Type in a postcode and read Wikipedia's strangest stories from your own streets.
What the article says
- Local Wikipedia shows only the Wikipedia articles about places within a half hour walk of you.
- Enter a town, landmark or postcode and it pulls up nearby pages, all taken straight from Wikipedia.
- If your area is sparse, it widens the circle to about six miles. If it is crowded, it keeps the closest few hundred.
- Your location stays in your browser, and shared links don't include it.
What HN is saying
- Most people agree that facts feel more interesting when they happen on your doorstep, even a random fire station or old plane crash. thread ↗
- Several pointed out it resembles Wikipedia's built-in Nearby page, but it adds the useful trick of typing in a place instead of sharing your location. thread ↗
- Rural and non-English readers say results are thin or full of roads and housing estates, and many want other language editions or a configurable radius. thread ↗
- A pedant noted it is really a three kilometre circle, not a true half hour walk, which would need walking routes. thread ↗
- People shared similar projects, like location-unlocked audio stories and a game that turns nearby Wikipedia places into quests. thread ↗
A dev builds his own container runtime and finds Docker is mostly Linux tricks in a trenchcoat.
What the article says
- Containers are not real. They are ordinary processes with restrictions, and you can find them in /proc.
- There is no single Linux container feature. Docker stitches together namespaces, cgroups and some networking magic.
- An image is just a folder, and Docker swaps it in as the process's root directory.
- Go is an awkward fit for a runtime. It has no real fork, so runc leans on a C module that runs first.
- Docker has long needed root, and rootless networking means rebuilding the network stack in userspace.
What HN is saying
- Docker's creator showed up to say hello and take questions. He later plugged his new project, Dagger. thread ↗
- Commenters split on what made Docker win. Some say the Dockerfile and registry, others say momentum and ease of use. One researcher points out the layered image idea appeared in academic papers years earlier. thread ↗
- Several readers found nothing weird in the list. One says the real oddity is the client and server design, which makes permissions painful. thread ↗
- A former Docker engineer notes the key part of runc is actually C. Another replies that by language stats it is still mostly a Go project. thread ↗
- Fans of Podman praise its rootless, daemonless approach. One user also points out that containers never became secure enough for multitenant use without a VM underneath. thread ↗
- Early Go bugs had real costs. Goroutines hopping between threads quietly broke network namespace cleanup. thread ↗
A big Chinese open-weight model rivals the best US ones, with its hacking skills deliberately trimmed
What the article says
- Kimi K3.1 Pro is Moonshot's new open-weight model, with a huge context window and a faster Highspeed version. Full weights are promised by November 4.
- It is pitched as close to the top proprietary models on coding, 3D modelling and finance work, with a sharp drop in made-up facts.
- The cybersecurity section is the surprise. They stripped hacking-related data from training and say the public model stays below what they could build.
- They argue that letting this skill spread slowly helps defenders audit old code.
What HN is saying
- Many paid subscribers say the usage limits are brutal. A single task can burn through the five hour allowance, so people lean on the API or cheaper rivals instead. thread ↗
- Independent rankings put it near the top, and one commenter reads that as China being only weeks behind. Others reply that the US labs are holding back stronger unreleased models. thread ↗
- The cyber data removal split people. Some say models start from zero on skills they never saw in training. Others doubt that cutting data really works, or suspect it is a smart shield against political pressure. thread ↗
- A cheap Pac-Man coding test had it tying Sonnet 5.5 at lower cost and time. thread ↗
- Fans like that it pushes back bluntly instead of flattering you. One worries that an annoyed attitude could be risky in agents that use tools. thread ↗
A small security firm just showed iPhones being fully taken over from a PDF, with no patch yet.
What the article says
- Calif showed four live demos at the Hexacon conference in Paris, all aimed at the iPhone.
- The first hits the Signal app with one click. The second does the same through iMessage.
- The third gets a root shell on the latest iOS. The fourth is a persistent break that survives restarts.
- The page is mostly videos with little text. Technical details are promised in the coming weeks, after the bugs are reported to Apple.
What HN is saying
- Many asked why PDFs keep getting phones hacked. The answer: the format does far too much, from scripting to ancient image codecs, so the attack surface is huge. thread ↗
- Lockdown mode would probably block the PDF route, but commenters say it is too restrictive for most people to live with. thread ↗
- Persistence is the standout. A truly untethered jailbreak has been rare for about ten years, and people think the flaw sits early in boot, before first unlock. thread ↗
- A split on jailbreaking itself: some want root on their own phone, others only object to a stranger's PDF giving someone else root. thread ↗
- Thai Duong of Calif joined the thread. He says they take no bounties, sell nothing, and have reported close to 100 bugs to Apple this year. thread ↗
A man's HN post may have blown a secret PHP security fix. Who is to blame?
What the article says
- The author says he posted about a serious PHP remote code execution flaw on Hacker News, and is now apologising for it.
- He says nobody told him an embargo existed. His contact said there was none, and her own source said the same.
- Reports suggest his post helped push the planned announcement forward, and he says he has barely slept since.
- He takes full responsibility for posting without checking the facts. (Inferred from the title and comments, since the post body was not captured.)
What HN is saying
- Most commenters say he did nothing wrong. An embargo binds the people who agreed to it, and someone earlier in the chain dropped the warning. thread ↗
- A sharper view: a post-mortem should blame the process, not a person, or people will hide mistakes next time. thread ↗
- Pushback: if you hear about an unannounced flaw, you should still report it privately first, and he did know an announcement was coming. thread ↗
- Some mock the phrase "take full responsibility" as empty. Others reply it simply means the buck stops with him. thread ↗
- A bleaker take: with AI finding bugs so fast, embargoes may matter less, since anyone could find the same flaw. thread ↗