Published on May 10, 2026 at 06:00 CEST (UTC+2)
Show HN: Building a web server in assembly to give my life (a lack of) meaning (58 points by imtomt)
This project is a web server written entirely in ARM64 assembly for macOS, using only system calls and no libc. It follows a fork-per-connection model and is designed to be portable, though it currently requires Apple Silicon. The author humorously frames the endeavor as a search for meaning through low-level systems programming.
Bun's experimental Rust rewrite hits 99.8% test compatibility on Linux x64 glibc (458 points by heldrida)
Bun, the JavaScript runtime, has reached 99.8% test compatibility on Linux x64 glibc with its experimental Rust rewrite. The tweet by creator Jarred Sumner highlights significant progress in migrating from Zig to Rust for core components. This milestone suggests Bun is moving closer to a production-ready Rust-based implementation.
Casio S100X Japanese Lacquer Edition (JP Page Only) (27 points by dr_kiszonka)
Casio has released a premium calculator, the S100X Japanese Lacquer Edition, featuring a lacquered finish and high-end design. It is available only on the Japanese Casio website and targets collectors or enthusiasts of luxury stationery. The calculator represents a niche intersection of traditional craftsmanship and modern electronics.
Internet Archive Switzerland (561 points by hggh)
The Internet Archive is launching a new Swiss foundation based in St. Gallen to preserve endangered archives and collect generative AI models. The initiative partners with the University of St. Gallen’s Gen AI Archive project to archive AI models, addressing an emerging preservation need. The location was chosen for its thousand-year archival tradition and academic environment.
Gemini API File Search is now multimodal (4 points by gmays)
Google DeepMind’s Gemini API File Search now supports multimodal retrieval-augmented generation (RAG), processing images and text together. It introduces custom metadata and page-level citations for better grounding and transparency. The update leverages the Gemini Embedding 2 model to give applications “photographic memory” for visual assets.
I’ve banned query strings (304 points by susam)
The author Chris Morgan explains why they have banned query strings from their web projects, arguing they cause caching, SEO, and analytics problems. The post likely proposes alternative URL structures or server-side solutions. It reflects a strong opinion on web architecture and URL hygiene.
I'm writing a history of Visual Basic, Chapter 1 is up (49 points by speckx)
A multi‑chapter history of Visual Basic has begun, with Chapter 1 covering the language’s origins from 1964 to 1992. It focuses on Microsoft’s BASIC dynasty, the acquisition of a California developer for the visual frontend, and Bill Gates’ early pitch. The author aims to tell the human story behind VB, not just the technical or corporate narrative.
The Serial TTL connector we deserve (62 points by kohlschuetter)
This article discusses the common practice of using messy Dupont jumper wires for UART TTL connections on embedded devices. The author proposes a more robust, reliable connector design for serial debugging. It appeals to hardware hackers and those who work with low‑level board recovery.
Local privilege escalation via execve() (107 points by Deeg9rie9usi)
A security advisory for FreeBSD reveals a local privilege escalation vulnerability in the execve() system call. An operator precedence bug in the kernel can cause a buffer overflow, allowing an unprivileged user to overwrite adjacent argument buffers and gain superuser privileges. All supported FreeBSD versions are affected, and patches have been released.
Show HN: Rust but Lisp (96 points by thatxliner)
This project, “Rust but Lisp,” is a transparent s‑expression frontend that compiles directly to Rust, providing Lisp macros on top of Rust semantics. It has no runtime or GC and currently supports a subset of Rust syntax. The author frames it as an experimental exploration of language interoperation rather than a production tool.
Multimodal Retrieval-Augmented Generation is becoming mainstream
Google’s Gemini API adding multimodal file search (text + images) and page-level citations lowers the barrier for building verifiable, grounded AI agents. This trend indicates that next‑generation RAG systems will seamlessly handle diverse data types, making applications like visual asset retrieval and document understanding far more practical.
Preserving AI models is an emerging critical discipline
The Internet Archive Switzerland’s Gen AI Archive project highlights a need to archive not just datasets but also model weights, architectures, and training configurations. As generative AI proliferates, the risk of model loss or historical amnesia grows—foundations and archives will likely invest heavily in this space, creating new standards and infrastructure.
Rust is increasingly the go‑to language for performance-critical AI infrastructure
Bun’s migration from Zig to Rust for its runtime (now at 99.8% compatibility) and the existence of projects like “Rust but Lisp” show Rust’s dominance in systems-level AI tooling (e.g., tokenizers, inference engines, runtimes). This trend promises faster, safer, and more portable foundations for AI applications, especially on the edge.
Low‑level systems programming is being revisited for AI efficiency
Writing a web server in ARM64 assembly (Article 1) is a niche hobby project, but it reflects a broader push to extract maximum performance from hardware—essential for AI workloads. As models grow and inference costs rise, hand‑tuned assembly or close‑to‑metal optimizations may become relevant for specific bottlenecks (e.g., matrix multiplication, kernel launching).
Language interoperation and metaprogramming are hot areas for experimentation
The “Rust but Lisp” transpiler combines Lisp macros with Rust’s semantics, showing developer interest in flexible compile‑time metaprogramming for AI/ML pipelines (e.g., custom DSLs for model definitions or gradient computation). Such experiments could influence future frameworks that blend expressive syntax with high performance.
Security vulnerabilities in OS kernels remain a blind spot for AI‑deployed systems
The FreeBSD execve() privilege escalation (CVE-2026-7270) is a reminder that kernel bugs can undermine the security of cloud and edge AI infrastructure. As AI workloads move to bare‑metal or container‑based deployments, hardening the kernel (especially memory safety) becomes a critical part of MLOps and model serving.
Analysis generated by deepseek-reasoner