Hacker News
1Hormuz Minesweeper – Are you tired of winning?
- This appears to be instructions for a Minesweeper-style game where players use left-click to reveal tiles, right-click to flag potential mines, and double-click to chord. The game has a unique twist where mines only appear on water tiles rather than land.
Political commentary and satire
Users debate the appropriateness of making light of serious geopolitical situations through gaming satire. Discussion includes Trump references ("MAGA"), concerns about making fun of deadly conflicts, and whether comedy/satire serves as valid political critique or trivializes human suffering.Game mechanics and technical issues
Players discuss missing minesweeper features like double-clicking/chording, mobile compatibility problems on iPad/iPhone, alternative versions with ship navigation, and suggestions for improvements like hex tiles and nuclear options.Real situation in Hormuz Strait
Users analyze the actual maritime situation, debating whether it's truly a minefield vs. selective targeting, ship traffic data, insurance coverage, AIS tracking being turned off, and Iran's strategy of uncertainty to deter shipping.
2Tinybox – A powerful computer for deep learning
- Tiny corp develops tinygrad, a simple neural network framework that breaks down complex networks into three operation types. They're hiring software engineers and sell tinybox computers for deep learning, ranging from $12,000 to $10M with various GPU configurations.
Performance and feasibility of 120B models
Users debate whether the red v2 can actually run 120B parameter models with its 64GB VRAM, with many arguing heavy quantization would be needed, causing performance issues and context limitations around 4k tokens.Pricing and value proposition concerns
Discussion centers on whether these AI boxes are overpriced for their target market - too expensive for hobbyists but companies needing scale can build equivalent systems themselves for much less money.Business model and customer approach criticism
Users criticize the company's rigid no-customization policy and website-only ordering for expensive B2B equipment, calling the approach arrogant and impractical for enterprise customers.
3The three pillars of JavaScript bloat
- The e18e community has been addressing "dependency bloat" in npm packages through cleanup initiatives. The article identifies three main types of bloat: packages supporting very old runtime engines with safety features that most developers don't need, atomic architecture that breaks code into unnecessarily small packages, and ponyfills for features that are now natively supported but haven't been removed. These compatibility layers and micro-packages increase supply chain risks, bandwidth costs, and maintenance overhead while providing little benefit to most modern development environments.
Dependency-free JavaScript development
Developers advocate for vanilla JS over frameworks, citing better maintainability and performance. Many share positive experiences building apps without dependencies, using web components and modern browser APIs. Some create tutorial resources promoting this approach.Micro-packages and ecosystem bloat
Discussion focuses on packages with just a few lines of code creating massive dependency trees. Criticism targets specific maintainers who allegedly create many tiny packages for financial gain through download counts and sponsorships.Legacy browser support driving bloat
Developers debate supporting extremely old JavaScript engines (ES3/ES5). Many argue this creates unnecessary polyfills and transpilation overhead, suggesting teams should update compilation targets and drop ancient browser support.
4Ubuntu 26.04 Ends 46 Years of Silent sudo Passwords
- Ubuntu 26.04 LTS will change a 46-year tradition by displaying asterisks when typing sudo passwords, replacing the historically silent prompt. This switch from the classic C-based sudo to the new Rust-based sudo-rs aims to improve user experience, though it has sparked debate over security implications. Users can revert to the silent behavior through configuration if desired.
User experience improvements
Users frequently struggle with silent password prompts, especially over high-latency SSH connections where they can't tell if keystrokes register. Many share stories of typing passwords in wrong places or being confused by lack of feedback. This change aligns sudo with GUI password fields that already show asterisks.Security implications
Debate over whether showing password length reduces security. Some argue it's minimal risk since shoulder surfing could already reveal length, while others worry about livestreams and recordings. Most agree that for strong passwords, length disclosure isn't meaningful, and usability improvements may enhance overall security.Historical context and alternatives
Discussion of why silent prompts existed (shared terminals, teletypes) and creative alternatives like Lotus Notes' random hieroglyphs or spinning indicators that don't reveal length. Some suggest configuration options to preserve old behavior for those who prefer it.
5The future of version control
- Manyana is a project demonstrating CRDT-based version control that eliminates merge conflicts by always succeeding merges while providing clearer conflict presentation. Unlike traditional systems showing opaque conflict blobs, it shows exactly what each party did with detailed markers. The approach enables permanent line ordering, informative non-blocking conflicts, and rebase without destroying history.
Merge conflict handling and tools
Discussion centers on improving merge conflict resolution through better tools like p4merge's 4-pane interface, Git's diff3/zdiff3 config options, and various IDE merge tools. Users debate whether switching VCS is necessary or if better tooling around Git suffices for conflict resolution.CRDTs effectiveness for version control
Debate over whether CRDTs are appropriate for version control, with arguments that conflicts are fundamentally necessary in VCS to catch semantic issues. Some see CRDTs as producing garbage code, while others argue they can provide better merge semantics and conflict handling.Git workflow preferences and alternatives
Strong opinions on Git workflows emerge, with some advocating for rebase-only approaches avoiding merge commits, while others discuss alternatives like Jujutsu and Pijul. Discussion includes team scaling considerations and the role of AI in conflict resolution.
6Cloudflare flags archive.today as "C&C/Botnet"; no longer resolves via 1.1.1.2
Archive.today DDoS attack
Archive.today has been conducting an ongoing DDoS attack against gyrovague.com for over two months, using JavaScript on their site to make visitors' browsers spam the target with requests. This has resulted in Cloudflare categorizing archive.today domains as malware/botnet.Cloudflare DNS filtering and blocking
Cloudflare's malware-blocking DNS resolver (1.1.1.2) now blocks archive.today domains, categorizing them as "CIPA Filter", "Reference", "Command and Control & Botnet", and "DNS Tunneling". The blocking appears related to the DDoS behavior.Doxxing allegations and privacy concerns
Debate over whether blogger Jani Patokallio's investigation into archive.today's ownership constitutes doxxing. Archive.today appears to be retaliating against the blog post, with some arguing the DDoS is worse than the research.
7Professional video editing, right in the browser with WebGPU and WASM
- A browser-based professional video editor offering GPU-accelerated rendering, multi-track timeline, keyframe animation, real-time effects, and zero installation requirements. Built with WebGPU and Rust/WASM for native-like performance while keeping all media files local.
Browser compatibility and WebGPU support
Users discuss WebGPU browser support limitations, particularly Firefox on Linux/Android. Some report crashes and compatibility issues, while others successfully use Chromium-based browsers like Brave without problems.Licensing and open source concerns
Community debate over the project's licensing choice, with users pointing out the initial non-commercial license wasn't truly open source. The developer responded by changing to ELv2 license after feedback.Feature limitations and workflow issues
Users report basic video editing limitations like inability to add separate audio tracks or properly handle MP3 files. Discussions compare the tool to desktop alternatives like DaVinci Resolve and highlight missing features.
8Project Nomad – Knowledge That Never Goes Offline
- Project NOMAD is a free, open-source offline server that provides Wikipedia, AI models, maps, and educational content without internet. It runs on personal hardware, offering emergency preparedness and off-grid capabilities with tools like offline Wikipedia, local AI assistants, OpenStreetMap navigation, and Khan Academy courses.
Preparedness and doomsday prepping
Discussion about the stigma around "preppers" vs reasonable preparedness, with users sharing perspectives on maintaining offline resources, local documentation, and knowledge bases for internet outages or government censorship scenarios.Technical implementation and alternatives
Technical discussion about ZIM file format, Kiwix, compression methods, and alternative approaches to storing offline knowledge bases, including suggestions for better compression and hardware platforms like Raspberry Pi.Offline knowledge and past internet experience
Users sharing experiences from dial-up internet days and current practices of maintaining offline documentation, local wikis, and backup resources for programming, emphasizing that network access shouldn't be taken for granted.
9Windows native app development is a mess
- A long-time Windows developer attempted to create a simple utility app to black out monitors during gaming, but found the modern Windows development ecosystem to be a complete mess. Despite Microsoft's numerous UI framework iterations over the years (Win32, MFC, WinForms, WPF, WinRT, UWP, WinUI 3), basic functionality still requires dropping down to old Win32 APIs through painful workarounds. Distribution is equally problematic, with expensive code signing requirements and poor unsigned app experiences. The developer concludes that this fragmented, half-finished state explains why developers choose Electron over native Windows development.
Win32 API advocacy
Many developers strongly recommend sticking with the classic Win32 API for Windows development, citing its exceptional backward compatibility, small executable sizes (potentially under 8KB), and stability. They argue it's better than newer Microsoft frameworks despite memory safety concerns.Microsoft framework mess
Discussion highlights the confusing state of Microsoft's UI frameworks (WinUI, UWP, WinAppSDK) and recommends avoiding them in favor of older, stable options like Win32, MFC, WinForms, or WPF. Microsoft's tendency to abandon frameworks halfway through development is a major concern.Native vs web-based development
Debate over whether to use native Windows development or web technologies like Electron/Tauri. While native apps offer better performance and user experience, the complexity of Windows development often pushes developers toward web-based solutions despite their drawbacks.
10Flash-MoE: Running a 397B Parameter Model on a Laptop
- Flash-MoE is a pure C/Metal inference engine that runs a 397 billion parameter Mixture-of-Experts model on a MacBook Pro with 48GB RAM at over 4 tokens per second. The system streams the 209GB model from SSD using custom Metal shaders and achieves production-quality output including tool calling without Python frameworks.
Hardware requirements and accessibility
Discussion about whether high-end MacBooks with 48-128GB RAM can be considered "consumer hardware" and concerns that clickbait titles make local LLM seem accessible to average users when the required hardware costs $3000+ and is out of reach for most people.Quality degradation from extreme optimization
Debate over whether 2-bit quantization and reducing experts from 10 to 4 severely compromises model quality, with users reporting that 2-bit quants produce broken JSON output and lose reasoning ability, making the approach more of a proof-of-concept than practical solution.Performance benchmarks and real-world viability
Users sharing actual performance data showing 5-20 tokens/second depending on context length, with some arguing the slow speeds and quality issues make it unsuitable for real work despite impressive technical achievement of running 397B model on consumer hardware.