digest · hn front page
Wed, Sep 16, 2026
54 stories
-
Training a 4B model to produce 81% faster query plans than Postgres rohanbansal.com
A small 4B parameter language model can be trained using supervised fine-tuning and agentic reinforcement learning to produce Postgres query plans that significantly outperform default system plans.
-
EU chief opens door for Canada to become 'associate member' bbc.com
European Commission President Ursula von der Leyen proposes making Canada the first 'associate member' of the EU to strengthen ties in technology, defense, and energy.
-
Small programming tricks will-keleher.com
Engineering productivity is significantly boosted by accumulating small, high-leverage nuggets of knowledge that require little mental overhead to implement.
-
Mistral X Mozilla: Private, Multilingual AI Browsing mistral.ai
Mistral and Mozilla are partnering to integrate Mistral's open-weight AI models into the Firefox Smart Window assistant to provide a private and multilingual browsing experience.
-
Hackers Got Inside a Flock Camera wired.com
A hacker collective successfully reverse-engineered a Flock Safety camera, exposing how these devices capture vast amounts of data on vehicles and pedestrians.
-
Xiaomi Mimo 2.6 live post-training dashboard mimo.xiaomi.com
Intermittent connection drops and reconnection attempts are common technical hurdles in real-time communication systems.
-
Apple Reference Image: A New Approach for Verified Photography security.apple.com
Apple is introducing Apple Reference Image to combat AI-generated imagery by creating a cryptographically verified chain of trust from the camera sensor to the final image.
-
Show HN: I made a flight simulator, except you're just a passenger inflightsimulator.com
The provided text is a loading screen for a flight search tool.
-
Building a Linux GPU Driver for the M4 Mac Mini in One Month codyho.dev
The authors successfully developed a fully OpenGL ES 3.0 compliant GPU driver for M4 Mac Mini and MacBook Neo in one month by leveraging advanced LLMs and a custom hypervisor.
-
The Google Play app review process now regularly takes longer than a week gultsch.social
The Mastodon web application requires JavaScript to function properly.
-
Backups Aren't Simple filipovski.net
Effective data backup is not a simple act of copying files, but a complex system of managing snapshots, rotation, and redundancy.
-
Performance Improvements in .NET 11 devblogs.microsoft.com
.NET 11 introduces significant performance improvements by focusing on reducing runtime overhead through advanced JIT compiler optimizations and a redesigned async/await infrastructure.
-
PS5 Linux lead quits: "a bunch of noobs using LLMs" that "they don't understand" frvr.com
Andy 'TheFlow0' Nguyen has abandoned the PS5 Linux project because the open-source community has been flooded with inexperienced developers using LLMs to generate code they do not understand.
-
Tell the speakers that you liked their talks ohhelloana.blog
Speakers often feel vulnerable and uncertain about how their presentation landed because they lack the immediate, public feedback they once had.
-
Original Sony PlayStation 2 security chip 'broken wide open' after 26 years tomshardware.com
A retro hardware enthusiast has successfully reverse-engineered the MechaCon security chip in the original PlayStation 2 after four years of intensive work.
-
The engineering behind the US Strategic Petroleum Reserve johnjwang.com
The US Strategic Petroleum Reserve solves the complex challenge of storing hundreds of millions of barrels of oil securely and cheaply by utilizing underground salt caverns.
-
Salesforce Global Outage status.salesforce.com
Modern web applications often rely on heavily on JavaScript to render content and manage interactive elements.
-
Learning Programming in an Age of LLMs blog.ploeh.dk
While LLMs can quickly bridge the gap between an idea and a working prototype, they risk creating a dangerous competence gap where developers build systems they cannot understand or maintain.
-
Reversing Factorio's RNG gegell.github.io
The author demonstrates that Factorio's pseudo-random number generator (PRNG), based on the taus88 algorithm, is deterministic and can be reversed through linear algebra.
-
Breaking the 1.58-bit Barrier for Ternary LLMs arxiv.org
Standard ternary Large Language Models use a fixed packing format that ignores the uneven distribution of weights, wasting memory.
-
A warning about 'model welfare' mustafa-suleyman.ai
Artificial intelligence systems are sequence completion engines without feelings, consciousness, or biological substrates, and treating them as 'moral patients' is a dangerous mistake.
-
Japan's book scene is moving from bookstores to libraries untranslatedjp.substack.com
Local libraries in Japan are evolving from boring, functional spaces into vibrant community hubs and travel destinations.
-
Claude Cowork and chat are now one Claude claude.com
Anthropic is integrating Claude Cowork and chat into a unified interface to eliminate the friction of choosing where to start a task.
-
Doing Everyone Else's Job yosefk.com
Taking ownership of tasks that fall outside your official job description is a strategic advantage that allows you to learn how a system works and ensure critical goals are met when others fail to act.
-
Dream-RSI: Recursive Self-Improvement through Evolving Worlds arxiv.org
Recursive self-improvement for AI agents is hindered by the high cost and slow feedback of online exploration in complex search spaces.
-
OpenSpec – A lightweight and configurable AI spec framework openspec.dev
OpenSpec is a lightweight, configurable framework designed to align software development teams and AI coding agents through structured specifications.
-
Recreating Voodoo Graphics and a Late-1990s Gaming PC on an FPGA nand2mario.github.io
The author successfully implemented a SystemVerilog version of the 3dfx Voodoo Graphics (SST-1) chip, called zSST, and integrated it with a z486 CPU to create a functional late-90s DOS PC on a Xilinx KV260 board.
-
The DeepMind Institute institute.deepmind.com
The arrival of Artificial General Intelligence (AGI) requires a multidisciplinary approach to manage its societal, economic, and safety implications.
-
DeepSeek v4.1 Flash Is Now Our Best Hacking Model enclave.ai
DeepSeek V4.1 Flash achieved a perfect 11/11 score on a hacking benchmark by gaining code execution on all vulnerable targets while keeping fixed targets secure.
-
Anecdotally, programmers dislike "reduce" evanhahn.com
Many developers find the 'reduce' function difficult to read and less intuitive than 'map' or 'filter'.
-
Vectorized and performance-portable Quicksort (2022) opensource.googleblog.com
Google researchers have developed a vectorized Quicksort implementation that is up to ten times faster than the C++ standard library.
-
Intelligence per Watt: Measuring Intelligence Efficiency of Local AI arxiv.org
Local language models can now handle a significant portion of real-world queries, potentially reducing the reliance on centralized cloud infrastructure.
-
ER visits for gambling disorders doubled after expanded online gambling market temertymedicine.utoronto.ca
The legalization of single-event sports betting and the opening of Ontario's online gambling market to private companies led to a significant increase in emergency department visits for gambling disorders.
-
Can we stop with the uptime percentages? blog.jim-nielsen.com
Uptime percentages are an inefficient way to communicate service reliability to the general public because they fail to convey the magnitude of downtime.
-
Saving Jet Fuel tech.marksblogg.com
Optimizing flight paths based on real-time wind conditions and aircraft-specific fuel models can significantly reduce fuel consumption and operational costs.
-
Negativland, Culture Jamming, and the Art of Making Something New blog.archive.org
Negativland is a multimedia collective that uses sound collage and 'culture jamming' to critique American power structures and capitalism.
-
Anatomy of a Texture agentlien.github.io
Converting texture data between gaming platforms is surprisingly complex because modern hardware requires non-linear memory layouts to optimize rendering performance.
-
How good are frontier models at physics? arxiv.org
Current physics benchmarks for large language models are flawed due to incorrect reference solutions and ambiguous questions.
-
The Beauty of Roundabouts gruhn.me
Roundabouts are a superior alternative to traffic light intersections because they replace complex, error-prone active control systems with a simple, static structure.
-
How big are factorials? eli.thegreenplace.net
The number of digits in a large factorial can be estimated using a simplified version of Stirling's approximation.
-
Stay discoverable in search while disallowing AI training blog.cloudflare.com
Cloudflare is introducing a 'Disallow AI Training' setting to solve the conflict where website owners had to block search engine crawlers entirely just to prevent AI training.
-
Show HN: How Stale Is Your AI? Release age and training cutoff for 20 models stale.jock.pl
AI models suffer from a knowledge gap because their release dates differ from their training cutoffs.
-
Douglas Adams and the exterminated Doctor Who adventure bbc.co.uk
A 1979 Doctor Who story titled 'Shada' remained unfinished for decades because a BBC strike prevented the completion of essential studio scenes.
-
Accurate Models of AMD Matrix Cores arxiv.org
Modern GPU matrix multipliers deviate from IEEE 754 standards, leading to inconsistent results across different hardware architectures.
-
Scaling Golang CI by Replacing actions/setup-go cloudx.ai
The official GitHub actions/setup-go action slows down parallel CI workflows because its immutable cache keys are too generic, leading to stale data and race conditions between jobs.
-
This Code Is CRAP (2011) testing.googleblog.com
The CRAP (Change Risk Anti-Patterns) metric provides a way to quantitatively identify risky code by combining cyclomatic complexity and test coverage.
-
A software thing I built: GPS on a 25MHz 486-SX forum.vcfed.org
In the early 1990s, it was possible to create a real-time GPS moving map display on a low-power 486-SX CPU without a math co-processor by utilizing a specialized math library called Hipparchus.
-
WalShadow: Sub-second Postgres replication to ClickHouse from physical WAL clickhouse.com
WalShadow is a new open-source engine that replicates Postgres data to ClickHouse using physical Write-Ahead Logs (WAL) instead of traditional logical replication.
-
Datamimic – don't let your coding agent invent its own test world github.com
DATAMIMIC provides a deterministic data engine designed for regulated enterprises to generate synthetic test data and pseudonymize PII without relying on random output.
-
The Siberian Ice Maiden and the Scythian World patrickwyman.substack.com
The discovery of the Siberian Ice Maiden provides a rare, well-preserved glimpse into the Pazyryk culture and the broader Scythian world of the Iron Age.
-
Training Text-to-Image Models 3.6× Faster linum.ai
Traditional Latent Diffusion Models (LDMs) are limited by the compression ceiling of Variational Autoencoders (VAEs), which increases the computational cost of attention mechanisms.
-
Sierra digital cameras on the Apple II colino.net
The author expanded the Quicktake for Apple II project by implementing a driver for 'Sierra-class' digital cameras, enabling a wide range of vintage hardware to interface with an obsolete computer.
-
MartyPC – A Cycle-Accurate IBM PC/XT Emulator github.com
MartyPC is a cycle-accurate emulator for early IBM PCs and compatibles, developed in Rust to provide a precise environment for retro software development.
-
Better routing, probe fixes, plugin updates in Freenet/Hyphanet 0.7.5 build 1507 hyphanet.org
Hyphanet release 0.7.5 build 1507 introduces critical updates to privacy, system performance, and development infrastructure.