the story should be on HN at all, with some calling it off-topic politics while others insist a tech billionaire decamping to a "libertarian paradise" is exactly the kind of phenomenon worth discussing.
wolfSSL releases a new product; wolfCOSE a zero alloc C embbedded COSE stack [comments]
87 points · 17 comments · github.com · 11h ago
wolfCOSE is a new lightweight C library from wolfSSL that implements CBOR Object Signing and Encryption (COSE) for embedded systems, boasting zero dynamic allocation, a tiny footprint, and support for post-quantum signing with ML-DSA. The HN thread quickly clarified that COSE is essentially a binary version of JOSE (the JSON-based JWT family), with CBOR being a binary JSON alternative that avoids some JSON parsing ambiguities—though one commenter pointed to the RFC’s own warnings about CBOR encoding pitfalls. The main debate landed on the “zero alloc” claim: several people argued that while the library avoids heap allocation (no malloc), it stacks sizable local arrays, which can be just as dangerous on memory-constrained microcontrollers where stack overflow silently corrupts memory. Others countered that “zero alloc” has always meant no dynamic heap allocation, that stack usage can be statically analyzed with the right tooling, and that the library explicitly states all buffers are caller-provided—so the real issue is managing stack depth, not a misleading label.
Generated 2026-05-31 08:19 UTC
Generated by Sauron from Hacker News discussions and linked articles.