digest · hn front page

Sat, Sep 19, 2026

53 stories · 0 worth reading in full

Filter stories
  1. AI-generated posters don’t have to be horrible john.hartnup.uk

    AI-generated posters often suffer from a repetitive, generic aesthetic because users rely on default settings.

  2. I built non-autoregressive decision models with RL a year ago laya.convaiinnovations.com

    Modern AI pipelines often waste resources using slow, generative LLMs for simple, structured decisions that only require a reflex-like response.

  3. How to Write with an LLM sockpuppet.org

    Large Language Models should be used as copyeditors rather than ghostwriters to prevent writing from becoming generic and artificial.

  4. Two parallel neural ectoderm progenitors contribute to the developing brain med.stanford.edu

    Stanford Medicine researchers discovered that the human brain consists of two distinct organs that evolved independently and arise from different progenitor cells.

  5. If math is more than proof, we need to better celebrate the rest of it terrytao.wordpress.com

    As AI becomes capable of generating proofs without human understanding, the mathematical community must shift its rewards and academic credit from the mere production of proofs to the creation of 'motivated explanations.' By formally recognizing work that clarifies how a…

  6. GPT-6 Astra Solves a WWI German Radio Cipher prinzai.com

    GPT-6 Astra successfully decoded a previously unsolved World War I German radio message by identifying the correct encryption key and verifying the result against historical records.

  7. San Francisco Onion Futures Company onionfutures.com

    The San Francisco Onion Futures Company provides a way for individuals to buy private contracts for the physical delivery of yellow onions in the future.

  8. The first new cat species discovered in 100 years nationalgeographic.com

    Researchers have identified a new species of small wild cat, Leopardus tilcayo, marking the first new felid species discovered in over a century.

  9. English: A vs. An redblobgames.com

    Choosing between 'a' and 'an' in English depends on the spoken sound of the following word rather than its first letter.

  10. I think you should almost never use AI to write erichgrunewald.substack.com

    Delegating substantive writing to AI undermines the critical thinking process because the act of writing is where ideas are refined and flaws in logic are discovered.

  11. Brood War Bench bw.swerdlow.dev

    Current large language models cannot play StarCraft: Brood War beyond a beginner level, as they struggle with real-time decision making and complex macro-management.

  12. The Secret Life of Circuits blog.coredump.cx

    The author has released a new hardcover book titled The Secret Life of Circuits, which aims to provide a practical and intuitive guide to electronics.

  13. What Zig felt like, coming from Rust besok.github.io

    Transitioning from Rust to Zig reveals a shift from a high-level, functional-influenced systems language to a more explicit, imperative one.

  14. Tin: full-text search for Postgres planetscale.com

    TIN is a new full-text search extension for Postgres designed to outperform existing options by using Postgres' own physical tuple identifiers (ctids) as the native index format.

  15. Measure internet censorship ooni.org

    OONI Probe is an open-source tool designed to collect evidence of internet censorship by allowing users to perform network tests.

  16. How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip spectrum.ieee.org

    OpenAI significantly accelerated the development of its Jalapeño AI accelerator chip by integrating large language models into the design process.

  17. How Hacker News ranking works: scoring, controversy, and penalties (2013) righto.com

    Hacker News rankings are not a pure meritocracy of upvotes but are heavily influenced by a hidden system of penalties.

  18. Microsoft director: AI scraping 'the largest theft of labor in human history' tomshardware.com

    The New York Times lawsuit against OpenAI and Microsoft reveals internal documents suggesting that AI leadership knew their scraping practices were detrimental to publishers.

  19. Science Is Open Software jepedersen.dk

    Computational science is fundamentally dependent on software, meaning that for scientific results to be truly valid and reproducible, the underlying code must be open source.

  20. We made Playwright 2x faster and 80% more token efficient github.com

    Stagehand is a specialized SDK designed specifically for AI agents to interact with the web, moving beyond the testing-centric focus of tools like Playwright.

  21. Alibaba open-sources AI model that can detect cancer and nearly 150 conditions scmp.com

    Alibaba's Damo Academy has released an open-source vision-language model called Damo Radar that can identify nearly 150 abdominal conditions from CT scans.

  22. New evidence for hidden chambers beyond Tutankhamun's tomb nature.com

    The website cannot load its required content because JavaScript is currently disabled or blocked in the user's browser.

  23. You can run Git on object storage if you re-make packfiles tigrisdata.com

    Standard Git packfiles are optimized for local disk access and memory mapping, making them prohibitively slow when used with object storage due to excessive network round-trips.

  24. Why building a Rust LSP is hard rust-glancer.github.io

    Building a Language Server Protocol (LSP) for Rust is significantly more complex than building a compiler because it requires producing useful, real-time answers from partial, often incorrect, or incomplete code.

  25. Compiler-style optimization for drawing via Skia arxiv.org

    Modern 2D rasterization libraries like Skia are powerful but often execute inefficient instruction sequences because their internal logic is too complex and opaque for developers to optimize manually.

  26. Learning another language may be one of the best ways to keep your brain healthy theconversation.com

    Learning a new language acts as a powerful form of cognitive exercise that enhances neuroplasticity and builds cognitive reserve.

  27. A graphical desktop for the ZX Spectrum github.com

    The author creates a functional, overlapping-window GUI for the 1982 ZX Spectrum 48K, demonstrating that high-level desktop metaphors can be achieved on extremely limited hardware through precise T-state timing and rigorous measurement.

  28. Deodands put a price on objects that caused death daily.jstor.org

    The medieval law of deodands, which required the owner of an object that caused death to forfeit its value, served as an early form of accountability for nonhuman entities.

  29. ZK-JPEG: Zero-Knowledge Image Editing and Compression eprint.iacr.org

    Deep fake photographs are increasingly common, making it is essential to have tools that can verify that an image was captured by a physical camera.

  30. Black Holes or Black Hole Stars? Astronomers Spar over 'Little Red Dots' quantamagazine.org

    Astronomers are debating whether mysterious, bright red pinpricks of light discovered by the James Webb Space Telescope are standard supermassive black holes or a new class of objects called 'black hole stars.' These proposed black hole stars would be colossal hydrogen shells…

  31. Suzanne Ciani's Buchla Cookbook echo.orpheusinstituut.be

    Suzanne Ciani's 1976 report to the National Endowment for the Arts serves as a practical manual for live electronic music performance using the Buchla 200 series, specifically the Model 248 MARF.

  32. Flock Offers Employees Buyouts as Customers Flee wired.com

    Flock Safety is offering voluntary buyouts to its employees as a growing number of US cities terminate their contracts for the company's license plate readers.

  33. UFO Series Home Page: "UFO" TV Series from 1970 ufoseries.com

    This website serves as a comprehensive digital archive for the 1970 British science fiction show UFO.

  34. Show HN: CUA-S1 – A System One Model for Computer Use github.com

    Cua provides a comprehensive open-source ecosystem designed to enable AI agents to interact with computers like humans do.

  35. Ctenophores: Wonders of Biology quantamagazine.org

    Ctenophores, or comb jellies, provide critical insights into the earliest stages of animal evolution because they may be the first branching lineage of animals.

  36. Typesafe-computer-use drives a Mac toward a goal for 1/50th of a cent per step github.com

    Current frontier models are too expensive and slow for computer use because they rely on constant screenshot analysis and high-level planning for every step.

  37. AI Safety Is Mostly a Sex Cult bsky.app

    The AI safety movement is driven by more than just technical concerns about existential risk; it is structured as a a high-control group with shared ideological

  38. Goroutine Leak Profiles go.dev

    Goroutine leaks occur when goroutines block indefinitely on concurrency primitives, leading to memory and CPU degradation over time.

  39. LispBM is a concurrent Lisp for microcontrollers with message passing lispbm.com

    LispBM provides a reliable way to measure the same set of benchmarks across different Lisp same-set benchmarks across different Lisp implementations.

  40. Mayday Mysteries maydaymystery.org

    The Mystery Maze community is expanding its reach and communication channels through the growth of its Facebook group.

  41. NASA-IBM Lunar Foundation open-Source Geospatial AI Model newsroom.usra.edu

    NASA and IBM have released an open-source Lunar Foundation Model to help scientists analyze massive, diverse lunar datasets more efficiently.

  42. From Stonemasons to Carpenters thelastsoftwareengineer.substack.com

    Software development is transitioning from a 'stonemason' approach of writing code line-by-line to a 'carpenter' approach of designing structural forms.

  43. The Farnese letter simonklee.dk

    A long-unsolved 1542 letter from Cardinal Alessandro Farnese was deciphered using a combination of statistical analysis, a five-gram Italian language model, and simulated annealing.

  44. Adventures in Microcontroller Circuit Debugging bigmessowires.com

    A batch of disk emulators failed QA due to unpredictable boot behavior and system glitches, which were eventually traced to unstable clock signals.

  45. I built the fastest PHP webserver in the world qbixserver.com

    Qbix is a pure PHP web server that achieves superior throughput compared to traditional PHP-FPM or coroutine-based alternatives by utilizing copy-on-write memory management through pcntl_fork().

  46. Ray Ozzie and the Optimism of Being Early reproof.app

    Ray Ozzie's journey from early networked computing on PLATO to creating Lotus Notes illustrates that pioneering new technology is more about enabling value and exploration than achieving immediate perfection or industry standardization.

  47. Agreement between the USA and Denmark (1951,2004) [pdf] state.gov

    The 2004 US-Denmark agreement updates the 1951 Greenland defense pact, confirming Thule Air Base status, NATO SOFA integration, and joint environmental committees.

  48. People who know the most often sound the least certain vrash.substack.com

    AI researchers often struggle to communicate complex ideas to the public, leaving a vacuum that is filled by overconfident but less informed commentators.

  49. Column built an issuer processor from scratch column.com

    Column is eliminating the complexity of launching card programs by combining the functions of an issuing bank, issuer processor, and capital provider into a single integrated system.

  50. Veronese's Dogs publicdomainreview.org

    Paolo Veronese frequently incorporated dogs into his large-scale religious and allegorical paintings to humanize divine scenes and provide a grounded, emotional perspective for the viewer.

  51. Show HN: I wrote a custom assembler for CHIP-8 in C++ github.com

    The author is developing a custom assembler specifically designed for the CHIP-8 interpreted language.

  52. Stepfun Step 5 Preview (LLM): On AA Pareto frontier artificialanalysis.ai

    Step 5 Preview is a high-performance proprietary reasoning model that offers a strong balance of intelligence, speed, and cost-efficiency.

  53. Suppress vulnerabilities applying Kubernetes context to scans github.com

    Vex8s is an experimental tool designed to reduce the number of reported vulnerabilities in Kubernetes clusters by determining which CVEs are actually exploitable based on the cluster's configuration.