Today’s Highlights#

Start with the TV that would not sleep. LG smart TVs caught logging audio with screen off and snooping on local devices turned the quietest box in the living room into the day’s most talked-about story: the screen goes dark, the microphones and network scans keep going. It resonated because it made an abstract privacy policy audible in packet captures. Follow that thread and the rest of the day comes into focus. On one side, control over personal tools — whether compression benchmarks are fair, how a Python interpreter can fit in 1,024 bytes, and how a million surviving buildings can be stitched into a draggable history of Los Angeles. On the other, broader questions of stewardship — how a 210-petabyte public memory stays online on monthly gifts, why a neutral state is moving 3,000 workstations off a U.S. cloud, why repairability labels rarely lead to a repair, and how a deliberately slow vacation can retrain attention.

Policy and Governance#

LG smart TVs caught logging audio with screen off and snooping on local devices A TV that keeps listening after you turn it off#

Reporting summarized by Notebookcheck describes testing by Gamers Nexus with Level1Techs and independent researchers on several LG OLED sets running webOS, the operating system that powers LG TVs. According to the investigation, the sets kept capturing microphone audio while in standby with the screen off, buffered it locally, and uploaded once reconnected, while also sweeping the local network for phones, watches and other devices, collecting IP addresses, nearby Wi-Fi names and signal strength, and fingerprinting what is on screen across inputs via ACR, a system that samples audio and video to identify what you are watching. The piece says the data feeds LG Ad Solutions, which claims to link hundreds of millions of secondary devices in the U.S. The researchers also reported remotely exploitable flaws in webOS now in coordinated disclosure, and for now advise disconnecting the sets or using an external streamer.

HN readers largely treated the findings as a trust threshold. Many said they already keep TVs offline and use them as dumb displays, while others warned that blocking is not simple — commenters cited spoofed MAC addresses, hard-coded DNS and attempts to hop onto nearby hotspots. The thread’s other strand questioned enforcement, arguing that vague consent screens are little protection without verifiable limits on sharing data with third parties.

Discussion: Hacker News thread

Switzerland’s Federal Government Is Replacing Microsoft on 3k Computers Moving 3,000 workstations off Microsoft 365#

The Swiss federal administration is piloting a move from Microsoft 365 to the German open-source suite openDesk on about 3,000 workstations — roughly 7% of its 54,000 machines — with about CHF 9 million budgeted and completion targeted for the end of 2027, as covered in Switzerland’s Federal Government Is Replacing Microsoft on 3k Computers. A 172-person proof of concept found core office work and email solid, with large video meetings still rough. The military’s Cyber Command is set to move faster, aiming for October. The article frames the push around three risks: exposure to foreign access claims under U.S. cloud law, operational dependence on a single vendor, and rising license costs, against the backdrop of a 2024 law that makes government-built software open by default and a federal focus on digital sovereignty.

The HN thread split between pragmatism and principle. Some readers argued that reinvesting even a fraction of license savings into projects like LibreOffice could improve the alternative over time and loosen platform lock-in. Others cautioned that the switch is not just software but organizational change, with device support and document fidelity still uneven on the desktop. A second theme was geopolitical: with current pressures, cost was no longer the decisive line, sovereignty was.

Discussion: Hacker News thread

Smartphone makers don’t bother to comply with EU repairability requirements Repairability labels that lead nowhere#

A year after the EU’s repairability rules for phones and tablets took effect in June 2025, a review of the European Product Registry for Energy Labelling cited by The Register in Smartphone makers don’t bother to comply with EU repairability requirements found that of 2,334 phone listings, only about 18% provided a working link to spare-part prices and repair instructions; roughly half were blank, nearly a fifth pointed to generic product or support pages, and a few even pointed to Temu or AliExpress. Even when a link existed, readers found it often took several clicks to reach a price, with wide ranges and missing labels elsewhere. The advocacy group Right to Repair Europe criticized self-assigned repairability scores without published backing and called for open documentation and easier reporting; the European Commission had been asked for comment with no reply at press time, while a user-replaceable battery mandate is due in February.

Comments focused less on whether the law is good and more on whether it is enforced. Readers pointed to the UK’s food hygiene scores and Euro NCAP as examples of rules that matter only because inspections and penalties are visible and timely. Others argued for independent testing and published evidence behind each grade, warning that without that, high self-scores will keep passing.

Discussion: Hacker News thread

Society and Culture#

Keep Our Servers Running Keeping 210 petabytes online, one monthly gift at a time#

The Internet Archive’s September drive, described in Keep Our Servers Running, offers a two-to-one match on new recurring gifts of $25 or more, turning $25 into $75 of impact for the month. The post stresses that the Archive is free to use, carries no ads, does not sell data, and builds its own servers, storage and crawlers that now hold about 210 petabytes, including the Wayback Machine. The organization says its typical gift is around $25 and that recurring support helps budget for storage, power and staffing as costs climb.

HN discussion largely defended matching as a proven fundraising tool, arguing that a large donor can multiply participation and that both sides feel a stronger sense of contribution. Pushback questioned why the money is not simply given, or flagged the checkout experience — several readers noted that the default flow felt subscription-like and that canceling required emailing support unless the gift was tied to an account.

Discussion: Hacker News thread

De-Brainrot Vacations Two slow weeks to de-brainrot#

In De-Brainrot Vacations, a software engineer traces a shift from intense early-career learning to a present where short video, games and AI assistance fill most gaps, leaving thinking slower and reading rarer. He used a two-week stay with family in the countryside to deliberately slow down — walks with his dog, mate with relatives, board games — and read several books, including Mary Beard’s SPQR and a popular physics title. Curiosity about the math in that book led him to rework pre-calculus systematically with web-native materials like Paul’s Notes and Active Calculus before returning to a university physics text. The author presents the stint not as a productivity hack but as a reminder of learning for its own sake, reporting less mental laziness afterward.

Top comments likened today’s attention environment to the shift from physical labor to desk work that once required inventing the gym. Many readers argued that ambient thinking, like ambient movement before, has been engineered out of daily life and must be deliberately reintroduced. Others stressed the value of boredom itself, saying endless feeds fill the gaps where reflection would otherwise do its work.

Discussion: Hacker News thread

Babylonian Lamb Stew with Beets (1750–1730 BCE) A lamb and beet stew from nearly 4,000 years ago#

The Yale Babylonian Collection’s reconstruction of Babylonian Lamb Stew with Beets (1750–1730 BCE) draws on tablets from about 1750–1730 BCE. The modern rendering calls for diced lamb leg, rendered sheep fat, beer and water, with onion, arugula, shallots, cilantro, cumin, beets, leek and garlic, simmered until thick and finished with coriander seed and a kurrat-cilantro paste, alongside a barley-based beer-bread for serving. The page notes that the beets are much as we know them today and links the dish to later Iraqi Jewish Hamed stews, suggesting a long local thread in ingredients and technique before the Columbian Exchange.

Readers focused on how continuous the lineage feels and how practical the recipe is now. Some noted that in Australia and other lamb-heavy food cultures, saving roast fat or sourcing tail fat from a halal butcher is straightforward. Others traced names and crops across languages and eras, cautioning that modern recreations are plausible approximations rather than exact replicas.

Discussion: Hacker News thread

Tech and Products#

bzip3 The successor to BZip2 and the fine print on benchmarks#

bzip3 presents itself as a faster, stronger follow-on to BZip2, built around an order-zero context-mixing entropy coder, a suffix-array Burrows-Wheeler transform, and a prediction stage that blends LZ77 string matching with PPM-style modeling. The project page benchmarks a big tar of every Perl 5 release against xz, bzip2 and zstd at varying block sizes, then adds a second round after long-range deduplication with lrzip, listing compressed sizes and decompression times and offering build instructions via source or Homebrew. A disclaimer notes the algorithms’ complexity and a small but non-zero risk of edge-case flaws.

HN’s main critique was comparability. Commenters pointed out that the original numbers gave bzip3 a 512MB block while leaving zstd near its default window around 8MB — a mismatch that is especially flattering on concatenated source histories — and showed that passing –long shrinks the zstd result by an order of magnitude. The thread’s consensus was to weigh compression ratio together with decompression time and memory, rather than crowning a winner on one number.

Discussion: Hacker News thread

Making a Python interpreter in 1024 bytes Fitting Python into 1,024 bytes#

Austin Henley recounts a weekend attempt to squeeze a Python-like interpreter that can run a FizzBuzz with def, if, for and while into 1,024 bytes of C in Making a Python interpreter in 1024 bytes. The design skips tokenization and bytecode and instead parses and executes directly with recursive descent, keeping state in a few globals and a pointer; loops and calls work by remembering a source position and jumping back to re-parse. Single-letter variable names simplify the symbol table. To shave bytes, the author leans on GNU C89 quirks — implicit ints, zero-initialized globals, one-letter names, ternary and comma operators, and bit tricks — compressing a readable version of about 4,800 bytes down to exactly 1,024 and publishing the golfed source with a demo.

Commenters admired the compression while debating the cost of indentation-as-syntax. Several noted that significant indentation makes the lexer non-regular and requires an indent stack, which is awkward at this scale. Others framed the project as an exercise in trade-offs: drop error handling and generality to buy size, and you learn more about taste and limits than about a practical language.

Discussion: Hacker News thread

Watch Los Angeles get built, one building at a time (1880–2026) Lighting up a city by year built#

Watch Los Angeles get built, one building at a time (1880–2026) is an interactive map that stacks roughly a million surviving buildings by the year they were built, with filters for decade and height and views for Downtown, the Wilshire Corridor and beyond. The site explains it shows what remains today, not the city as it was, and highlights bursts in the 1920s and the postwar decade — each over 230,000 buildings still standing — with slower stretches in the 1930s and after the 1970s. The author describes stitching together fragmented public sources on Socrata and ArcGIS and notes the animation batches buildings into five-year buckets for performance rather than popping each one at its exact year.

Readers enjoyed spotting their own blocks and then turned to why certain decades surged. Top comments tied the pattern to zoning and housing policy, citing how Los Angeles was zoned for around ten million in the 1960s and roughly four million in the 1980s, with knock-on effects for population growth and prices. Others praised the data wrangling itself, calling the fragmentation of civic datasets the hardest part of the project.

Discussion: Hacker News thread

Business and Platforms#

‘You Can See Everything’ Review: Nathan Fielder’s Doc About Elizabeth Holmes When a startup story becomes a self-sealed narrative#

Variety’s review of the 174-minute A24 film ‘You Can See Everything’ Review: Nathan Fielder’s Doc About Elizabeth Holmes — co-directed by Nathan Fielder and Lance Oppenheim and shown as a surprise at Telluride — follows the filmmakers into Elizabeth Holmes’s Del Mar home in the 34 days before her prison term. The reviewer describes Holmes asserting she never lied, recasting Theranos’s failure as chemistry still to be solved while the device itself was sound, and recounts the documentary’s second half, where actor Amanda Seyfried performs transcript readings, creating a hall of mirrors between life and portrayal. The review argues the film does not excuse its subject; it studies how a sealed narrative can make harm feel abstract to the person telling it.

The HN thread asked why the story traveled so far. Commenters pointed to a moment when identity and promise aligned with what many wanted to believe, weakening diligence, and drew parallels to cases like Wirecard. On the film itself, readers were startled that figures who looked like actors were the real participants, and debated the line between public reckoning and voyeurism.

Discussion: Hacker News thread

TiVo to charge money for skipping commercials in your own recordings A skip button that used to be free is now a subscription#

Cord Cutters News reports that TiVo will end free one-button and automatic commercial skipping on November 2 and test a paid add-on, TiVo to charge money for skipping commercials in your own recordings, with a 30-day trial and pricing to come. The existing SkipMode, which relied on cloud markers to jump between program segments, will no longer be included; manual 30-second skip and fast-forward remain. The change lands as parent company Xperi steps back from DVR hardware toward smart-TV software and ad tech, turning a signature convenience into an optional paid tier.

Reactions split between fatigue and workaround. Some readers said they would simply pay to reclaim quiet viewing, while others cited counterparts abroad — blocked fast-forward on recordings in France or platforms that insert their own ads when you skip — and pointed to local tools like Tvheadend with comskip. A smaller group argued the clearest leverage is to watch less.

Discussion: Hacker News thread

Science and Research#

Caltech Mathathon – first hackathon ever devoted to research level mathematics A 40-hour hackathon for research-level math#

A student-led Caltech Mathathon – first hackathon ever devoted to research level mathematics is set for Oct 30–Nov 1, inviting about a hundred teams to chase open conjectures or propose new theory with frontier models over 40 hours, then defend their work to mathematician judges. The site cites recent AI-assisted results — an August disproof of a decades-old Erdos conjecture on unit distances and an explicit non-sofic group construction — to pose two questions: how fast can ideas move to peer-reviewed publication with AI, and what becomes of the mathematician’s role. Organizers advertise more than $2M in model credits and two prize rounds, one for promising, well-explained results at the event and one after community verification, and say chat logs will be recorded and prior work must be disclosed.

HN’s debate balanced opportunity and fit. Supporters argued that choosing the right problem and understanding the proof remain the hard parts, and that the format could reward clear exposition and verification. Skeptics questioned whether a sprint suits mathematics, which often depends on slow reflection, and worried that long model runs would make the event feel like waiting for output. Organizers replied that novelty and understanding during the event would be the judging focus.

Discussion: Hacker News thread

Icy Moons Are Ocean Worlds Six proven oceans that rewrite the outer solar system#

Maciej Cegłowski’s essay Icy Moons Are Ocean Worlds traces how three missions — Voyager, Galileo and Cassini — plus Hubble and Webb, turned icy moons from a niche curiosity into prime places to look for life. The piece counts six confirmed global oceans beneath ice — Europa, Ganymede and Callisto at Jupiter; Enceladus, Titan and Mimas at Saturn — with candidates including Dione, Pluto and several Uranian moons. It walks through the evidence: tidal heating, induced magnetic fields and plume samples, noting Enceladus’s south-polar jets flown through by Cassini with salts, organics and hydrogen, and Titan’s double interest as a hydrocarbon-surface world with a salty subsurface sea, and closes by ranking targets on rock-water contact, age and activity.

Readers appreciated the synthesis and added nuance on radiation, pointing out that Europa’s punishing surface dose does not apply to Ganymede, which has its own magnetic field, or to far-out Callisto, among the calmest spots in the system. Others questioned whether floating volcanic rocks could plausibly provide a rock-water interface under high-pressure ice, expecting such material to be rare and short-lived.

Discussion: Hacker News thread

Closing#

Together, today’s threads pull in one direction: capability is getting cheaper, but trust is getting more expensive. Whether the device in question is a TV, an archive, a phone or a brain on vacation, the recurring question is whether we can see and verify the boundaries we depend on — and who pays when they slip. The hopeful counter-thread is also practical: clearer disclosures, independent checks, and small, repeatable practices — reading, walking, rebuilding attention — are still within reach. See you next time.