HN Brief: 2026-06-17
Today's front page was dominated by a single, sprawling question: are the tools actually good enough now, or is the whole thing a bubble about to pop? A $60 billion SpaceX acquisition of a coding assistant sparked furious debate over whether AI tools have any real moat, while the news that Meta is forcing its best engineers to label training data was read as either the most cynical soft-layoff ever or a sign that the company sees its own talent as expendable mules. A third thread, on whether running local models is finally practical, turned into a raw argument about global economics—half the world can't afford the hardware—and the psychological value of owning a tool versus renting one from a cloud provider you don't trust.
The threads most worth clicking: "Running local models is good now" because the cost debate buried inside it is the honest conversation the industry keeps avoiding; "SpaceX to buy Cursor for $60B" for the sheer, unvarnished referendum on whether AI coding has any durable value at all; "Is Meta destroying its engineering organization?" for the cold, inside-Meta accounts of what Zuck actually told his engineers; "Apple's weird anti-nausea dots cured my car sickness" for the genuine surprise and the sensor-fusion debate it kicked off; and "Stop Killing Games fails to secure EU law" for the sobering reality check on what happens when a million signatures meet a room full of industry lobbyists.
Running local models is good now [comments]
1259 points · 495 comments · vickiboykis.com · 17h ago
The linked article argues that running local AI models has become genuinely practical, thanks to better models like Qwen 3 MOE, faster inference via llama.cpp and Metal on M2 Macs, and mature tooling like Open WebUI and Continue.dev—claiming a $2,000 used Mac can now handle 80% of daily tasks, making cloud subscriptions unnecessary. The thread immediately pivoted to a sharp cost debate: while the author frames the hardware as a reasonable $2k investment, many pushed back that this ignores global reality, noting that half the world’s population could never afford that, and that for professionals, paying $20/month for a better cloud model is often cheaper than amortizing a depreciating asset. Others countered that local models offer privacy, control, and freedom from the whims of commercial model providers, and that the psychological value of owning your own tools is worth the premium—though a significant faction argued that for hard reasoning tasks, the cloud still beats local hands-down. The conversation also took a tangent into whether companies will start running on-prem AI clusters, with some saying the trend is toward outsourcing everything, while others pointed out that local models don’t require the same infrastructure headaches as self-hosting servers, making the comparison to cloud compute a false one.
SpaceX to buy Cursor for $60B [comments]
1002 points · 1502 comments · www.reuters.com · 21h ago
The linked article wasn't available to this summarizer; from the discussion, the splash is that SpaceX is buying the AI coding tool Cursor for $60 billion in an all-stock deal, using shares it just raised publicly. The big split in the thread is over valuation—some argue Cursor's $4 billion in annual revenue makes this a 15x price-to-sales bet on insane future growth, while others counter that revenue is a garbage metric when the tool is just a thin wrapper on an open-source editor and faces a commoditized market with zero moat. A vocal faction is dead set that the acquisition will tank Cursor's goodwill because it’s now tied to Musk and Grok, which they argue is a worse coding model tainted by forced ideological fine-tuning, not just a technical underdog. The thread largely derailed into a heated referendum on whether AI coding assistants are even worth $60B, with many pointing out you can switch to a dozen open-source or YC-backed alternatives overnight, and a handful of people openly calling the whole thing a bubble built on monopoly paper.
Apple's weird anti-nausea dots cured my car sickness [comments]
732 points · 227 comments · www.theverge.com · 15h ago
The Verge's Thomas Ricker wrote up Apple's "Vehicle Motion Cues" feature—animated dots around the screen that shift with the car's movement—and described how it let him read and type in a moving camper van for the first time. The thread immediately split into two camps: those who swear the feature works like magic (including one user's wife who went from instant nausea to reading for hours) and those who say it did nothing for them after a genuine try, even on trains and buses. A deeper debate emerged about the underlying mechanics, with several commenters arguing Apple's implementation relies on accelerometer-based inertia modeling while the Android alternatives use gyroscopes to simulate a horizon—and one person pointed out that's physically impossible as a pure accelerometer approach, meaning Apple must be doing full sensor fusion. The conversation also circled around the practical challenge of building a hardware equivalent for paper books, and a few people noted the feature is hidden in Accessibility when it should really be a front-line toggle.
GrapheneOS has been ported to Android 17 [comments]
700 points · 305 comments · discuss.grapheneos.org · 11h ago
GrapheneOS, the hardened privacy-focused Android fork, has been fully ported to Android 17 and its developers are pushing out a public release within the next day or two. The thread quickly split into two distinct camps: one side was deeply appreciative of the team's speed and professionalism, while a vocal group of current users jumped in to share their real-world experience with the switch from stock Android. The main friction point they raised had nothing to do with security—it was that GrapheneOS ships the barebones AOSP keyboard and messaging app, which means you lose features like spacebar cursor-scrolling and proper inline RCS emoji reactions, though the consensus was that both problems are easily solved by swapping in Gboard or FUTO's keyboard and granting them network access. A deeper thread spiraled into a full-on indictment of Google's RCS and SMS ecosystem, with multiple people arguing that RCS is an "embrace, extend, extinguish" power grab that locks out third-party apps and forces you into Google's walled garden, and that anyone serious about privacy should just move to Signal.
Mechanical Watch (2022) [comments]
683 points · 116 comments · ciechanow.ski · 20h ago
The linked article is Bartosz Ciechanowski’s stunning interactive explainer of mechanical watch internals, from the mainspring to the escapement—widely considered one of the best technical deep-dives on the web. This was a re-post of a 2022 piece, and the thread immediately split into two camps: one side thrilled to see it again, the other mildly disappointed it wasn’t new, with a few pointing to subsequent reposts and wondering what Ciechanowski has been up to since. The real conversation took a sharp turn into the economics and culture of watch ownership, where people pushed back hard on the idea that expensive watches are the interesting ones—arguing instead that the real appeal is the mechanical engineering, not the cost, and that cheap, repairable vintage pieces and affordable automatics like the Tissot PRX are where the real community lives. There was a strong consensus that the watch repair scene is brutal, with several people sharing stories of broken sentimental pieces that nobody will touch, and one recommending a YouTube channel like Wristwatch Revival for old watches most shops refuse to handle.
Feds freaked over Fable 5 after 'fix this code', not jailbreak, say researchers [comments]
567 points · 333 comments · www.theregister.com · 22h ago
The article reports that the Trump administration's move to block Anthropic's advanced models was triggered by a surprisingly simple "Fix this code" prompt, not a sophisticated jailbreak, according to bug bounty pioneer Katie Moussouris. The HN crowd was deeply split: some argued the whole thing was a political pretext to screw with Anthropic rather than genuine security concern, while others pointed out a fundamental contradiction—Anthropic spent months hyping these models as incredibly dangerous, then released them with security that obviously isn't bulletproof against an LLM's infinite attack surface, which made the federal freakout look predictable. A strong contingent pushed back that the real issue isn't the prompt but that you can't discern intent from an LLM, so any "safety" system is a farce, and the government's solution of ID-gating access to frontier models is laughably easy to bypass with stolen identities and dark web credentials. The consensus hardened around the idea that pulling the best tools from defenders while adversaries are grinding through prompts is a farcical move, since the black hats will get their answer one way or another.
Is Meta destroying its engineering organization? [comments]
564 points · 491 comments · newsletter.pragmaticengineer.com · 15h ago
The linked article wasn't available to this summarizer; from the discussion, it's about an alleged plan at Meta to forcibly reassign 30-50% of core engineers to data-labeling and RLHF work, effectively using them to train the AI that will replace them. A major split emerged: many found the premise ridiculous, arguing that using expensive Silicon Valley engineers for grunt labeling work is a waste of talent and money, but people claiming to be inside Meta or connected to it pushed back hard, saying the story is real and that Zuck himself told them in a town hall their remaining value to him is as training mules for his AI. The thread went deep on the cynicism of the move—that forcing engineers to label data means they’ll either sabotage the models or feed garbage—and drew parallels to *Severance*, the bootstrapped startup Mercor, and the old "hot dog" meme, while the consensus was that this is either the most efficient way to lose your best engineers or a soft layoff dressed up as "useful" work.
Apple is about to make Hide My Email useless [comments]
479 points · 293 comments · arseniyshestakov.com · 13h ago
A developer named Arseniy Shestakov posted that Apple is changing its iCloud+ "Hide My Email" and "Sign in with Apple" feature to issue aliases on the new `@private.icloud.com` subdomain instead of the existing `@icloud.com` domain, making it trivially easy for any website to block all those relay addresses with zero false positives. The HN consensus is that this move fundamentally breaks the privacy feature's core value—plausible deniability—since services can now simply ban the entire `private.icloud.com` subdomain, just like they already do with temporary mail services like `anonaddy` or `10minutemail`. A lot of people pushed back on the article's alarmism, pointing out that Apple's old `@icloud.com` aliases already had pretty weird-looking patterns that services could have been blocking for years, so this change just formalizes what was already possible. The real split in the thread is between those who think this is Apple actively caving to anti-privacy pressure and those who suspect the company is just scared of its own `@icloud.com` domain getting blacklisted as spam, since the aliases are now increasingly used for outbound, not just inbound, mail.
U.S. pulling ocean sensors a 'shock' for Canadian research as El Niño nears [comments]
416 points · 254 comments · www.timescolonist.com · 12h ago
The article reports that the U.S. is pulling a network of over 900 ocean sensors off the coasts of Oregon, Washington, Alaska, North Carolina, and Greenland, a move that blindsides Canadian researchers just as an El Niño pattern is developing. The thread quickly zeroed in on the fact that this isn't just defunding science—it's actively spending millions to dismantle a system that could have simply been left in place, which a bunch of people argue is intentional sabotage to make it harder for a future administration to restart monitoring. The consensus is that this has nothing to do with cost savings; several people point to Project 2025 as the playbook, and there's a split where some see it as an illegal act Congress could stop but won't, while others take the darker view that the majority party is quietly onboard. A few commenters got into a heated, derailing argument about whether the U.S. is about to pay Iran $300 billion in reparations from a war that supposedly didn't happen, but the main takeaway is that the administration is spending money to destroy data that would embarrass its donors in the fossil fuel industry.
Calvin and Hobbes and the price of integrity [comments]
399 points · 173 comments · therepublicofletters.substack.com · 16h ago
The Substack piece tells the story of Bill Watterson’s decision to end *Calvin and Hobbes* at the height of its fame rather than sell out—opening with a dorm-room fresco he whitewashed over, and closing with him turning down Steven Spielberg’s movie call and burning a box of plush toys to kill any licensing deal. The thread mostly lands on Watterson’s integrity as the real point, with the top comments arguing that his refusal to merchandise is precisely *why* the strip still hits as hard today, while a few push back that he could have let the strip evolve without ruining it, or that the “pissing Calvin” bumper stickers are a bitter legacy of that choice. A split emerges over whether the modern world has room for that kind of purity: some say the internet would have wrecked his creative control by forcing him to build a website, while others counter that the syndicate was his only real bottleneck, and the same pressures—algorithms, platforms, “more”—would have eventually ground him down. A handful of people point out that Watterson’s own stated reasons for stopping, “I’d rather stop than run out of ideas,” are the same logic that kept *Cowboy Bebop* one season and *Firefly* one, and that the thread’s real takeaway is how rare it is for someone to simply choose done over infinite.
TIL: You can make HTTP requests without curl using Bash /dev/TCP [comments]
390 points · 183 comments · mareksuppa.com · 15h ago
The article shows how to make quick HTTP requests using bash's built-in `/dev/tcp` redirection, useful for checking connectivity in minimal Docker containers that lack curl or wget. The Hacker News crowd immediately split into two camps: one group loved this as a clever debugging trick for stripped-down images, while the other pushed back hard that this is not a real HTTP client—it can't handle TLS, redirects, chunked encoding, or anything else curl does quietly, and the author himself conceded it's only for quick checks. A long subthread erupted over whether running distro-less "FROM scratch" images actually reduces your real attack surface or just keeps security scanners off your back, with the pragmatic side arguing that Alpine adds only 10MB and saves you from having to spin up debug containers every time something breaks. Someone also pointed out that if the container already runs Python, you can just use `urllib` or `socket` directly, which is cleaner than hand-rolling HTTP over a bash file descriptor, and the whole thing got people worried about how easily they're starting to recognize the "acknowledge mistake, thank for pushback" cadence as AI-pilled writing.
Stop Using JWTs [comments]
363 points · 209 comments · gist.github.com · 15h ago
A developer posted a strongly-worded manifesto arguing that people should stop using JWTs for browser-based user sessions, because they're insecure, poorly specified, and better replaced with regular cookie-based sessions. The HN thread immediately split into two camps: one side agreed that JWTs are a terrible choice for user login, pointing to the history of JWT libraries with "alg=none" vulnerabilities and the fundamental problem that you can't revoke a stateless token without maintaining a revocation list anyway. The other side pushed back hard, arguing that the article overstates the case — JWTs are perfectly fine for service-to-service communication and have been hardened over time, and one commenter pointed out that AWS's own identity services rely on them for machine-to-machine auth. There was also a lively tangent about Macaroons as an alternative, with the Fly.io team chiming in that they run one of the world's largest Macaroon deployments.
Correlated randomness in Slay the Spire 2 [comments]
295 points · 87 comments · tck.mn · 22h ago
A blog post dives deep into how Slay the Spire 2 has correlated random number generators, revealing that seemingly random events like what curse you get or what potion drops are actually deeply predictable based on other early-game RNG calls. The HN crowd immediately zeroed in on the fact that this is a direct consequence of the developers using C#'s `System.Random` with seeded instances, which has a famously linear output, instead of a better algorithm — a decision that locks them into this weird behavior forever because changing the seedable `Random` class would break every seeded run. There's a split in the comments: some see this as a fixable bug that trivially should have been caught, while others argue the developers deliberately chose this design to preserve seed stability so that two players making the same choices get the same rewards, even if it means the "randomness" is technically exploitable. A few people pointed out the irony that Godot's own scripting language uses a better RNG (PCG32) that wouldn't have this problem, but the game was built on C# anyway.
Has AI already killed self-help nonfiction books? [comments]
275 points · 296 comments · tim.blog · 14h ago
Tim Ferriss published a detailed postmortem of his own book sales, showing that his self-help catalog—things like _The 4-Hour Workweek_ and _The 4-Hour Body_—has dropped roughly 80% since 2022, with the decline accelerating sharply right as LLMs took off. The thread largely agreed with his premise that prescriptive nonfiction is being eaten alive by chatbots, but a fair chunk of the pushback was about whether those books ever held much signal in the first place, with several people calling out specific titles like _Rich Dad Poor Dad_ and _The Secret_ as outright fiction or scams. Others pointed out that his data only covers print, and that audio books and YouTube have continued to grow, so the real story might be a format shift rather than a collapse of the ideas themselves. A long aside debated why people still prefer video over text for tutorials, with one camp arguing it's because a large chunk of the population is functionally illiterate and another saying it's just easier to monetize on YouTube. The split essentially came down to whether the "transformation" he's selling is genuinely irreplaceable, or whether it was always just information dressed up with a high enough price tag to feel legitimate.
Google Chrome update will close the door on ad blockers [comments]
273 points · 345 comments · 9to5google.com · 18h ago
Google Chrome is finally killing off Manifest V2 extensions, which means the end of popular ad blockers like uBlock Origin that relied on a lingering loophole. The HN crowd is split—some argue that effective MV3-based blockers like uBlock Origin Lite already exist and don't sacrifice features most people need, while others call the headline clickbait and insist Google's move is a slow-motion monopoly play to cripple ad blocking. A big part of the thread pushes back on the idea that this is the end of ad blocking entirely: Brave has built MV2 support for key extensions directly into its engine, and DNS-level tools like Pi-hole still work independently of Chrome's changes. The main pushback from the domain-knowledgeable side is that Firefox remains the only real long-term option, though some concede its UI feels slightly less snappy and note that Mozilla's own spending priorities don't inspire confidence. A few commenters point out that Google engineers describe the MV2 code as buggy tech debt, but others aren't buying that justification from a trillion-dollar company that's spent years deliberately shutting down the same ecosystem.
But yak shaving is fun (2019) [comments]
253 points · 72 comments · parksb.github.io · 17h ago
The article is a defense of "yak shaving"—the tendency to build things from scratch in a cascading series of sub-projects—and argues it’s actually fun and valuable, using examples like Donald Knuth inventing TeX just to write a book. The thread immediately split into two camps: people who enjoy the journey of building and learning, versus those who just want a working result as fast as possible. A lot of the pushback came from people who said the real split isn't about enjoyment, but about "slop tolerance" and how much time you have—with AI now let users choose which projects to lovingly handcraft and which to speed through with code generation. One recurring tangent was that the article’s image is actually a Highland cow, not a yak, and there was a long, meta side-discussion about how the term "yak shaving" itself has drifted to sometimes get used as a weapon to shame engineers for not just using existing frameworks.
Humiliating IIS servers for fun and jail time [comments]
219 points · 53 comments · mll.sh · 9h ago
The linked article is a detailed walkthrough of techniques for finding and exploiting vulnerabilities in Microsoft IIS servers, covering Shodan queries, Google dorking, tilde enumeration, and web.config path traversal. The comments immediately split into two camps: one celebrating the guide as a practical recon resource, and another pushing back hard on the "for fun" framing, arguing that publishing these tricks lowers the barrier for actual harm against legacy systems that keep critical infrastructure running. Several people dive into the technical weeds, confirming that IIS tilde enumeration still works on C: drives by default and noting that Microsoft itself recently created `C:\inetpub` on non-server Windows machines, which one commenter traced back to a 2004 Old New Thing post about reusing old ideas. A vocal minority dismisses the guide as LLM-generated or just "script kiddie stuff from the early 2000s," while others counter that the article has high signal-to-noise and explicitly calls out the "recon harder" takeaway. The overall vibe is that IIS is still everywhere—in banks, local governments, and the Navy—so the community is split between "this is a fun bug bounty compendium" and "please don't make it easier for randos to wreck stuff civilization depends on."
GPT‑NL: a sovereign language model for the Netherlands [comments]
200 points · 184 comments · www.tno.nl · 14h ago
The submission is about GPT‑NL, a publicly funded Dutch-language model built by TNO, SURF, and the Netherlands Forensic Institute to provide a sovereign alternative to big‑tech AI, with strict governance and transparency. The HN crowd split hard over whether this is a real product or just a years‑old proposal that hasn’t shipped, with some calling it “ridiculous” that two and a half years later it’s still not complete. A major camp argued the whole sovereignty framing is marketing fluff — since the model’s weights are under a controlled license, it’s sovereign for the government and data owners, not for the public, and a rival thread dismissed it as “in‑house” AI dressed up as a national project. Others pushed back hard, saying TNO is a semi‑DARPA applied‑research outfit that routinely “wastes” money on R&D nobody hears about, and that targeting a niche audience of Dutch‑speaking public‑sector clients with specific privacy and compliance needs is a legitimate, resource‑constrained start — not fraud, just boring. The most heated tangent devolved into a sprawling argument about whether Europe actually builds any computer hardware, with people flexing ASML and NXP as counterexamples, while others fired back that lithography machines aren’t chips and the continent still can’t fabricate its own AI silicon.
Leaked OpenAI financials show $38.5B loss and compute burn [comments]
198 points · 212 comments · runtimewire.com · 5h ago
A leaked set of OpenAI financials shows a $38.5 billion net loss for 2025, driven by $34 billion in total costs against $13 billion in revenue, with $19 billion going to R&D and nearly $6 billion to sales and marketing. The thread immediately split on whether a company burning cash that fast should even be allowed to IPO—some argued public markets can decide for themselves, citing Uber’s eventual profitability, while others pointed out that index funds would be forced to buy in, leaving 401(k)s holding the bag. A lot of pushback centered on the headline loss being inflated by a non-recurring $30 billion accounting charge tied to OpenAI’s for-profit conversion, making the operating loss a more meaningful $20.9 billion. The $5.7 billion sales-and-marketing line got heavy skepticism, with guesses ranging from bribes and subsidies to massive discounts on enterprise deals, and one commenter noted that if you strip out R&D and S&M, the business would actually have decent margins—but then it wouldn’t stay ahead of competitors. A side tangent dismissed Anthropic’s Claude as misused for coding tasks that a simple parser could handle, arguing its real value should be in homes and classrooms, not writing `tail -n5`.
'Ghost jobs' could soon be illegal in New York [comments]
195 points · 120 comments · www.fastcompany.com · 15h ago
A New York state bill would make it illegal for companies to post "ghost jobs"—listings for positions they don't actually intend to fill—with fines starting at $2,500 per 30 days. The discussion immediately split on whether the law would actually work: some argued companies will just keep pushing the "fill by" date back or treat the fine as a cost of doing business, while others countered that requiring a public, continuously updated "fill by" date creates a visible paper trail that turns the practice from an invisible crime into something candidates can actually point to and call out. A significant chunk of the thread argued the real problem isn't bad-faith employers but legitimate ones—for instance, companies legally required to post jobs publicly for H1B visa purposes even when they've already picked an internal candidate, or firms that need to keep a listing open because they're waiting on a contract that might fall through. The broader consensus was that even a flawed first attempt is worth it: let companies try to skirt the rules, then close the loopholes immediately, rather than waiting years for a perfect bill.
Stop Killing Games fails to secure EU law despite 1.3M signatures [comments]
192 points · 76 comments · www.dexerto.com · 6h ago
The linked article covers the European Commission's formal rejection of the "Stop Killing Games" petition, which gathered over 1.3 million signatures but only triggered a mandatory review, not a binding legislative process. The discussion on HN quickly split between people who saw this as a predictable failure of the EU's citizen initiative system and those who argued the campaign actually succeeded in shifting the fight to the European Parliament, where supporters are now trying to amend the upcoming Digital Fairness Act. Several commenters pointed out that the Commission spent most of its time with industry lobbyists rather than the petitioners, while others pushed back hard on the technical feasibility of the demand, with one game developer arguing that forcing publishers to release self-hosted server binaries after shutdown would be a massive rewrite for modern games built on third-party cloud services like DynamoDB and Kafka. A more cynical take emerged that this defeat was inevitable given the industry's financial power, though a few people countered by noting that Ubisoft's recent financial troubles suggest consumer boycotts might already be working better than any EU law ever could.
Claude: Elevated errors across many models [resolved] [comments]
184 points · 154 comments · status.claude.com · 14h ago
The upstream was a service outage at Anthropic — a roughly two-hour window where Sonnet and Opus models saw a ~10% error rate, with follow-on issues hitting Haiku and the newer Opus 4.8 as well. The thread immediately split into two camps: one side saw the outage as a routine inconvenience and wanted better retry logic baked into the API clients, while the other side dove into a much longer argument about whether Anthropic's own dogfooding of AI coding tools proves the "coding is solved" claim is hollow, given their own infrastructure is buggy and their uptime stats (under 99.5%) trail OpenAI's. A recurring sub-thread zeroed in on Claude Code's terminal rendering problems — specifically in tmux on Linux and in Windows Terminal — with some people swearing it’s fine and others reporting it’s been broken for months, and one person claimed they got it fixed by having Claude itself build a test harness that used tmux as an oracle. The outage also sparked the usual meta-jokes about what happens when the dogfooding AI goes down, with references to a "breakglass ChatGPT subscription" and a "meta-Claude that claudes Claude when Claude is down."
I've always wondered if anyone used sharing buttons on news sites and blogs [comments]
178 points · 98 comments · derekhanson.blog · 11h ago
The article digs into data from GOV.UK and others showing that social sharing buttons get clicked only about 0.2% of the time—roughly 1 in 476 visitors—with most people just copying and pasting URLs instead. A big chunk of the HN discussion actually pushed back at the author, arguing that 0.2% is a perfectly respectable conversion rate for a government site, and that calling it “nobody” is misleading; one person pointed out that MrBeast’s like-to-view ratio is under 5%, so by that logic “basically no one” likes his videos. Others chimed in that the real problem isn’t low usage but the tracking and bloat these buttons introduce, with several commenters advocating for the native `navigator.share()` API or a simple “Copy link” button as a cleaner alternative. The thread also split on whether the buttons ever make sense—some insisted they’re cargo-cult behavior that leaks data, while others recalled cases like Wordle’s share feature that actually drove viral growth, and noted that major platforms keep them for a reason.
Commodore Releases Flip Phone [comments]
174 points · 105 comments · commodore.net · 22h ago
The submission is a long, manifesto-style pitch from the CEO of the revived Commodore brand for a new flip phone called the Callback, which runs Sailfish OS and blocks social media and web browsers at the system level while still supporting WhatsApp, Signal, Maps, and a Commodore 64 emulator. The HN thread is sharply split, with one vocal camp dismissing it as a cynical, overpriced nostalgia play that costs $500 for what amounts to a glorified dumb phone with arbitrary parental controls, while another camp argues the exact opposite—that this is the first device that actually understands their desire to ditch smartphones without losing essential messaging and navigation. A lot of the pushback centers on the price, with many noting you can get a rugged, military-grade flip phone with better specs for under $200, and several commenters point out that the real competition isn't other flip phones but the $90 Android phone that already runs WhatsApp. There's also a minor, unexpected tangent where the thread debates whether blocking social media at the OS level is genuinely helpful or just insulting to adults who should be able to exercise self-control, with the CEO’s own "Obi-Wan" quote about absolutes getting some playful pushback.
Qwen-Robot Suite: A Foundation Model Suite for Physical World Intelligence [comments]
168 points · 28 comments · qwen.ai · 18h ago
The linked article wasn't available to this summarizer; from the discussion, Qwen announced a new "Robot Suite" foundation model meant to give LLMs physical world intelligence and control over hardware, like a general-purpose robot brain. The HN crowd immediately couldn't load the page because of obnoxious JavaScript, but once past that, the real debate was whether the robotics industry should care or if this is just another research toy — someone building a snow-clearing robot dug into the architecture and laid out why it's not ready for prime time, pointing out the models need a safety layer, are too expensive for edge hardware, and haven't released weights or training scripts. A few commenters argued the total addressable market for physical robots dwarfs coding or services, and that this kind of integrated system could produce simple products next year and production in a few years, while others dismissed the whole thing as a solution in search of a problem and pointed to existing snow plow competition hardware that already works. The split came down to whether you buy that an LLM agent loop is the right way to orchestrate a robot, or if you think "traditional" robotics R&D with purpose-built systems already has this beat.
Wolfram Language and Mathematica version 15 [comments]
154 points · 69 comments · writings.stephenwolfram.com · 8h ago
Version 15 of Wolfram Language and Mathematica just shipped, adding a built-in AI assistant, major time-series upgrades, symbolic music, and a unified model-fitting framework—7,000+ primitives deep after 38 years. The thread immediately split on the walled-garden problem: heavy pushback that the high cost and closed source make Mathematica a non-starter for enterprise and academic reproducibility, with Python, Julia, SageMath, and even cracked copies coming up as the workaround. Several people who use it professionally in physics, finance, and engineering defended the productivity gains—especially for small teams—but even fans admitted the new Wolfram AI assistant is embarrassingly bad compared to Claude or Gemini, and that charging for it was borderline criminal. A separate vein went after Stephen Wolfram himself, branding him a crank on par with Eric Weinstein, though others argued his independent research is harmless and worth having around.
I Fired Google [comments]
148 points · 127 comments · www.theartofdoingstuff.com · 17h ago
The article is a personal rant about replacing a Google Home with an Alexa after a Gemini update turned simple queries into long-winded, useless lectures—asking how old Geena Davis is just stopped working. The HN thread overwhelmingly agreed with the author, piling on examples of Gemini getting verbose, mishearing commands, and refusing to shut up, with many noting the same regression in Android Auto. Several people pointed out that Alexa is just as bad with its own "by the way" spam and ads, so jumping to Amazon is swapping one enshittified product for another. A strong contingent argued these assistants peaked a decade ago at timers, music, and weather, and that every "improvement" makes the core experience worse—one long comment described giving up entirely after discovering there’s no documentation to become a power user, calling the entire category an "utterly unserious product." A minor split formed over the article itself, with some dismissing it as breathless or complaining the site is unusable without an ad blocker.
Russian artist and Putin critic shot dead in Poland [comments]
145 points · 70 comments · www.bbc.com · 19h ago
The article reports that Russian satirical artist Semyon Skrepetsky (real name Robert Kuzovkov) was shot dead execution-style in a Polish town near the Belarusian border, with two Belarusian citizens detained. The Hacker News discussion immediately split: one camp treated it as an obvious Russian or Chechen assassination, pointing to his caricatures of Putin, Lukashenko, and Kadyrov, while others argued the headline was misleading—Skrepetsky also criticized Ukrainian authorities and appeared on Ukraine’s Myrotvorets database, muddying a clean “Putin critic” narrative. Several commenters pushed back on calls to wait for an investigation, citing the 2019 Berlin Tiergarten assassination as a pattern of daylight hits by Russian agents, and a deeper thread erupted over whether prisoner exchanges like that of Vadim Krasikov legitimize political killings. A vocal minority derailed into accusing the BBC of hyping Russian crimes while downplaying Israeli ones in Gaza, but most of the thread stayed focused on the mechanics of state-sponsored assassination and the bleak reality that dissidents abroad are not safe.
10Gb/s Ethernet: switching to a Broadcom SFP+ module [comments]
134 points · 125 comments · www.gilesthomas.com · 14h ago
The article walks through replacing a hot-running Marvell-based 10GBASE-T SFP+ module with a Broadcom-based one to stop the network from crashing when the module thermally shuts down. The HN discussion quickly turned into a broader debate about whether anyone should bother with copper 10GbE at all—multiple people argued fiber or DAC is strictly superior, pointing out that 10GBASE-T runs stupidly hot and power-hungry, and the only real reason to use it is if you already have Cat6/6a in the walls and don't want to pull new cable. There was a lively subthread about the fact that the replacement module lies about its identity in its EEPROM (pretending to be an Intel fiber optic module), which is apparently common for compatibility, and that spawned a whole tangent about reprogramming SFP modules using tools like Ubiquiti's SFP Wizard or even a Banana Pi. Several commenters also pushed back on the idea of staying with copper for future upgrades, recommending single-mode fiber as the only future-proof choice for new installs, since MMF is hitting a dead end at higher speeds.
The UK's Teen Social Media Ban Is Political Theater, Not Child Safety Policy [comments]
125 points · 128 comments · www.techdirt.com · 9h ago
The article argues that the UK’s proposed ban on under-16s using social media and its anti-sexting demands are cynical political theater from an unpopular Prime Minister, not real child safety policy. The thread split sharply: some dismissed Mike Masnick’s critique as self-serving because he sits on Bluesky’s board, pointing out that social media companies’ own internal research shows clear harm, while others countered that the real issue is adults having systematically destroyed physical spaces where kids could gather, leaving digital spaces as the only option. A few commenters noted Australia’s ban is already a joke—kids just don’t create new accounts—and questioned why the UK would undermine its own just-finished Online Safety Act with rushed, unenforceable laws. There was also a sub-thread arguing that calling “ban social media” authoritarian is naive when it’s the rare policy with 90% public support, though that claim got pushed back with evidence that “think of the children” is always a cover for something else.
Generated 2026-06-17 08:35 UTC
Generated by Sauron from Hacker News discussions and linked articles.