Switzerland gets 25Gbit home internet because regulators treated fiber as shared infrastructure, not a product.
What the article says
- The core argument: fiber is a natural monopoly, like water pipes. Building it once as shared, neutral infrastructure and letting ISPs compete on top is better capitalism than letting incumbents carve up territory.
- Switzerland mandates a dedicated four-strand fiber line to every home, terminating in an open hub any ISP can use. You switch providers with a phone call and no technician visit.
- The US model handed incumbents geographic monopolies and let them build shared, neighborhood-splitting networks. Your gigabit plan is shared with dozens of neighbors; your only alternative is often decade-old DSL.
- Germany went the other way, allowing competitive overbuilding, but wound up with redundant trenches and courtroom battles rather than faster internet for anyone.
- Switzerland's system was nearly dismantled in 2020 when Swisscom tried switching to shared architecture. A small ISP filed a complaint, regulators blocked the move, and courts backed them up.
What HN is saying
- The sharpest pushback: the article never mentions population density. Switzerland is a small, wealthy, dense country, and several commenters argue the comparison breaks down at continental scale.
- Others counter that density is a red herring. US suburban areas have plenty of density, and electricity reached rural America long before fiber did. The real issue is private infrastructure with no sharing mandate.
- A few commenters point out that Utah's Utopia network has run an open-access fiber model for years, proving the idea works on US soil, just nowhere near enough people.
- Several people note the Speedtest global rankings don't actually show Switzerland dominating, suggesting real-world performance may be more complicated than the headline speed implies.
- The thread splits between 'this is a useful policy argument' and 'anecdote plus agenda,' with one commenter summarizing the piece as 'America Bad, upvotes to the left.'
Virginia just made it illegal to sell your precise location data, and it took effect July 1.
What the article says
- Virginia amended its consumer privacy law to ban selling geolocation data, following Maryland and Oregon.
- The law took effect July 1, 2026. California, Massachusetts, Vermont, and Washington are considering similar bans.
- Virginia's definition of 'sale' is narrower than other states: it covers money changing hands, not all exchanges of value.
- The trend follows a 2024 federal settlement that banned a data broker from selling location data entirely.
What HN is saying
- Key correction: the law only covers precise location, within about 1750 feet. Fuzzy location data is still fair game, and people note that enough fuzzy data can still profile you.
- Enforcement skeptics point out that companies with no Virginia presence may simply ignore it, and the 'sale only' framing leaves a loophole for barter-style data sharing.
- Several commenters cite real harms the law would address, like a company that tracked visits to Planned Parenthood locations and sold the data for anti-abortion ad campaigns.
- Some wonder why sharing was left legal alongside selling, suspecting the distinction is mostly cosmetic.
Rivian refuses to add CarPlay, and a frustrated Apple podcaster explains why that argument falls apart.
What the article says
- Casey Liss argues that Rivian's chief software officer is wrong: CarPlay doesn't have to take over every screen, and in cars like the Volvo XC90, it coexists alongside the manufacturer's own UI.
- The deeper point is that CarPlay is opt-in. If Rivian's software is genuinely great, drivers simply won't use it. Supporting it costs Rivian nothing and opens the door to thousands of apps no car OEM will ever build natively.
- CarPlay Ultra is the full-takeover version nobody is actually asking for. Plain CarPlay is a windowed option, and Rivian is arguing against something that isn't on the table.
- Liss says he won't buy a car without CarPlay and was seriously considering a Rivian R2 until this. The post ends as an open letter: ship it, and I'll be first in line.
What HN is saying
- The thread broadly agrees CarPlay is table stakes. Several people say they crossed GM off their list the moment it dropped support, and Rivian risks the same fate.
- The strongest counterpoint is from Tesla and Rivian owners who say those native interfaces are genuinely good enough that they don't miss CarPlay, especially once Spotify and Google Maps are integrated.
- One sharp comment argues automakers are resisting CarPlay not for UX reasons but to protect a data business. CarPlay blocks manufacturers from tracking what music and apps you use, and that telemetry has real commercial value.
- Consistency across rental cars and shared vehicles comes up repeatedly as an underrated benefit: plug in, get your own interface, no setup required.
- A few commenters push back on the whole premise, saying Bluetooth and a phone mount solve the same problems, but most replies treat that view as missing the point.
Someone translated the entire Rust compiler into 46 million lines of C, then got it to compile itself.
What the article says
- Three years and fourteen attempts produced cilly, a Rust compiler backend that outputs C instead of machine code, letting any platform with a C compiler run Rust.
- The author used it to translate the full Rust compiler source into C, which you can then build with GCC and make. The result is a working rustc.
- The main goal is obscure hardware that C compilers reach but LLVM does not, such as old Plan 9 machines. The author already got a Hello World running on an x86 Plan 9 VM this way.
- Cilly probes the target C compiler for what it supports before generating code, so it adapts to weird or ancient toolchains rather than assuming standard behavior.
- The full toolchain is not yet released publicly. The author is juggling a new job, a thesis, and a blender-related hand injury.
What HN is saying
- The Guix project already bootstraps Rust through a handwritten C++ compiler called mrustc, and someone confirmed it produces byte-identical output to the official compiler, clearing it of backdoors.
- A commenter pointed out the deeper appeal: cilly could become a second bootstrapping path for Rust, making it easier to build a trustworthy compiler chain from near-scratch.
- One skeptic noted the generated C is more like a compiler artifact than readable source, so it may not meet the standards bootstrappable-software projects actually require.
- The thread dissolved into gentle jokes about the irony of rewriting Rust in C, inverting years of "rewrite it in Rust" memes.
- Several people were more curious about the blender incident than the compiler.
A petition to protect your right to run AI models locally, before laws quietly make it illegal.
What the article says
- The site asks people to sign a petition and call their state legislators supporting safe-harbor language for owning, modifying, and running AI models locally.
- The core ask is proactive, not reactive: enshrine the right before any specific bill closes it off.
- The article body is sparse, inferred from description and comments. The site is more a mobilization page than a policy brief.
What HN is saying
- Commenters can't find what specific bills the campaign is fighting, making it feel vague and hard to act on.
- The sharpest concern is that big AI companies will lobby to restrict open-weight models to protect their cloud revenue, the way Microsoft once tried to push back against Linux.
- One commenter points out that you technically already have the right today, but worries the real threat is a future law making local model possession a felony.
- Several people agree that cloud-only AI feels fragile and invasive, and that consumer hardware is nearly powerful enough to run capable models locally anyway.
Linux silently stopped wiping disk encryption keys on suspend for over two years.
What the article says
- A kernel refactoring in Linux 6.9 broke the tool that wipes your LUKS disk encryption key when you suspend your laptop, and nobody noticed for two years.
- The fix is one line, but the bug persisted because the failure was silent: everything appeared to work, the key just stayed in memory when it was supposed to be gone.
- Anyone who seized your powered-but-suspended laptop could have dumped RAM and recovered the encryption key, bypassing full-disk encryption entirely.
- The bug was found via careful git bisection, and a NixOS integration test now guards against it coming back.
What HN is saying
- Most commenters note this only affected distros using the optional cryptsetup-suspend addon, mostly Debian and its derivatives, not stock Linux installs.
- The sharpest thread is about whether Rust or better tooling would have caught this: the consensus is no, because the bug was a violated high-level system invariant, not a memory-safety error any compiler can see.
- Several people point out that hibernation is the only solid defense against cold-boot attacks, since it powers off RAM entirely rather than just hoping keys get wiped.
- A commenter cuts through the alarm: if you never re-enter your passphrase after waking from sleep, your key was always in memory anyway, so this only matters if you were explicitly using the suspend addon.
A 2017 essay on why every simple task hides an unexpected jungle of details, and why that gets smart people stuck.
What the article says
- Building basement stairs looks trivial until you hit warped lumber, brackets that won't seat straight, and screws that poke through the tread. Each fix reveals another snag.
- This isn't a carpentry quirk. Boiling water has the same property: superheating, nucleation, vessel material, all quietly waiting to bite you.
- Experts forget this because once you've absorbed the details they become invisible. What looks obvious to them looks complicated to beginners, and vice versa.
- The danger is getting intellectually stuck: the key detail you're missing is also the detail you can't see, because you don't yet know to look for it.
- The fix is deliberate attention. Slow down, notice what you normally skip, and when someone smart disagrees with you, find out which detail they're weighing that you aren't.
What HN is saying
- Readers pile on with their own crooked-wall moments: a perfectly square bookshelf that didn't fit a tilted wall, bulkhead doors built square for a parallelogram frame.
- Programmers debate whether code is special. One view: programming is uniquely tractable because you can snapshot, undo, and test. The counter: until the next OS update breaks everything.
- Several comments land on the sharpest implication: the sentence 'you could be intellectually stuck right now, with the evidence in front of your face' unsettled people most.
- One commenter notes the flip side: once you've truly learned a domain, you gain the power to ignore the details that don't matter, which is its own kind of freedom.
The Zachtronics puzzle game that made assembly feel fun, not terrifying.
What the article says
- Exapunks casts you as a hacker in 1997 who writes programs for mobile agents called EXAs to infiltrate networks, trade for medicine, and survive.
- The game ships with a physical-style zine called Trash World News as your manual, a loving nod to magazines like 2600.
- It sits in Zachtronics' broader catalog of programming puzzle games, alongside TIS-100 and Shenzhen I/O, all built around writing real assembly-like code to solve problems.
- Zachtronics has since closed, but the creator is still making games under a new studio called Coincidence Games.
What HN is saying
- Multiple people say these games gave them the confidence to stop fearing assembly and low-level code, often crediting them with shaping their careers.
- There is genuine debate over which Zachtronics game to start with: Opus Magnum gets recommended most as the friendliest entry point.
- One commenter shares a half-built game idea that sounds like Exapunks meets Starcraft, played over a raw TCP socket in Forth, and the replies are enthusiastic.
- A dissenting voice calls the whole genre pointless busywork that teaches nothing practical, and the thread mostly ignores them.
The Trump administration just banned the math that keeps Census data private.
What the article says
- A Commerce Department directive issued in June bars noise infusion and differential privacy, the core techniques statisticians have used since the 1990s to share granular data without exposing individuals.
- The only permitted alternative is coarsening, which the post proves is worse: a simple algebra example shows how combining several coarsened statistics lets anyone reconstruct the exact employee count at every business in a county.
- The stated motive from Project 2025 architects is explicit: differential privacy makes it impossible to identify citizenship status from census records, even if a citizenship question is added.
- Civil servants now face a bind: comply with the directive or comply with the Census Act, which makes it a crime to publish data that can identify any individual. The likely outcome is less data or data that is legally but not practically private.
- The post is cosigned by Cynthia Dwork and other leading researchers, and calls for contacting Congress and archiving methodology pages that are already disappearing from government websites.
What HN is saying
- Most commenters accept the technical argument without dispute: coarsening was known to be weaker than noise infusion long before this directive.
- The sharpest disagreement is over whether calling your legislator does any good. One long commenter argues the only lever that works is primarying incumbents, while an Australian respondent says direct contact with MPs has worked for them.
- Several people note the political motive openly stated in the CRA explainer: killing differential privacy is specifically about making citizenship data traceable.
- A commenter with genuine nuance flags that some statisticians did criticize the 2020 Census's use of differential privacy for distorting apportionment data, so this is not entirely a settled scientific consensus vs. politics story.
- The thread drifts into broader campaign finance frustration after one commenter ties the directive to Project 2025 and a recent Supreme Court ruling removing donation limits.
PeerTube is a working, federated YouTube alternative with no ads and no algorithm.
What the article says
- PeerTube lets anyone host a video server that federates with others via ActivityPub, so viewers can follow channels across instances.
- Bandwidth costs are partly offloaded to viewers through peer-to-peer sharing in the browser, which helps small hosts survive traffic spikes.
- There is no ad system and no recommendation engine. Creators can add a donation link, but the platform does not handle payments.
- You can follow channels from Mastodon or subscribe via RSS without creating an account on the host instance.
What HN is saying
- The central debate is monetization: a working YouTuber argues you cannot sustain quality production on donations alone, while others say PeerTube is simply not for that use case.
- Several people see a clear niche: KDE, Blender, universities, and conference organizers do not need ad revenue and would benefit from owning their distribution.
- One commenter cuts through the noise by noting YouTube handles discovery, monetization, hosting, and delivery while PeerTube only handles the last two.
- A sharp observation: the $500 to $1,000 per video that YouTube pays smaller creators may be precisely the minimum Google calculated to keep them from building alternatives.