Published on April 26, 2026 at 18:00 CEST (UTC+2)
Asahi Linux Progress Linux 7.0 (330 points by elisaado)
The Asahi Linux project releases its progress report for Linux 7.0, marking the end of the 6.x kernel series. The report highlights a long-overdue update to the installer, which had not been refreshed since June 2024 due to a cumbersome manual release process. The team automated the installer pipeline to avoid future delays, reflecting ongoing upstreaming efforts and improvements in Apple Silicon Linux support.
Statecharts: hierarchical state machines (173 points by sph)
Statecharts provides an introduction to statecharts, describing them as a visual formalism for complex systems that extend finite state machines. The approach helps avoid state explosion by adding hierarchy, concurrency, and communication capabilities. The site emphasizes that statecharts make behavior easier to understand, test, and modify independently of the components that use them.
Why SWE-bench Verified no longer measures frontier coding capabilities (58 points by kmdupree)
OpenAI explains why they no longer use SWE-bench Verified as a benchmark for frontier coding capabilities. The article likely argues that the benchmark has become saturated or fails to measure genuine improvements in AI-assisted software development, signaling the need for more challenging and realistic evaluation methods.
Amateur armed with ChatGPT solves an Erdős problem (565 points by pr337h4m)
A 23-year-old amateur with no advanced math training used a ChatGPT Pro subscription to solve a 60-year-old Erdős problem. The AI suggested a novel method that human mathematicians had not thought of, though experts caution that AI solutions to such problems can vary in originality and significance. This case demonstrates how large language models can assist in mathematical discovery even without deep domain expertise.
The Nintendo Switch Switch (2019) (39 points by zdw)
A hacker describes turning a Nintendo Switch console into an actual network switch by installing Linux via the switchroot project. They connected a USB Ethernet dongle and verified the device appeared correctly with lsusb. The project is a proof-of-concept showing how the Switch’s hardware can be repurposed thanks to community homebrew and Linux support.
Show HN: Turning a Gaussian Splat into a videogame (112 points by yak32)
PlayCanvas demonstrates turning a Gaussian splat (a cloud of oriented blobs used for photorealistic 3D scenes) into a playable first-person shooter videogame. The process involves adding physics colliders, baked lighting probes, a Recast navmesh, and NPC behavior on top of the splat. The entire experience runs in a browser, showcasing how splat-based rendering can be combined with traditional game mechanics.
Why has there been so little progress on Alzheimer's disease? (323 points by chiefalchemist)
The Freakonomics podcast episode questions why progress on Alzheimer’s disease has been slow despite decades of research. It likely explores failures in drug trials, the complexity of the disease, and potential misdirection in scientific approaches, reflecting broader challenges in biomedical research.
The West forgot how to make things, now it’s forgetting how to code (801 points by milkglass)
The article argues that Western industry has lost the ability to manufacture complex physical goods (e.g., Stinger missiles) and is now losing the ability to write good software. The author draws parallels between declining hardware production skills and the rise of AI-generated code, warning that over-reliance on automation may erode deep technical understanding and craftsmanship in software engineering.
Tell HN: An app is silently installing itself on my iPhone every day (409 points by -x-)
A user reports that the Headspace app silently reinstalls itself on their iPhone every day despite automatic downloads being turned off. Multiple users on Reddit confirm the issue across various iPhone models. The cause might be a system bug or a conflict with daily reminder notifications triggering a reinstall, though Apple has not yet officially addressed it.
USB Cheat Sheet (2022) (414 points by gwerbret)
Fabien Sanglard provides a comprehensive USB cheat sheet covering marketing names, signal speeds, cable lengths, and encoding schemes across USB 1.1 through USB4. The table clarifies confusing naming conventions and gives real-world effective throughput estimates. It serves as a handy reference for developers and engineers working with USB hardware.
AI as a co-scientist for mathematical discovery – The article about an amateur solving an Erdős problem with ChatGPT shows that large language models can propose genuinely novel solution methods, even for problems that stumped experts. This trend suggests AI will increasingly augment human reasoning in fields like mathematics and theoretical science, lowering the barrier for outsiders to contribute. The implication is that researchers should treat AI as a collaborator, not just a coding assistant, and that AI benchmarks should include open-ended creative problem-solving tasks.
Benchmark saturation and the need for harder evaluations – OpenAI’s decision to stop using SWE-bench Verified for frontier coding capabilities indicates that existing benchmarks no longer discriminate among top AI models. This mirrors the common “AI benchmark treadmill” where performance quickly plateaus. The trend demands the creation of more dynamic, real-world, and adversarial evaluation suites that measure true software engineering skill (e.g., system design, debugging, or long-term maintenance) rather than static patches.
Gaussian splatting + AI unlocks new game development paradigms – The PlayCanvas demo of turning a Gaussian splat into a videogame represents a convergence of neural rendering (a ML-based 3D representation) with traditional gameplay code. As splatting becomes cheaper and more realistic, game developers can bypass manual 3D modeling entirely, sourcing scenes from real-world scans. The AI/ML trend here is the blending of learned representations with game physics and AI-driven NPCs, reducing art costs while preserving photorealism.
AI-assisted code generation threatens deep coding skills – The essay on “forgetting how to code” warns that reliance on LLM-generated code may erode the foundational understanding needed to debug, optimize, or reason about complex systems. This is a critical trend: as AI coding tools become ubiquitous, the software industry must invest in education and practices that preserve human expertise, otherwise we risk a skills hollowing similar to that seen in physical manufacturing. A key takeaway is that companies should pair AI coding with rigorous code review and deep-dive training.
Silent app installations highlight AI-driven app store and OS bugs – The iPhone Headspace bug (and similar past issues) shows how automated download/reinstall systems can malfunction. While not directly AI/ML, the trend is that as AI increasingly controls app recommendations and installations on devices, such bugs can become more frequent and harder to diagnose. For AI/ML practitioners, this underscores the need for robust testing, rollback mechanisms, and transparency in automated decision systems that affect user devices.
Statecharts as a formal method for AI behavior design – The resurgence of statecharts (hierarchical state machines) is relevant to AI/ML because they provide a structured way to model complex agent behaviors, especially in robotics and game AI. Unlike end-to-end neural policies, statecharts offer interpretability and testability. The trend is a move toward hybrid systems where statecharts handle high-level logic and ML handles perception or low-level control, balancing predictability with learning.
Open-source AI infrastructure benefits from long-term maintainers – The Asahi Linux installer story illustrates how manual release processes lead to neglect, even in technically strong projects. For AI/ML projects that depend on open-source libraries, the trend is that automated CI/CD pipelines and community contribution models are essential for sustainability. Without continuous investment, even foundational AI tools can stagnate, hindering reproducibility and progress. A practical takeaway is to prioritize infrastructure automation in ML research groups and open-source AI projects.
Analysis generated by deepseek-reasoner