HN Brief: 2026-11-03
Today's Hacker News was shaped by two major throughlines: the AI industry's financial reckoning and a systemic erosion of trust in critical infrastructure. The cluster of Anthropic threads—debating its slowing growth, trillion-dollar compute obligations, and IPO strategy—presented a sector wrestling with its own economics, while the Ubuntu patch delays and the sabotaged supermarket freezers exposed deep vulnerabilities in software supply chains and operational technology. A secondary undercurrent of nostalgia for simpler, more repairable hardware offered a quiet counterpoint to the narrative of inevitable progress.
The threads most worth clicking into are 'Astra can read and write weights to improve Qwen' for the genuine shock over a potential paradigm shift in AI, 'WTF Ubuntu?' for the furious debate on Canonical's breach of trust, 'If you love a website, back it up' for the explosive war over personal archiving and the open web, and 'Supermarket freezers across Europe hit by second suspected cyber-attack' for the damning indictment of operational tech negligence.
Astra can read and write weights to improve Qwen [comments]
2466 points · 772 comments · rikiyat.github.io · 13h ago
A blog post argues frontier LLMs can natively read and write neural net weights, editing a Qwen model to fix its reasoning loops or building game-playing networks from scratch without gradient descent. The reaction on the thread was genuine shock, quickly cleaving into a debate: is this a fundamental new capability that makes symbolic optimizers look like archaic artifacts, or is Astra doing brute-force mechanistic interpretability inside its context window? The consensus settled on a paradigm shift either way—the concept of "weight-space programming" is here. The conversation also took a characteristically HN turn into a heated meta-argument about whether this makes superalignment trivial or just accelerates inequality.
WTF Ubuntu? [comments]
1470 points · 619 comments · chalmovsky.com · 17h ago
A furious blog post excoriates Canonical for taking over two weeks to patch a critical remote root exploit that Debian volunteers had already fixed, branding Ubuntu’s “timely patches” commitment a hollow marketing lie. HN is unanimous that the delay is a catastrophic failure for a company selling security subscriptions, but the discussion immediately splits over whether this forces a distro change or just confirms declining trust in the Ubuntu brand. People who run servers for a living push back on the simple switch-to-Debian reflex, pointing out that Ubuntu's HWE kernels and predictable LTS life cycle solve real problems that pure Debian stable doesn't cover without backports or running testing. A totally unexpected tangent erupts over whether current agentic AI makes migrating distros a solved weekend project, which gets shut down by the practical crowd who rightly refuse to hand their private keys to an LLM, leaving the thread with a grim consensus—the breach of trust with Canonical is real, but the migration path away from Ubuntu looks a lot fuzzier than the blog's anger suggests.
Show HN: Subvocal – Talk to your computer without making noise
663 points · 331 comments · news.ycombinator.com · 14h ago
This was a Show HN for a YC-backed startup building glasses that use radio waves to decode silent mouth movements, letting you dictate text at high speed without making a sound. The discussion instantly split between people genuinely impressed by the tech and those who wanted it in a less conspicuous form factor—floated alternatives included in-ear buds and baseball caps—but the founders explained why cameras hit a fundamental ~17% word error rate wall (visemes versus phonemes) and why the RF sensors had to be that close. The biggest turn in the conversation came when the creators revealed that the system actually doesn't require visible lip movement at all, just the internal shaping of your mouth with lips closed; that "internal speech" detail reframed the whole pitch from a niche hack for libraries into something genuinely futuristic for private use. The rest of the conversation tackled the practical tradeoffs, like whether speaking code will ever beat typing it, why the landing page buried the lede on how the tech actually works, and how the open-source Subvox project compares as a free alternative.
If you love a website, back it up [comments]
514 points · 112 comments · blog.mistys-internet.website · 18h ago
A web archivist wrote a tragic tribute to a lost fansite and a tutorial begging everyone to back up the web, naturally built around his own company’s tools. The thread completely bought the mission but completely ignored the vendor pitch, exploding into a detailed war over the actual best workflows for personal archiving. SingleFile and the Linkding+SingleFile combo crushed it for dead-simple, browser-side captures, while ArchiveBox caught real heat for breaking on any site behind a Cloudflare wall. The loudest vein of the discussion then vaulted past bitrot into a political panic, arguing the real urgency is the active, biased rewriting of history by LLM providers and the centralization of information. What started as a “here’s how to save a website” post ended as a consensus that the open web is disintegrating and no one trusts the platforms to preserve anything.
Ubuntu still has not released patch for Linux 0day RCE in almost 3 weeks [comments]
456 points · 128 comments · freenode.net · 13h ago
This is a thread about Ubuntu taking three weeks and counting to patch a CVSS 9.8 kernel RCE across its LTS line, with no fix in sight. A chunk of the thread veered into criticizing the source — the current stewards of freenode.net are widely considered hostile to open source, making the domain a weird choice to highlight this story even if the facts hold up. On the substance, the split was sharp: one camp said everyone should just drop IP options packets and move on, while the other pointed out that Red Hat explicitly says that hack can't fully stop it, and that Ubuntu dragging its feet while RHEL shipped a fix days ago is a real failure of the LTS promise. A few people volunteered they'd already jumped ship to Debian or HardenedBSD, and a brief, pedantic argument flared up over whether a three-week old public bug can still be called a "zero-day".
Supermarket freezers across Europe hit by second suspected cyber-attack [comments]
389 points · 431 comments · www.theguardian.com · 20h ago
The thread covers a second wave of coordinated attacks sabotaging supermarket freezers across Europe by remotely cranking the temperature to 50C, destroying stock mere weeks after an identical incident in France. HN's reaction completely bypassed the article's hand-wringing about state-sponsored hybrid warfare, focusing its fury instead on the staggering engineering negligence that allowed it. A deep split emerged between defenders who said remote connectivity is unavoidable for grid balancing and compliance, and a far louder camp arguing any system with remote write access to temperature controls is fundamentally broken by design. The harshest pushback came from industry insiders who treated the first French attack as an unmistakable red flag, making the failure to even segment the vulnerable Danfoss controllers before the second wave an act of gross negligence rather than sophisticated espionage. A bleak consensus settled that operational tech security is a decade-deep structural failure, and that the attackers are simply exposing how absurdly cheap and easy mass sabotage has become.
US boy names that became girl names and vice versa [comments]
373 points · 357 comments · namedaisy.com · 15h ago
The article crunches the data on 125 US baby names that swapped gender associations, and its headline finding is that girls are the prime movers on both ends of the see-saw—names tip toward girls when girls adopt them, and toward boys when girls drop them. The thread zeroed in on the obvious social corollary, broadly agreeing that the one-way stigma against femininity in men resolves the asymmetry: parents avoid uncertain boy names far more than uncertain girl names. Corrections and tangents kept it grounded, with people pushing back on the notion that a name like Arya represented a true "flip" from boy to girl (it barely existed before pop culture adopted it for daughters), and a massive side debate broke out over whether patrilineal naming traditions structurally anchor the male name pool. At the end of the day, the whole thread kept arriving back at the same conclusion: the data just reflects the society we've got, where moving toward masculine is neutral or positive and moving toward feminine is a risk.
The SEC Should Block Anthropic's IPO to Protect the Public [comments]
318 points · 236 comments · www.techpolicy.press · 9h ago
The linked article argues the SEC should block Anthropic’s IPO because the company carries nearly a trillion dollars in non-cancelable compute contracts it treats as debt, supported by slowing revenue and unaudited financials. The conversation immediately bypasses the SEC’s legal theory and fixates on whether retirement index funds tracking the S&P 500, Russell, or CRSP indices will be forced to buy Anthropic when it lists—replaying the entire SpaceX fast-track debate. This splits the room cleanly between people who think the market should just be allowed to price the risk and people who see mandatory index buying as a backdoor bailout for insiders and VCs before the crash. The strongest pushback comes from domain experts explaining that most index funds rebalance over months and have broad latitude to skip risky new positions, and that the bottom half of Americans own almost no equities anyway—which undercuts the article’s core premise that ordinary retirees are the ones exposed.
Anthropic's internal AI use, at retail prices, now exceeds its employee pay [comments]
317 points · 248 comments · www.businessinsider.com · 15h ago
The article highlights Anthropic's IPO disclosure that the value of its internal Claude usage, calculated at what it charges customers, has overtaken its total employee payroll—a metric presented as a signal that AI is economically surpassing human labor. The discussion immediately seized on the article's own caveat that this isn't actual cash outlay, framing the disclosure as a carefully crafted piece of IPO theater designed to show investors the future is now. The split came down to whether this is a legitimate opportunity-cost metric in a compute-constrained world or pure accounting nonsense, with one side comparing it to Microsoft charging itself full retail for every Office license and the other defending the retail valuation as the correct cost of burning revenue-grade compute. The thread then burrowed into side debates over whether Anthropic’s consumer plan actually delivers the value claimed at API rates and whether running this much inference masks an obsession with benchmarks over reliability. The deeper consensus rejected the headline's sweep, concluding the stat tells you more about the unique incentives of a pre-IPO AI lab than about the economy broadly shifting talent spend to tokens.
BYD sold more EVs last month than the US did all quarter [comments]
262 points · 269 comments · electrek.co · 10h ago
The submission centers on BYD selling over 300,000 EVs in October, outpacing the entire US third-quarter total on its own. The central split in the reaction was a refusal to accept the “West is falling behind” frame, since the US is the obvious laggard while Europe and Australia are buying EVs at a China-like pace. A huge stats argument erupted over whether China is “doubling down on coal”—the side accusing the US of cherry-picking per-capita or raw totals clashed directly with those pointing at China’s absolute coal capacity additions. The real focus ended up being the structural advantages behind BYD, with general agreement that vertical integration and a hyper-competitive home market leave American automakers flat-footed. This was reinforced by detours into the reliability problems of rivals like Kia, which lingered as a case study in why legacy EV makers keep trailing.
Show HN: Warsaw in voxels – An interactive 3D map of the city centre [comments]
259 points · 41 comments · p.migdal.pl · 15h ago
The Show HN is a real-time 3D voxel map of downtown Warsaw stitched together from OpenStreetMap, LiDAR, and the city's tree database, complete with weather toggles and the ability to possess a pedestrian. The thread's main battle was between people running it smoothly and everyone else (Linux, Safari, old GPUs) hitting a terrible framerate, leading the author to scramble and ship performance fixes live. Locals called out the missing Old Town and rough building shapes, and the author copped to it — the raw open data is full of floating trees and bombed-out-looking bridges that require heavy manual polish, so the map expands carefully from the center. A tangent compared the honest data projection to recent AI-generated traffic slop, and while a few pedants debated the definition of a voxel, the thread landed on the open data mashup being the real draw, warts and all.
Anthropic reports $17B in Q3 revenue, commitments of $895B in revised S-1 [comments]
230 points · 271 comments · www.cnbc.com · 17h ago
The article covers Anthropic's S-1 filing, highlighting $17B in Q3 revenue from a 47% quarterly growth rate (a steep dive from triple digits), a $104B net loss mostly attributable to convertible note GAAP accounting, and a staggering $895B in committed infrastructure spend. The thread immediately seized on the growth deceleration and the sheer size of the compute obligations, splitting sharply between those who dismissed the quarter-to-quarter slowdown as noise on a healthy S-curve and those who argued the whole narrative depends on compute costs cratering or growth somehow re-accelerating to absorb the infrastructure burn. A surprisingly loud and specific debate erupted over moats, with one side arguing these financials confirm the model layer is a commodity where users will leap to DeepSeek or whatever is cheapest and good enough next week, while the counter-argument insisted that enterprise integration, brand trust, and a genuine quality gap on the very best coding models provide a real defensibility that pure revenue churn numbers miss. No one really settled whether Anthropic is a disciplined long-bet business that can survive a growth lull or a capital-intensive bet that needs frictionless hypergrowth to justify its pre-IPO pricing and infrastructure debt.
Show HN: I've updated my Linux System Call website and API [comments]
219 points · 52 comments · syscall.sh · 22h ago
The submission is an update to syscall.sh, a site that compiles the Linux syscall tables for every architecture across kernel versions, including calling conventions and a JSON API. The thread quickly zeroed in on a heated debate over whether the roughly 400 syscalls represent an inflated count—a philosophical tangent the reference site itself avoids entirely—given how many are obsolete variants (pipe/pipe2) or are being absorbed into the single io_uring interface. The forum split between people arguing the raw number artificially inflates complexity and those countering that distinct ABI numbers are the only objective technical reality, making similarity metrics an impossible quagmire. On the practical side, someone caught that the x86_64 calling convention was wrong—the kernel reads eax for the syscall number, not rax—which the author promptly confirmed and fixed with a note about the behavioral change around kernel 5.14. The discussion also expanded into how glibc hides radically different syscall tables from developers, with newer architectures like RISC-V strictly dropping legacy calls, making the resource essential for navigating the gap between the POSIX API and the actual kernel ABI.
Ask HN: Who is hiring? (November 2026)
213 points · 168 comments · news.ycombinator.com · 15h ago
The November “Who is hiring?” thread landed overnight, stacking up a dense mix of job listings posted directly by founders and hiring managers rather than recruiters. The usual startup spread leans heavily into AI engineering, DevOps, and infrastructure, but there's a strong showing of privacy-focused bare-metal shops and European remote gigs mixed in with the cloud-native fintech. The liveliest exchange broke out under Simple Analytics, where the listed €5k/month salary for a remote role drew pushback as low for the Dutch market, prompting a direct back-and-forth on what full-time actually means in a global remote setup. Elsewhere, the same founder explicitly begged applicants to leave their LLM off for ten minutes and own their voice, a plea that resonated in a thread full of people building agents to automate that exact kind of application. The split running through the whole thread is pretty clean: well-funded Series B insurance and fintech shops paying big-city salaries versus bootstrapped tooling teams offering equity, autonomy, and noticeably smaller paychecks.
Show HN: Hacker Times, a New York Times-inspired version of Hacker News [comments]
206 points · 67 comments · hackertimes.ca · 6h ago
The submission is a Show HN that reformats the Hacker News feed into an hourly, NYT-style printed broadsheet to make the firehose more skimmable. The thread split hard and fast—a contingent praised the layout and scannability, but a bigger chunk immediately wrote it off as another LLM-generated clone of dead projects like Paper HN, dragging the whole thing into a familiar HN civil war over vibe coding. The pushback was almost as loud: defenders argued the implementation here dodged the usual AI slop aesthetic and that the tool didn't warrant dismissing the result, which escalated into a broader argument about HN upvoting the very GenAI outputs the community claims to hate. From there the conversation drifted into nostalgic laments for extinct digests like N-gate and Instapaper's Kindle magazine, an esoteric font debate over the use of Manufacturing Consent, and genuine head-scratching over the copyright of republishing HN stories in such a rigid format. The site itself was celebrated as a fun remix, but the thread mostly became a mirror for the community grappling with its own conflicted relationship with generative AI remixes.
Anthropic's growth just flattened [comments]
203 points · 234 comments · www.regenerator1.com · 9h ago
The linked piece parses Anthropic's newly shared Q3 financials to argue that sequential revenue growth has flattened sharply over the summer, suggesting the AI spending bubble might be losing air. The thread’s biggest pushback was simple timing: Opus 5.5 dropped in late September, so its impact is a ghost in this data, making the whole "flattening" diagnosis feel premature to a substantial camp. Another large contingent spent the time litigating whether the revenue is structurally sound, debating if subsidized inference and VC circular financing are inflating the books beyond what the unit economics truly support. The deepest fight was over the balance sheet math, splitting the room between those who think the required returns to justify the trillion-dollar infrastructure bills are simply "a lot" versus those who call them mathematically unsalvageable without hypothetical moats that don't exist yet. A grim underlying consensus did float to the surface: every single one of these businesses is a knife-edge away from irrelevance if they slip to second place in coding, which is a zero-moat position to be in.
Self-Employed? Solopreneur? NIST Has a Cybersecurity Guide Just for You [comments]
199 points · 54 comments · www.nist.gov · 14h ago
NIST released a new cybersecurity guide explicitly tailored to solopreneurs and small business owners without dedicated IT staff, translating its broader Cybersecurity Framework into plain language. Opinion immediately split on whether the guide actually hits its layperson target—some praised it as a solid, well-linked primer, but others argued even this "simplified" version leans too heavily on technical jargon without properly grounding concepts for a true beginner. A significant portion of the reaction completely veered away from the guide's actual content, latching instead onto the document's mandated use of "Super Intelligence" (SI) instead of AI, leading many to dismiss the entire document's credibility over what they saw as a performative political loyalty test rather than a genuine security resource. Beyond the political static and the jargon debate, the general consensus conceded the underlying advice is sound NIST guidance, but cautioned that a PDF is no substitute for actually finding a trusted consultant or managed service provider to lean on.
Intel stock collapse worsens as $3.5T chip rout spreads [comments]
196 points · 151 comments · finance.yahoo.com · 7h ago
The Yahoo Finance article covers the $3.5 trillion chip rout and the alarm around Intel losing half its value, with experts fighting over whether it's a bottom or the start of major train wrecks. The thread immediately zeroed in on the Nvidia/Intel divergence, landing on a sharp consensus where Nvidia's iron grip on AI training and inference plus its insane margins make it a toll collector that Intel can't touch—especially since Intel bet the farm on laptop chips just as the whole market pivoted to data center AI. The pushback split into two camps: one insists Nvidia's dominance is fragile, pointing to Google and AMD chips finally getting traction and future architectures like 1-bit nets that could gut its moat, while the other camp piled onto Intel for losing server share to AMD and ARM and letting its fabs fall so far behind that a decade of catch-up is still unfolding. A solid minority of the thread zoomed out to argue this is the first real crack in the AI bubble, with datacenter loans freezing and enterprises tightening budgets, making the whole chip sector look like meme stocks riding a speculative wave that could drain out just as fast as it crested.
Scaling Git at OpenAI [comments]
192 points · 108 comments · openai.com · 8h ago
OpenAI detailed a new Git backend called Pando built specifically to handle their monorepo exploding in size from AI agents pushing an insane volume of concurrent branches and commits. The thread mostly ignored the engineering merits of Pando and instead split hard on whether that staggering growth is a genuine asset or an outright liability—one camp argued code volume is only a problem for limited human engineers, while the counterargument was that every bit of unchecked, agent-generated slop just pollutes the context window and makes future agents dumber in a self-reinforcing cycle. A huge thread sub-debate formed around whether any of this code needs to be reviewed by humans or if verification is just another job for agents, with skeptics pointing to nightmare scenarios where hundreds of thousands of LOC get generated for what used to be a simple project, creating a debt bomb no one can unwind. The mood carried an undercurrent of existential dread about the profession, especially once people tied it back to Anthropic’s similar post and realized both camps are just arguing about how fast the floor is burning instead of whether it’s burning at all.
My Favourite Laptop [comments]
188 points · 136 comments · blog.jsbarretto.com · 15h ago
A developer made the case for their beat-up, £100 Thinkpad X250 as a perfect tool—cheap, repairable, running Linux, with a ten-hour battery—positioning it as a quiet revolt against consumption culture. The thread wholly embraced the frugal-hardware ethos, but the top-line debate immediately crystallized around the disappearance of physical mouse buttons from modern business laptops, a change widely blamed on everyone trying to copy the MacBook trackpad. Another major trench involved the logistics of keeping this dream alive—swapping specific part numbers for aftermarket batteries that actually deliver ten hours, and litigating whether Lenovo has "enshittified" the modern ThinkPad to the point that the old X250 era is unrecoverable. A significant faction dismissed the whole local-compute premise, arguing the ideal machine is now just a durable thin client for remoting into a home server or AI coding agent, making the CPU and battery wars irrelevant. The split ultimately wasn't about whether cheap used hardware is smart—everyone agreed it is—but whether the modern options to pursue it have been ruined, and whether the job profile has changed so much that a dumb terminal is the real endgame.
Net worth for people aged 30–33 in the U.S. [comments]
186 points · 90 comments · www.keepingup.com · 4h ago
The site is a net worth percentile tool for 30- to 33-year-olds, giving a median of $68,000 and a top-10% cutoff of $440,000 based on the Fed's Survey of Consumer Finances. The discussion immediately fractured over the methodology, pointing out the household-level figures inflate single-person wealth and the 2022 data is a pandemic anomaly that has since corrected. The site itself took almost as much heat, widely dismissed as a Claude/GPT-generated lead-gen funnel whose green-button UI and lack of historical context made it impossible to take seriously. The debate pivoted to whether young people can realistically build wealth today, with one side claiming AI, robotics, and biotech are open frontiers while the other side countered that those industries demand too much capital for low-barrier entry compared to the early IT or construction booms. The final consensus was that the data might be interesting, but the packaging was too transparently designed around harvesting emails from high-net-worth individuals to be trusted at face value.
Let Me Ch... [comments]
186 points · 78 comments · blog.alxgsv.com · 14h ago
The story is a work of fiction about a lead developer chasing a time-desync bug between his flower delivery company's server and its warehouse, who gets hit by a car and finds himself permanently stuck in the "no-man's minute" between life and death that he spent the whole story trying to observe. The thread split into a proxy war over whether the prose was AI-generated or human-authored, with one camp citing Pangram scores and stylistic tics like staccato rhythm and labored metaphors as definitive tells a model wrote it. The other camp defended the writing by pointing to its tightly woven thematic structure, the emotional payoff of the cut-off title, and the careful layering of the cheese subplot, arguing the accusations said more about the critic's cynicism than the text. That debate hit a serious factual correction when someone recognized the Russian wordplay on "poka" (meaning both "goodbye" and "for now"), which suggested the story was translated rather than generated, and reframed those supposed LLM tells as artifacts of translation. The thread also went deep into a sprawling philosophical riff on "meat suit firmware" and whether the tech industry engineers enjoyment out of the journey on purpose, giving the story's defenders the perfect real-world example to argue the detractors were proving the author's point by trying to debug a piece of art.
Oracle files to sell $20B in new stock, bonds to finance AI buildout [comments]
185 points · 166 comments · www.cnbc.com · 12h ago
Oracle is selling $20 billion in stock and bonds to finance its AI buildout, an already previewed plan that still sent shares down 9% and fanned existing fears about the company's towering debt load. The thread immediately got locked in a protracted, foundational argument over whether the new share issuance actually dilutes current holders, with one camp insisting fresh capital expands the pie proportionally and the other countering that yes, your ownership stake literally gets smaller—and the stock is already down 8%. Sentiment toward the company itself was overwhelmingly hostile, with commenters openly hoping this overleveraged bet on commodity AI infrastructure blows up before the hardware is even fully online. A separate discussion pointed out the quiet irony of Oracle gutting its ERP cash cow to chase a hyperscale buildout where it owns no critical IP, arguing the AI arms race could just as easily render those datacenters obsolete before the debt matures. There was also a familiar side thread on whether holding equity in these companies even hedges against AI taking your job, since the financing strategies themselves can tank the stock price just as fast.
We halved DNS build delay and cut memory use by 85% [comments]
176 points · 70 comments · blog.cloudflare.com · 17h ago
Cloudflare described reworking their DNS pipeline to swap full zone rebuilds for targeted builds, cutting memory by 85% and halving update propagation times. The stated throughput of 700 events per second was immediately met with skepticism from those running the numbers on domain churn rates, a challenge countered by explaining the long-tail distribution and the difficulty of ambiguous zone events. The dramatic memory savings prompted a referendum on contemporary software efficiency, with the post's careful optimization framed as a direct indictment of Electron and the economics of treating hardware as disposable. Beneath the broader finger-wagging, the real technical meat—proving the safety of a targeted build when a Kafka event can mean anything, and handling zones shared across multiple internal objects—largely carried the day as an impressively thorough solution to a genuinely gnarly distributed systems problem.
Jobless tech workers are being left out of San Francisco's AI boom [comments]
174 points · 247 comments · www.bloomberg.com · 17h ago
The Bloomberg piece follows laid-off Bay Area tech workers on a Mount Tamalpais hike, literally left behind as the Anthropic and OpenAI IPO billions prepare to wash through the city. HN almost entirely ignored the AI inequality angle, immediately landing on the idea that these workers are jobless because companies are flooding the market with indentured H1B labor and offshoring mid-skill roles. A fierce counter-argument rose up calling this a familiar, economically-illiterate scapegoat that ignores how the same corporate power grinds down both local and immigrant workers. The most pointed corrections sliced past the identity fight entirely, pinning the blame on the specific exploitative mechanics of the visa system itself—tethering workers to abusive employers, enabling 996 cultures, and letting companies cry talent shortage while fixing wages across the board. Under that entire argument sat a grimmer shared reality: San Francisco is a brutal winner-take-all monoculture where the hiking workers are probably better off just leaving the region entirely.
Moving our warehouse off DuckDB [comments]
171 points · 115 comments · posthog.com · 15h ago
PostHog explains why they’re ditching DuckDB for their cloud warehouse a bare five months after migrating to it from ClickHouse, saying that single-node scaling couldn’t handle the diverse, complex queries their customers ran without bankrupting them on big instances. The thread didn’t really bite on the anti-DuckDB pitch—most of the pushback honed in on the engineering churn itself, calling the post a product teaser for HogLake (their new Trino-based format) that was light on actual measurements. The author hopped in to admit they kept going too long out of sunk cost and that the blog context was missing, which fed right into the skeptical read that this was a five-month proof of concept rather than a serious engineering evaluation. A smaller side argued they should have just run Trino over Iceberg from the beginning, but the author countered that writes were too slow, leaving the takeaway split cleanly between those who think DuckDB genuinely doesn’t work at this scale and those who think PostHog’s rapid stack-hopping is the bigger red flag.
Why are Soviet math textbooks so hardcore in comparison to US textbooks? (2017) [comments]
169 points · 236 comments · www.quora.com · 16h ago
The Quora answer linked here argues that Soviet textbooks were so hardcore because scarce computing and lab equipment forced students to rely on deep analytical thinking and theoretical insight rather than running simulations. The HN discussion barely touched the historical Soviet context and instead sparked a bitter fight over whether the US is dumbing down math in the name of equity, with a vocal camp citing California's supposed removal of algebra as the prime example. That specific alarm was met with immediate pushback from people pointing to California's SB1410 law protecting advanced math tracks, arguing the panic over dumbing down is a political narrative that doesn't match classroom reality. The thread's real, irreconcilable split was over the underlying philosophy: the Soviet "throw the children down a well" efficiency that produces geniuses by writing off everyone else, versus the American ideal of universal accommodation, which the former camp sees as a guaranteed march toward mediocrity.
Never Complain, Never Explain [comments]
163 points · 63 comments · en.wikipedia.org · 20h ago
The article traces the "never complain, never explain" maxim from Disraeli through Elizabeth II to Prince Harry, as a straight history of a royal PR strategy. The discussion mostly ignores the Wikipedia framing and turns into a fierce referendum on the tactic itself—one camp sees it as a crucial "don't let them see you sweat" lever for managing drama, the other as a cynical shield for dodging accountability that looks increasingly ridiculous in a transparent age. Domain-aware participants push back on the tidy historical arc, noting the strategy wasn't gracefully updated but actively failed once the media turned hostile, with Prince Harry's accusation that the palace plants stories while publicly pledging silence exposing the whole thing as a convenient double standard. Modern analogies to Huberman and Trump test whether "ignoring the mob" is a smart crisis tactic or just a privilege of power, with mixed verdicts, and the Churchill quote about not stopping to throw stones at every dog gets deployed like a battle standard. The deepest fault line runs through Bagehot's "don't let in daylight upon magic"—half the room thinks it's a brutally honest insight into how institutions preserve authority, the other half thinks it's an outright confession of a system built on keeping the public stupid.
DiscImageStudio: burn visible artwork onto CDs, DVDs and Blu-ray discs [comments]
158 points · 35 comments · github.com · 6h ago
DiscImageStudio is a tool that encodes images directly into the writable data layer of optical discs, essentially a modern, vastly more ambitious take on LightScribe using the actual burnable surface. The project immediately sparked heavy flashbacks to LightScribe, DiscT@2, and Labelflash, splitting sharply between those who found the slow burn times and faint contrast not worth the hassle and the few who had the settings dialed in for crisp results. That initial nostalgia was quickly met with deep technical validation from someone who built a near-identical CTF challenge, confirming that reliably calibrating a disc image is an incredibly hairy problem this project seems to have cracked better than most. The conversation also veered into a debate over the heavy use of anime girl art, with some finding it off-putting while others contextualized it as standard aesthetic in Chinese developer circles. The broad consensus was that this is a genuinely impressive technical solution to a mostly dead medium, carving out a practical niche that somehow still feels magical.
Egoless Mathematics in the AI Era [comments]
154 points · 105 comments · terrytao.wordpress.com · 13h ago
The guest post is a philosophical appeal for mathematicians to embrace an "egoless," Platonic view of their field—treating mathematical truth as independent of its discoverer—to overcome the existential anxiety created by AI's growing capability. The thread immediately divided into those who found the idea spiritually resonant and those who found it dangerously detached from reality. The pro-Platonist camp, largely practicing mathematicians and engineers, defended the idea that the *understanding* itself is the reward, but they were met with a fierce rebuttal that the essay ignores the brutal economics of academia, where credit, reputation, and publications are the only currency that matters. Beyond this core debate, the discussion broadened into whether AI represents the fourth "narcissistic wound" to humanity and whether the fundamental purpose of learning math changes if a machine can outperform you at it. The sharpest critique, however, labeled the whole essay as a Silicon-tinged appeal to humility that conveniently serves the interests of those who would rather not fairly compensate the humans doing the work, roughly following the logic of a linked satire called "Lean Me Softly."
Generated 2026-11-03 08:08 UTC
Generated by Sauron from Hacker News discussions and linked articles.