HN Brief: 2026-06-03
Today's HN was split between fresh frustrations with Big Tech's overreach and nostalgic retreats to simpler tools. The Gmail rant catalyzed a broader reckoning with AI forced into everything, while the Adafruit vs. Flux.ai drama showed how quickly a security disclosure can turn into a pile-on against a half-baked product. A quieter thread about FidoNet offered the exact opposite energy—communal, permissionless, and weird. The Larry Ellison surveillance quote landed like a punch, and the crowd was in no mood for "both sides."
"Please don't spam people looking for employment. It's just cruel" — the raw, personal rant that ignited a real discussion about AI scraping of vulnerable job-seekers. "1-Click GitHub Token Stealing via a VSCode Bug" — a terrifyingly clean exploit, made worse by the author’s decision to go full public disclosure after Microsoft stiffed him on credit. "CT scans of BYD car parts" — a beautiful visual essay that devolved into a brutal debate on whether right-to-repair is dead or just sleeping. "Adafruit receives demand letter from Fenwick legal counsel on behalf of Flux.ai" — the thread where everyone discovered Flux.ai is apparently a token-burning scam, and the lawyers just made it famous. "A walking tour of surveillance infrastructure in Seattle" — the argument wasn't about surveillance; it was about whether Seattle even has effective policing to worry about.
Please don't spam people looking for employment. It's just cruel
919 points · 259 comments · news.ycombinator.com · 18h ago
The linked article wasn't available to this summarizer; from the discussion, the submission is a raw, personal rant from an unemployed man who posted in HN's "Who wants to be hired?" thread and got spammed by what appears to be an AI-driven service pitching development work, crushing the tiny glimmer of hope each email gives him. The thread largely rallied around him, with many sharing similar stories of getting scraped from those threads by bot-like spam—one person flagged a specific tool called "Alya," which brands itself as the creator's "daughter," and a lot of commenters found that branding deeply creepy rather than endearing. A significant split emerged over whether this is a tech-enabled scam versus just garden-variety spam; people pointed out that some of these emails are actually North Korean intermediaries trying to get US-based contractors to launder their access, and others noted the emails are so generic they clearly misread "Who wants to be hired?" as "Who's hiring?" There was also a practical back-and-forth about defenses: a few people argued you should put your email only in your HN profile, not in the post body, but others shot that down by saying the scraping is already moving to profiles anyway.
Gmail thinks I'm stupid, so I left [comments]
884 points · 539 comments · moddedbear.com · 12h ago
The author details their frustration with Gmail's aggressive AI features—unsolicited message summaries, auto-generated replies, and constant prompts to use a "help me write" button—which they feel signal that Google thinks they're incapable of handling their own email. The thread largely agrees this is a death-by-a-thousand-cuts driven by OKR culture, though some push back that they haven't seen these features and suspect it's a phased rollout or tied to specific settings. A major split emerges over whether Apple is any better: one side argues Apple's track record of opinionated design and creeping ads means they'll inevitably shove AI into everything too, while others counter that Apple has so far been the least aggressive about it and their current restraint actually feels like a feature. The conversation also veers into a broader frustration with Gmail's spam filtering getting worse, with a cynical take that the big email providers have built an oligopoly where they trust each other's mail but make it nearly impossible to self-host.
Adafruit receives demand letter from Fenwick legal counsel on behalf of Flux.ai [comments]
648 points · 267 comments · blog.adafruit.com · 22h ago
Adafruit got a demand letter from Fenwick & West on behalf of Flux.ai, a PCB design tool, after Adafruit apparently discovered a server misconfiguration that exposed Flux’s data and planned to publish on it. The HN thread quickly became a massive naming-confusion cleanup, with people clarifying this has nothing to do with Black Forest Labs’ image model, the f.lux screen warmth app, FluxCD for Kubernetes, or the Flux capacitor — someone even joked about Ben Afflux. The real meat of the discussion was people who’d actually used Flux.ai piling on to say the product is essentially a token-burning scam that ate $50–$140 with nothing useful to show, calling it “Software-as-a-Casino.” Others pointed out the Streisand effect is working perfectly, since nobody had heard of Flux.ai before and now the entire thread is reading it as a warning about an AI PCB tool that can’t place a component on a schematic.
Why Janet? (2023) [comments]
464 points · 244 comments · ianthehenry.com · 22h ago
The article is a detailed pitch for the Lisp-like language Janet, arguing it’s ideal for side projects, scripting, and CLI tools because of its tiny core, easy embedding, native-compiled executables under a megabyte, and a PEG-based text parsing system that replaces regex. Hacker News mostly agreed it was a well-written, refreshingly earnest argument, but the real energy went into the perennial Lisp syntax debate—some people found the brackets genuinely unreadable, others pointed out that editors handle it and that every language has terrible syntax if you don’t grow up with it. A few people dug into performance, noting it’s roughly as fast as non-JIT Lua and about 5–10× faster than Python on long-running tasks, and someone even demonstrated a working infix math macro to address the readability complaint. The thread also spun off into a broader tangent about whether any online community can survive the coming bot and AI onslaught, with WebAuthn hardware attestation and invite-only systems like Lobsters floated as partial solutions—but no one had a silver bullet.
MAI-Code-1-Flash [comments]
454 points · 201 comments · microsoft.ai · 13h ago
Microsoft launched MAI-Code-1-Flash, a 5-billion-active-parameter coding model optimized for real GitHub Copilot workflows, claiming it beats Claude Haiku 4.5 on every benchmark while using up to 60% fewer tokens. The HN crowd immediately zeroed in on the benchmark comparisons, with several people pointing out that Microsoft is comparing against Haiku 4.5 rather than the newer Opus models, making the "outperforms" claims feel carefully scoped—one person noted Haiku 4.5 is currently Anthropic's smallest model. A heated sub-thread argued over whether the model was actually trained on the SWE-Bench evaluation sets, with some citing the "hill climbing" announcement as evidence of potential contamination, while others defended the paper's decontamination procedures in appendix A.4. The lack of open weights drew real frustration, since Microsoft has released the Phi models openly for years, and multiple people argued that the 51% SWE-Bench Pro score isn't reliable enough for hands-off use—you still have to watch it like Tesla FSD, which defeats the purpose—while others countered that it's competitive with Claude Opus 4.6 on the same benchmark and the real play is pairing a small fast model for execution with a heavy reasoning model for planning.
A walking tour of surveillance infrastructure in Seattle (2020) [comments]
394 points · 270 comments · coveillance.org · 18h ago
The article is a detailed 2020 field guide mapping out specific surveillance technologies in downtown Seattle—cameras, Amazon Go stores, automated license plate readers, and fusion centers—explaining how each works and what social implications follow. The HN thread split sharply between people who found the guide surprisingly mild and those who thought it exaggerated real risks; several locals pushed back on the alarmism by pointing out that in Seattle the problem isn't too much surveillance but too little effective enforcement against visible public disorder. A technical subthread corrected the article's claim about Wi-Fi probe requests leaking past network lists, with multiple people noting that all major OSes stopped broadcasting preferred SSID lists by 2017 and now randomize MAC addresses by default. The most contentious debate circled around whether cameras inevitably re-define crime to match what cameras can catch, with one side arguing this algorithmic scoping erodes privacy and biases enforcement, while the other side countered that the real-world failures of Seattle policing make the privacy concerns feel academic.
Love systemd timers [comments]
371 points · 249 comments · blog.tjll.net · 22h ago
The article argues that systemd timers are a superior replacement for traditional cron jobs, offering clearer scheduling syntax, better logging via journald, and features like persistent timers that catch missed runs after downtime. The thread largely agrees that timers are technically better than cron, but it pivots hard into a familiar HN battleground: the ugliness of systemd's INI-style configuration syntax. A substantial chunk of the discussion is people arguing that the syntax is fine—compared to XML or YAML, it's clean and well-defined—against those who find it ugly or finicky, with a notable subthread revealing that many users didn't realize whitespace around the equals sign is actually allowed, leading to a minor "life changed" moment. The NixOS crowd chimes in heavily, saying defining systemd units in Nix configuration is far nicer than wrangling raw INI files, which sparks its own debate about whether Nix syntax is itself an improvement or just a different flavor of ugly.
CT scans of BYD car parts [comments]
355 points · 180 comments · www.lumafield.com · 11h ago
The article uses CT scans of four BYD components — a battery cell, window switch, portable charger, and key fob — to illustrate the company’s extreme vertical integration, from lithium mines to shipping, and argues this system is why BYD can sell cars for $15K and dominate global EV volume. Hacker News immediately zoomed in on the trade-offs: while everyone agrees BYD’s integration slashes costs and speeds iteration, a loud contingent pushed back hard on the repairability and longevity of these sealed, high-integration parts, pointing to EV clinics and forums where failed e-axles or coolant seals force whole-unit replacements at enormous expense. The comments split into two camps — one side argues this is just the automotive version of the phone industry’s “throw it away” model and that economics will win, while the other insists people actually want repairable cars, not cheap replacements, and that the EU is hypocritical for regulating phone batteries while ignoring far costlier vehicle right-to-repair issues. A tangential but heated subthread debated whether Tesla is truly a next-gen car company or just an EV-swapped Toyota with worse reliability and vanishing resale value, with several owners sharing specific, expensive failures.
1-Click GitHub Token Stealing via a VSCode Bug [comments]
333 points · 47 comments · blog.ammaraskar.com · 16h ago
A security researcher published a detailed writeup showing how a single click on a link to a Jupyter notebook on github.dev can silently steal your full GitHub OAuth token — not scoped to one repo, but granting read/write access to every repository you have access to. The attack chain works by exploiting VSCode's webview security model: the browser-based editor lets untrusted JavaScript in notebook outputs listen for and simulate keyboard events, which lets the attacker programmatically open the command palette, install a malicious extension that bypasses the trusted-publisher check by placing itself as a local workspace extension, and then exfiltrate the token. The HN discussion split between two main tensions: first, many people argued the root cause is that github.dev hands out a god-tier, per-user token rather than a per-repo scoped token like Codespaces uses, calling that an architectural "original sin"; second, the thread spent significant energy on the author's decision to go full public disclosure because Microsoft's MSRC team had previously silently patched his bugs without credit or acknowledgment. Several people with experience in bug bounty programs pointed out that MSRC's process is broken — developers often quietly fix reported vulnerabilities before the security team evaluates them, leading to reports being closed as "invalid" — while others argued the author should have given Microsoft more than an hour's notice before publishing the exploit.
Larry Ellison: "Citizens will be on their best behavior because we’re recording" [comments]
303 points · 235 comments · www.techradar.com · 14h ago
The article covers Larry Ellison’s September 2024 prediction that AI-powered surveillance—from street cameras to police bodycams—will make citizens behave because "we’re constantly recording and reporting everything." Hacker News immediately called out that Ellison wasn’t warning about this future; he was bragging about it, and that the real surveillance target isn’t crime but anyone who threatens people in power. The thread split into two angry camps: one side arguing this is straight-up authoritarianism where the rich exempt themselves while the rest live under a panopticon, and the other side—mostly pushed back against—pointing to China as a working example where surveillance delivers clean streets and public order. A recurring knife-twist was that engineers building these systems are making the choice to code this dystopia for a paycheck, drawing direct comparisons to "just following orders."
Apple rejected my dictation app for using the accessibility API [comments]
298 points · 162 comments · www.mitmllc.com · 20h ago
The article is a first-person account from a developer who built a Mac dictation app called WhisperPad to cope with a repetitive strain injury, only to have its update rejected by Apple for using the accessibility API to auto-paste text—a feature Apple deemed an unauthorized use of that API, even though earlier versions had been approved. The HN discussion immediately split into two camps: one arguing that Apple was being unreasonable and that this is the inevitable cost of a walled garden, with some suggesting the author should just switch to Linux, and the other pushing back hard that Apple’s restriction is a legitimate security measure against data exfiltration, with one person pointing out that the app is still fully functional via direct download outside the App Store. Several people jumped in to note that Google is similarly tightening access to its accessibility API, making this a cross-platform concern rather than just an Apple problem. A side debate erupted over whether suggesting Linux is a realistic alternative, with advocates claiming it “just works” for grandma and skeptics retorting that it’s a lifestyle change most people can’t afford, especially if they need specific software like Microsoft Office. The thread also surfaced a useful licensing workaround from another indie developer who lets App Store purchases unlock features in their direct-distribution version, though someone warned that could get you banned.
Use your Nvidia GPU's VRAM as swap space on Linux [comments]
276 points · 73 comments · github.com · 9h ago
This project repurposes unused VRAM on Nvidia laptops as swap space, using a userspace daemon that talks to the GPU via CUDA and serves a block device over the NBD protocol. HN generally found the idea clever for its specific niche—laptops with soldered RAM and an idle discrete GPU—but the performance numbers drew immediate technical pushback; the ~1.3 GB/s sequential throughput is far below the 64 GB/s PCIe bandwidth the card supports, and people pointed out that the NBD userspace path introduces heavy context-switching overhead that a direct kernel NVMe swap path avoids. The main debate shifted to whether this is actually useful: some argued it beats burning SSD write cycles, while others countered that modern SSDs have plenty of endurance and the real benefit is lower latency for sporadic page faults (the benchmark showed 335 µs average vs 9 ms for NVMe). A side conversation emerged about whether GPU RAM could ever be treated as system RAM, but the consensus was that it would be uncached and painfully slow from the CPU’s perspective. The thread also touched on why this approach exists at all—the obvious route using Nvidia’s P2P API is gated to Quadro cards, so the NBD hack is the only way to use consumer GeForce VRAM this way.
Stop Ruining It [comments]
270 points · 130 comments · seths.blog · 22h ago
Seth Godin’s post argues that customer delight, trust, and curiosity aren’t features you layer on—they’re what’s left when you stop ruining things through bad processes, marketing, or management. The thread took that concept and ran with it, largely agreeing on the principle but getting into a meaty debate about whether “not ruining it” is really just constant, active effort in disguise—someone pointed to Disney Parks’ obsessive detail work as a counterexample, though others argued that *is* the baseline. A big split emerged around advertising: plenty of people insisted most ad spend is wasted or actively creates negative brand association, backed by a reference to P&G cutting $200M in digital spend with no sales impact, while others pushed back that brand awareness ads still work for unknown products or for maintaining top-of-mind status. The conversation also took a sharp turn into how people are currently cashing in trust by switching to AI-generated content on LinkedIn and X, which destroys their reputation and turns a long-term asset into short-term slop, with one person tying this short-termism to a broader collapse in faith in the future.
Show HN: Eyeball [comments]
250 points · 78 comments · eyeball.rory.codes · 22h ago
The article is a minimalist browser game where you click on a line to guess where a given number falls within a range, then get scored on your accuracy. HN got immediately hooked, with people sharing near-perfect scores and calling it oddly addictive, but the thread quickly split into two camps: one side wanted more pressure (timed modes, moving targets, curved lines) while the other wanted less—specifically training modes to repeat missed guesses and build intuition. Several people pointed out the UI is confusing at first, needing clearer instructions that you click *on* the line itself, and others complained the low-contrast design makes it harder to see. A veteran developer spotted the project is clearly a riff on Matthias Wandel's classic "Eyeball" game, which tests geometric skills like angle bisection, not just linear guessing, and that kicked off a deeper conversation about what makes these games tick.
Three Ways to Get Paid (2018) [comments]
212 points · 138 comments · jasonzweig.com · 14h ago
Jason Zweig shared a three-part maxim from his father: you can get rich lying to people who want to be lied to, make a living telling truth to truth-seekers, or go broke telling truth to people who want lies. The HN thread immediately tried to extend the framework into a 2x2 matrix adding "lie to those who want truth" and questioned whether the third path necessarily means poverty—many pushed back, arguing plenty of people make good money with integrity and that the framing undersells honest work. A separate meta-argument erupted about the article itself being broken (the "read the rest" link goes nowhere), with some accusing the author of pulling a fast one, while others defended the brevity as refreshing in an era of LLM-padded content. Several people wryly admitted they wish they had fewer scruples, sparking a split between those romanticizing the sociopath's path to wealth and others insisting integrity actually builds sustainable community and income.
Coreutils for Windows [comments]
208 points · 220 comments · github.com · 15h ago
Microsoft shipped a preview of native Coreutils for Windows—Unix commands like `cat`, `cp`, and `ls` built as a single multi-call binary, maintained by Microsoft, based on the Rust `uutils` project. The thread immediately split: some people were genuinely excited this exists natively, but others dug into the mess of shell conflicts, noting that many commands clash with CMD or PowerShell built-ins and the documentation doesn't give clear guidance on how to actually get the right one to run. A big chunk of the discussion hammered on the "intentionally dropped" list—especially `uname`, `chmod`, and `dd`—arguing that leaving those out makes the whole thing half-baked, since you'd still need WSL or MSYS2 anyway, which defeats the point of frictionless script porting. There was also a sharp side-conversation about `shred` being dropped with the note "not particularly useful on Windows," which got kicked apart: shred never really worked on journaled filesystems or SSDs, so the omission is arguably honest, but people still wanted the tool. The underlying tension ran deeper—several people argued this is a GPL-avoidance move, since Microsoft chose the Rust reimplementation (MIT/Apache licensed) over GNU coreutils, and pointed out Ubuntu has done the same thing, framing it as an industry shift away from GPL code in foundational tooling.
Trump signs downsized AI order after weeks of reversals [comments]
205 points · 150 comments · www.politico.com · 15h ago
The linked article wasn't available to this summarizer; from the discussion, the piece covers a new executive order Trump signed that asks frontier AI labs to voluntarily submit powerful models for a 30-day government review before public release. The thread immediately splits into two camps: one side argues the order is a reasonable, bipartisan safety move that mirrors what Biden’s administration and the labs themselves originally wanted, creating a coordination framework and a regulatory moat for incumbents; the other side sees it as an expansion of the prior "Stop Woke AI" order, warning it gives the administration leverage to coerce companies into political censorship, propagandizing, or surveillance. Several people point out that executive orders aren't laws and can only command federal purchasing, not private companies, but others counter that the government is a massive customer and can effectively dictate model behavior through procurement—effectively making it a de facto regulation for anyone who wants to sell to the feds. A big technical angle emerges in the weeds: how exactly does a federal agency benchmark a model's "cyber capabilities" in 30 days using a classified process, and what stops a lab from gaming the evaluation the way VW rigged emissions tests?
The advertising cartel coming to your web browser [comments]
203 points · 57 comments · blog.zgp.org · 12h ago
The piece argues that Meta, Google, Apple, and Mozilla are collaborating on a built-in browser standard called "Attribution Level 1" that would let advertisers measure ad performance—framing it as a privacy-preserving feature while actually locking in Big Tech's advantage. HN pushed back hard on the author's premise, with several people noting the author is a VP at an adtech company whose real complaint is that the new system cuts him out, not that it harms users. A vocal contingent argued the proposal is strictly better than the current cookie-based tracking, since it uses differential privacy and aggregate reports rather than unique identifiers, and they pointed out that criticizing consent popups while demanding a per-site opt-in is disingenuous. Others countered that giving advertising a privileged seat in the browser itself is a fundamental corruption of what browsers should do, and that "anonymized" aggregate data is easily de-anonymized when combined with the hundreds of other tracking signals already in play.
My thoughts after using Clojure for about a month [comments]
201 points · 102 comments · www.acdw.net · 12h ago
The author wrote about their experience building a static site generator in Clojure after a month, praising its cohesive design and ergonomic data structures compared to Common Lisp and Scheme. The thread immediately veered into a debate about Clojure’s suitability for large-scale systems, with one camp arguing OOP is more intuitive for that scale, while others countered that Clojure’s immutability actually prevents exactly the kind of bugs that plague complex mutable codebases. A whole side discussion erupted around LLM code generation benchmarks, where Clojure consistently produces the best-performing solutions but has low success rates; several domain experts pushed back, saying hooking an LLM to Clojure’s REPL lets it test every piece before composing, making it dramatically more effective than static type feedback loops. Another thread dissected the pain of balancing mixed bracket types (parens, brackets, braces) without structural editing tools like paredit, with long-time users insisting it’s a solved problem. Someone also argued the real value isn’t syntax at all but the runtime—saying Java’s runtime still can’t match Erlang or Go for concurrency, while a reply shot back that pervasive immutability is itself a massive concurrency advantage that neither Go nor Java gives you.
AI outperforms law professors in Stanford Law study [comments]
199 points · 160 comments · law.stanford.edu · 8h ago
A Stanford Law study found that law professors preferred AI-generated answers to student questions over those written by other professors in 75% of blind comparisons, challenging assumptions about AI in fields without clear right-or-wrong answers. The HN thread immediately split into two camps: one side focused on the study's methodology, questioning whether the professors were simply wowed by the AI's polished prose and grammar rather than its legal reasoning, while others pointed to the risk of hallucinated case citations as a dealbreaker for real-world legal work. A strong contingent pushed back on the idea that this translates to replacing lawyers, arguing that the real danger is non-experts using AI for tasks where tacit knowledge—like knowing local judges' tendencies—can't be captured in training data. Thread regulars with legal experience kept pulling the discussion back to a core point: the AI is only as good as the human wielding it, and it's far more dangerous for an expert to atrophy their skills by over-relying on a 90%-accurate tool than for a layperson to dabble.
Morningstar values SpaceX at $780B, half its IPO target [comments]
198 points · 162 comments · www.reuters.com · 13h ago
The linked article wasn't available to this summarizer; from the discussion, Morningstar slapped a $780 billion valuation on SpaceX, about half of what the company reportedly hopes to command in a future IPO. The thread immediately split between people arguing that any valuation this high is divorced from fundamentals—pointing to revenue multiples and comparing the hype to Tesla's wild swings—and others insisting the real story is how SpaceX has rigged the game for itself. A major thread focused on how SpaceX has already secured fast-track index inclusion rules, meaning pension funds and 401(k)s that mechanically track the S&P 500 or Nasdaq-100 will be forced to buy in at whatever inflated price, essentially making the public the exit liquidity for early investors. Meanwhile, a dense sub-thread pushed back hard on the doomsaying about Kessler syndrome from Starlink, with orbital mechanics experts explaining that those low orbits self-clean in about five years and that debris-band scaremongering misunderstands how orbital decay actually works.
MAI-Thinking-1 [comments]
182 points · 76 comments · microsoft.ai · 13h ago
Microsoft just announced MAI-Thinking-1, a 35-billion-parameter active (1 trillion total) sparse MoE reasoning model they claim matches or beats Claude Sonnet 4.6 and Opus 4.6 on human preference and SWE benchmarks, trained from scratch on "clean, commercially licensed data" without distillation. The HN crowd was deeply skeptical of those bench numbers, arguing the model gets crushed by GLM-5.1 and Kimi K2.6 at similar sizes, and that the "no distillation" claim is mostly a hand-wavy flex since every other lab just distills from frontier models. The bigger fight was over whether "clean data" is even real—people pointed out Microsoft almost certainly scraped all of GitHub OSS (and possibly enterprise private repos), with the same old copyleft and attribution problems that no one has solved, making the ethical posturing ring hollow. The rest of the thread was almost uniformly about the website's scroll-jacking, which was so bad multiple people gave up reading entirely, despite the HN guidelines trying to police that complaint.
Preparing for KDE Plasma's Last X11-Supported Release [comments]
180 points · 217 comments · blog.davidedmundson.co.uk · 17h ago
KDE Plasma is dropping its X11 session entirely starting with version 6.8, removing the X11 login option and cleaning out all X11-specific code paths, though XWayland support will keep old apps running. The thread mostly agreed that KDE's Wayland transition has been remarkably smooth—many people reported switching and noticing zero difference or even lower latency and better frame pacing, with one person calling it "second-to-none" for X11 compatibility. A few dissenters pointed to real pain points: OBS recording latency spikes under Wayland, crashes on certain hardware during login (especially with Debian 13), missing backlight controls on old laptops, and the loss of foreign window embedding (e.g., embedding an mpv window inside another app). The biggest split is between users who say "it just works" and those who hit hard regressions they can't diagnose, but even the critics mostly conceded X11 itself is ancient cruft—the fight is over whether Wayland is ready *now* for every edge case, not whether the move itself is wrong.
Expanding Project Glasswing [comments]
173 points · 233 comments · www.anthropic.com · 18h ago
Anthropic is expanding Project Glasswing, a program that gives select organizations early access to its Mythos-class AI model to scan codebases for security vulnerabilities — the article reports 10,000+ high-severity flaws already found by the initial 50 partners. The Hacker News thread split sharply: some saw this as a brilliant marketing play that creates urgency and scarcity, cutting through enterprise procurement red tape, while others argued it's performative hype, noting that most targeted orgs already have Claude subscriptions and that the limited rollout risks letting competitors front-run them. A major thread debated whether even fixing every known vulnerability matters, given that social engineering attacks will soon be cheaply automated by any model — pushing the argument that authentication will have to become rigid and device-bound, which several people countered would create deeply uncaring systems that punish edge-case victims. There was pointed skepticism about Anthropic's "6 to 12 months" warning of catastrophic capability release, dismissed as the same timeline they've been recycling for years, and a technical debate over whether Mythos actually outperforms GPT-5.5-Cyber (which is also unreleased), with references to the UK's AISI benchmarks showing both models solve the same full benchmark but Mythos does it more consistently and with fewer tokens. A few commenters raised the irreproducibility of AI-generated patches and the geopolitical concern that access is concentrated in a handful of wealthy countries, reinforcing fears of a global AI underclass.
Agentic Mfw [comments]
171 points · 53 comments · agenticmotherfucking.website · 5h ago
The site is a profane, satirical manifesto written by Claude, riffing on the original "motherfucking website" to argue that clean code and maintainability are dead—replaced by vibe-coded, agentic slop where the burn rate of tokens is the pitch and complexity is the valuation. HN was split between people who found genuinely killer lines in the LLM-generated parody ("Light bends around it") and those who were exhausted by the hyperbole and profanity, especially once they realized an AI wrote it. Some commenters clocked the "Claude-ness" leaking through the tone and argued the whole thing was a meta-joke about LLMs writing slop rants about LLMs writing slop, while others pushed back hard on the "static sites are for people who can still read" line as fetishizing intelligence. The thread also veered into genuine lament about the state of SaaS, VC money loops, and whether anyone is actually reading anything anymore—with a few people wondering if the sheer commitment to the bit made it cathartic despite the AI authorship.
Fidonet: Technology, Use, Tools, and History (1993) [comments]
162 points · 66 comments · www.fidonet.org · 18h ago
This is a nostalgic dive into Randy Bush's 1993 technical paper on FidoNet, the pre-internet store-and-forward email network that ran over dial-up modems and connected 20,000+ hobbyist BBS nodes worldwide using MS-DOS, xmodem/zmodem transfers, and a hierarchical addressing scheme like 1:105/6. The HN crowd, many of whom were sysops or points back in the day, reacted with a flood of remembered addresses (2:463/1161, 1:153/7015, 5:7211/1.27 from Zimbabwe), which quickly turned into a debate over who had to "prove" themselves to get a node versus just being a point or a regular user. A strong counter-current emerged pushing back on the idea that FidoNet was an exclusive club—several people pointed out they were 14-year-old kids who walked in, got given an address for free, and ended up drinking with the sysops at user group parties. The thread also spun off into a detailed appreciation for Zmodem's simultaneous two-way transfers, a developer surfacing his retro BlueWave-compatible offline reader Wolverine, and a reminder that FidoNet and its alt-nets like fsxNet are still alive today.
HP re-releases classic computer science calculator: The HP-16C [comments]
157 points · 103 comments · hpcalcs.com · 13h ago
HP is re-releasing the HP-16C, the classic programmer’s calculator, as a Collector’s Edition after more than 35 years, preserving its RPN logic and bit-twiddling features while adding speed improvements and program save/load. The thread immediately lit up with nostalgia and practicality—people who still have their original 1980s 16C or 15C chimed in to say they’re still in daily use, while others pointed out this isn’t actually HP itself anymore, just an official licensee (with some ex-HP calculator staff involved). A major split emerged between those enthusiastic about the reissue and fans of SwissMicros, who argued the DM16L or DM42n are better-built alternatives, especially given complaints that the 15C Collector’s Edition from 2023 had worse key feel and display quality than the originals. Several commenters debated whether emulating vintage firmware on an ARM chip is wasteful versus pragmatic, and a few grumbled that for the same price you can find an original 16C on eBay, though others noted used prices have shot past $500.
CSS-Native Parallax Effect [comments]
135 points · 53 comments · dan-webnotes.com · 21h ago
A developer posted a clean implementation of a CSS-native parallax effect using the newer Scroll-driven animation timelines API, wrapping the whole thing in a single utility class with one adjustable variable. The thread immediately split into two camps: a large chunk of people who couldn't see the effect working at all—it's gated behind a flag in Firefox, only works in mobile Safari or Chrome for some, and several demo links were broken or flickering across Vivaldi, Edge, and desktop Chrome—and a separate group raising the classic rebuttal that CSS perspective with translateZ has done GPU-accelerated parallax for years with much wider browser support. The author responded directly to that comparison, arguing the new method is easier to set up and can control entry/exit ranges plus other properties like opacity, but acknowledged they aren't using it in production yet. A few commenters called out that the article itself had no interactive demo, which frustrated people looking to test it, while others appreciated the built-in `prefers-reduced-motion` support and the self-tuning scale/translate variable. Under the hood, the real tension was between developers who want this API shipped stable already and those who find the effect disorienting or wasteful, with one person noting Firefox was first to implement it behind a flag and is now the only major browser without it in stable.
Michael Burry says neither SpaceX nor Anthropic is worth $1T [comments]
128 points · 149 comments · www.businessinsider.com · 21h ago
Michael Burry, the “Big Short” investor, publicly stated that SpaceX and Anthropic aren’t worth their reported trillion-dollar IPO targets, calling the AI buildout a “false demand signal” driven by hype. The discussion largely split between people dismissing Burry as a one-hit wonder who’s been wrong plenty since 2008, and those arguing he’s making a solid, conservative point about valuation—namely that having the latest AI model isn’t a moat, and Anthropic’s massive revenue numbers are mostly unverified extrapolations. Several people dug into the idea that the broader market has been permanently broken by zero-interest-rate policies and meme-stock logic, making fundamentals irrelevant until the music stops. A notable counter-argument was that even if Burry is right long-term, he’s famously early and can stay solvent only so long against hype-driven momentum.
Open Repair Data Standard [comments]
125 points · 4 comments · openrepair.org · 12h ago
The Open Repair Data Standard (ORDS) is a shared specification for collecting and combining data from community electronics repair events, aiming to identify global trends in what breaks and what fixes work. The thread is extremely quiet for a 125-point submission, with only a handful of comments that mostly poke at the standard's limitations—one person notes the product data fields feel "anemic," especially the removal of the "model" field due to quality issues. Another commenter asks flatly if the project is abandoned, pointing to no major changes since roughly 2024. The actual spec link is buried in a correction, and there's no meaningful pushback or defense from the project side, so the overall vibe is that HN thinks the idea is worthwhile but the standard seems stalled and underspecified for real-world use.
Generated 2026-06-03 08:35 UTC
Generated by Sauron from Hacker News discussions and linked articles.