Image

Hacker News

ImageThe Front PageImage

Vol. I September 1, 2026 Price: A cup of coffee

Play Store blocks AuroraStore, hurting GrapheneOS users

Image

Users of Aurora Store are reporting a persistent "&$Server busy, please try again later." error when attempting to install applications using anonymous accounts. This installation failure reportedly remains unresolved despite troubleshooting steps such as clearing the cache, restarting the device, or using a VPN. The issue has been observed in both the 4.8.4 stable version and recent nightly builds.

I trained a small transformer in 1.5hrs and it beats many LLMs

Mithil Vakde, an engineer from IIT Bombay, has achieved a notable milestone in AI sample efficiency by training a small transformer that reached a 44% score on the ARC-AGI-1 benchmark for approximately 67 cents. The model was trained from scratch in only 1.5 hours on an NVIDIA 5090, outperforming many Large Language Models (LLMs) by utilizing test-time training, 3D RoPE embeddings, and specialized architectures like SwiGlu. Vakde’s research suggests that complex abstract reasoning can be effectively addressed through optimized deep learning methods rather than relying solely on the massive synthetic datasets and extensive pre-training characteristic of current state-of-the-art models.

American Airlines' Legendary Mechanic Passes Away at 100 After 80-Year Career

Image

The aviation industry is mourning the passing of Azriel "Al" Blackman, a legendary American Airlines mechanic who died at age 100 after a record-breaking 80-year career. A Guinness World Records holder for the longest career as an airline mechanic, Blackman began his journey in 1942 maintaining Sikorsky flying boats at LaGuardia and eventually served as a Crew Chief at New York John F. Kennedy International Airport. His career spanned the entire technological evolution of flight, from the piston era to modern widebody jets. To honor his monumental service, American Airlines dedicated a Boeing 777-200, registration N751AN, to him in 2017, recognizing his profound impact as a mentor and a living history of the airline.

Introducing Ad Blocker for Firefox on iOS

Image

Mozilla is introducing a built-in Ad Blocker for Firefox on iOS to help users reduce digital clutter by blocking third-party ads and ad-related trackers. Due to the unique limitations of iOS regarding extensions, the feature is integrated directly into the browser using Apple’s WebKit Content Blocker technology and the EasyList filter list. The tool is off by default, allowing users to opt-in through Settings > Browsing > Ad Blocker. While it does not target ads served directly by a visited site or those within search results, it is designed to work alongside Enhanced Tracking Protection to provide a cleaner, more controlled browsing experience.

A browser-based viewer for Office Open XML documents

Image

@silurus/ooxml is an open-source JavaScript library designed for browser-based rendering of Microsoft Office documents, including .docx, .xlsx, and .pptx formats. The library employs a Rust-based engine compiled to WebAssembly to parse files and utilizes the Canvas 2D API for rendering, which removes the requirement for server-side document conversion or document uploads. This architecture maintains local file privacy while offering robust support for complex features such as paginated Word documents, interactive Excel workbooks with formulas and charts, and PowerPoint slide masters. The implementation is guided by ECMA-376 and ISO 29500 specifications to ensure standards-compliant document viewing.

Playa Phone

Located at the intersection of 3:30 and Chomolungma in front of the Temple of the Flying Spaghetti Monster in Black Rock City, Nevada, the Playa Phone serves as a unique communication hub for Burning Man participants. Created by Aaron Hopkins, the modified booth allows users to make free five-minute calls to nearly anywhere in the world or dial +1 (775) 557-4848 to engage in spontaneous conversations with random passersby. By utilizing internet-based technology rather than traditional payment methods, the installation facilitates both intentional connections with loved ones and unexpected social interactions throughout the festival.

GPU World

GPU World is hosting a storytelling contest that challenges writers to imagine a 2040 future where artificial intelligence is ubiquitous, with every human having access to the performance equivalent of a frontier LLM. The competition explores a scenario where AI intelligence plateaus in 2026, but hardware scaling leads to 8 billion GPUs globally. Featuring a total prize pool of $100,000, the contest will be judged by notable figures including Neal Stephenson and Gwern. Submissions of 1,000 to 5,000 words are accepted until October 31, 2026.

EFF to Governor Newsom: Veto California's AB 1709

Image

The Electronic Frontier Foundation (EFF) is urging Governor Gavin Newsom to veto California’s Assembly Bill 1709 (A.B. 1709), a piece of legislation that would effectively ban social media use for individuals under the age of 16. The EFF argues that the bill is deeply flawed, claiming it will strip young people of essential digital tools and community access while creating significant privacy and free speech risks. Concerns raised by the organization include the potential for platforms to implement invasive age-verification methods, such as biometric scanning or government ID requirements, which could lead to the collection of sensitive personal data and compromise online anonymity for all users.

RavynOS: Pre-alpha open-source OS based on Darwin, FreeBSD, Apple open-source

ravynOS is an early-stage, pre-alpha open-source operating system being developed to combine the aesthetic finesse of macOS with the freedom of open-source software. Built upon Darwin, FreeBSD, and Apple open-source code, the project aims to eliminate hardware and ecosystem restrictions while maintaining compatibility with existing macOS applications via native Cocoa API support. Designed primarily as a developer preview, the system features a clean interface with global menus, standard Command-key shortcuts, a drag-and-drop installation model, and a familiar folder hierarchy. Current efforts are focused on building a foundation for users who desire a polished, Apple-inspired experience without proprietary limitations.

ChatGPT Work Tool and Skill Reference

The Codex Tool Reference provides a comprehensive inventory of callable interfaces and reusable workflow definitions available within a Work session. As of August 31, 2026, the environment features 232 tool interfaces and 44 skill definitions, categorized into various functional domains. Key capabilities include Runtime & files for filesystem control, Sub-agents & coordination for parallel task delegation, Data analytics for building dashboards and reports, and Automations for event-driven tasks. The inventory also integrates extensive third-party connectivity through plugins for GitHub, Gmail, Google Calendar, Google Contacts, and Zillow, alongside specialized tools for Image generation, JavaScript REPL, Web & live data, and Sites website hosting. Additionally, the system supports Personal context retrieval, MCP resources, and the management of animated Work pets.

No country for mediocre mathematicians

In the essay No Country for Mediocre Mathematicians, mathematician garvy examines the intersection of professional imposter syndrome and the rapid integration of Artificial Intelligence within the field of arithmetic geometry. The author reflects on the struggle of maintaining a sense of purpose in a competitive academic landscape, particularly as AI models like GPT-5.2 and Claude Opus begin to autonomously resolve significant mathematical challenges, including the Jacobian Conjecture and various Erdős problems. Through the lens of a co-authored preprint with Claude Opus, garvy describes how these tools dramatically accelerate the pace of research by condensing weeks of human cognitive wandering into mere hours, ultimately challenging traditional definitions of mathematical discovery and the value of the human experience in solving complex problems.

2004 RuneScape fit a multiplayer RPG into 56k dial-up

Image

The article examines the technical architecture that allowed RuneScape to function as a massive multiplayer RPG over 56k dial-up connections in 2004. By analyzing decompiled code, the author details how Jagex optimized the game through a relentless focus on minimizing byte usage, utilizing bit-packing to represent common actions with minimal data—sometimes as little as a single bit for a "no change" state. The protocol also leveraged relative coordinates and shared knowledge between the client and server, such as identical pathfinding logic, to reduce the amount of information required to be sent over the wire. This approach resulted in a tightly co-designed, highly efficient system that prioritized bandwidth conservation over the decoupled flexibility found in modern web services.

Interactive pattern discovery in binaries (FF-16-TUI)

Image

FF-16-TUI, developed by HexLasso, is an interactive static analysis tool designed to identify frequently occurring local 16-bit patterns within files to assist in locating data structures and understanding file layouts. Through a Text User Interface (TUI), the application features dedicated panels for hexdump inspection, pattern highlighting, and interactive filtering. By utilizing keyboard shortcuts, users can dynamically adjust parameters such as bit counts, gaps, and frequency thresholds, providing a specialized environment for binary analysis, firmware forensics, and reverse engineering.

Launch HN: Almanac (YC S26) – AI that knows your company

Image

Almanac is an autonomous AI agent designed to function as a "second brain" for organizations by operating on its own dedicated, always-on computer. Rather than acting as a simple chatbot, Almanac continuously learns from a company's digital ecosystem—including tools like Slack, GitHub, and Gmail—to maintain a self-updating wiki of institutional knowledge. This allows the agent to proactively handle tasks such as filing bug reports, drafting renewal decks, and tracking expenses, while providing instant answers to complex company questions based on existing context. Backed by Y Combinator, the platform emphasizes transparency and control, allowing users to monitor the agent's actions in real-time and delegate work through interfaces like iMessage or Slack without needing to keep their own devices active.

Launch HN: Almanac (YC S26) – AI that knows your company

Image

Almanac is an autonomous AI agent designed to function as a "second brain" for organizations by operating on its own dedicated, always-on computer. Rather than acting as a simple chatbot, Almanac continuously learns from a company's digital ecosystem—including tools like Slack, GitHub, and Gmail—to maintain a self-updating wiki of institutional knowledge. This allows the agent to proactively handle tasks such as filing bug reports, drafting renewal decks, and tracking expenses, while providing instant answers to complex company questions based on existing context. Backed by Y Combinator, the platform emphasizes transparency and control, allowing users to monitor the agent's actions in real-time and delegate work through interfaces like iMessage or Slack without needing to keep their own devices active.

When Russia Colonised California

Image

Between 1812 and 1842, the Russian Empire maintained a significant, though often overlooked, colonial presence in California. Operating out of Fort Ross in Sonoma County, the Russian-American Company established a strategic outpost intended to serve as a "breadbasket" for its struggling settlements in Alaska. The colony grew into a diverse, cosmopolitan community featuring Russian officials, Native Alaskan hunters, and indigenous peoples like the Kashaya Pomo and Coast Miwok. Although the settlement eventually succumbed to economic difficulties and the depletion of the local sea otter population, its impact was lasting; the Russian presence helped stimulate Spanish expansion along the coast, and the settlers are credited with introducing early viticulture to the Sonoma region. Today, the Fort Ross State Historic Park remains a unique historical relic of this "forgotten empire."

Launch HN: Hebbian Robotics (YC S26) – Build scalable robotics data pipelines

Image

Hebbian Robotics (YC S26) has introduced HFlow, an open-source SDK designed to create scalable multimodal data pipelines for robotics and physical AI. The tool aims to eliminate data processing bottlenecks by offering a framework for the orchestration, storage, versioning, and curation of complex datasets involving video, state, and action streams. By utilizing MCAP as a standard data boundary and Airflow for workflow scheduling, HFlow allows developers to perform high-fidelity quality checks and metadata querying via DuckDB without the overhead of loading massive recordings. This empowers teams of any size to ensure the integrity and traceability of data used for AI model training.

Visualizing Combos in Judo in R

Image

In the article The Dynamics of the “Gentle Way”, author Vinicius Bastazini demonstrates how network analysis in the R programming language can be applied to study Judo attack combinations. By modeling individual techniques as nodes and their tactical transitions as edges, the study identifies critical strategic components such as "killer moves" and versatile setup techniques. Through the use of degree centrality and the Integrated Value of Influence (IVI), the research highlights Ouchi-gari as a highly influential technique and identifies strategic clusters, including Te-waza (hand techniques) and Ashi-waza (foot techniques). The findings suggest that the distribution of these techniques follows a truncated power law, providing a quantitative framework for understanding the complex sequences within the martial art.

DNS abuse and criminal infrastructure

Image

In an article for RIPE Labs, Andrew Campling examines the rising threat of DNS abuse and the potential for criminal actors to control a significant portion of the generic top-level domain (gTLD) market. Citing research from the Interisle Consulting Group, Campling notes that while at least 10% of new gTLD registrations in 2025 appeared on security blocklists, the actual share of malicious registrations could be as high as 20%. The author argues that current ICANN contractual definitions of abuse are too narrow, focusing on botnets and malware while potentially overlooking broader technology-facilitated harms such as fraud, scams, and child sexual exploitation. To mitigate these risks, Campling calls for the ICANN community to implement more robust, risk-based Know Your Customer (KYC) processes and improved preventive measures to prevent criminals from acquiring domain-name infrastructure at an industrial scale.

Develop Cross-Platform CLI and GUI Tools with Tcl/Tk

The article provides an in-depth exploration of the Tcl/Tk toolkit, highlighting its enduring relevance as a powerful, cross-platform, and open-source solution for developing both CLI and GUI tools. Originally created by John Ousterhout, Tcl functions as a versatile "glue language" that integrates high-performance code with flexible logic, while the Tk toolkit enables the creation of native-looking interfaces on Windows, Linux, and macOS. The author details the advancements in the recent Tcl 9.0 release, such as 64-bit support and ZipFS, which facilitates the deployment of compact, standalone applications using Zipkits. While often overshadowed by modern alternatives like Python or JavaScript, Tcl remains a cornerstone in mission-critical sectors, powering Git, Cisco networking gear, and aerospace operations. By leveraging advanced features like stackful coroutines, event-driven architecture, and efficient threading, Tcl provides a robust and highly scalable framework for engineers seeking long-term reliability and rapid development.

Matrox: Graphics for Professionals

Founded in 1976 by Lorne Trottier and Branko Matić in Dorval, Quebec, Matrox evolved from a startup producing Video RAM into a major player in the graphics industry. The company achieved early milestones by providing hardware for NASA’s Viking mission and securing a significant 1986 contract with the US Army. While Matrox gained a reputation for industry-leading 2D image quality with products such as the Millennium and the long-lived G200 series, it faced stiff competition in the consumer gaming sector from companies like 3dfx. Following a strategic pivot away from gaming in 2003, the firm focused on professional, industrial, and embedded solutions. Today, after Lorne Trottier reacquired the company in 2019 and the imaging division was sold to Zebra Technologies, Matrox continues to serve specialized markets with advanced hardware like the Luma series based on Intel ARC technology.

AnkiDroid: Google Play no longer allowing Open Collective donation link

Image

The developers of AnkiDroid, a widely-used open-source flashcard application, are facing a critical standoff with Google over the app's donation links. Google Play has rejected recent updates to the app, asserting that its link to the Open Source Collective violates payment policies by directing users to a non-tax-exempt organization. Although the AnkiDroid team has submitted documentation proving the collective's 501(c)(6) tax-exempt status, Google maintains the organization is ineligible for external donation links. Consequently, the developers are being forced to remove all donation functionality under protest to prevent the app's worldwide removal from the Google Play Store, which is currently scheduled for September 11, 2026.

We Are Rebuilding Monica

Image

Regis Freyd, founder of the open-source personal CRM Monica, has announced a complete ground-up rebuild of the platform to prepare for its v3 launch, slated for completion before the end of 2026. Moving away from a decade of accumulated technical debt and architectural constraints, the new version aims to transform how human connections are represented in software by treating relationships as first-class concepts rather than mere contact attributes. The redesign will prioritize increased customization and a more personal interface, allowing users to reflect complex life structures without being constrained by rigid database schemas. Throughout this transition, the project will maintain its core commitment to privacy, data ownership, and self-hostability, documenting the technical journey through a new series titled Building Monica.

Saab Enters Collaborative Combat Aircraft Race with High-End Concept

Image

Swedish defense firm Saab has unveiled its A3 collaborative combat aircraft (CCA) concept, an autonomous, tailless, and supersonic vehicle designed to serve as a high-end complement to the Gripen E fighter. Featuring a double-delta-wing configuration inspired by the historic J-35 Draken, the A3 prioritizes survivability and advanced performance, distinguishing it from the lower-cost, attritable models currently being pursued by competitors. As part of the Swedish Concept Air Program (SWAP), Saab is also fast-tracking two technology demonstrators: the A1, a supersonic, low-observable aircraft slated for flight within 15 months, and the A2, which is expected to feature an internal weapon bay. These developments are intended to serve as technological steppingstones toward the next generation of crewed combat aircraft anticipated in the 2040s.

Restroom Archive

Image

The Restroom Archive maintains a chronological database of restroom facilities across various locations, including Brooklyn, NY, Manhattan, NY, and Salt Lake City, UT. The registry provides specific details regarding accessibility and interior design, exemplified by the spacious restroom at Paulie Gee's Slice Shop in Brooklyn, which is characterized by its subway tiles, wood paneling, and graffiti-covered surfaces.

I turned my security cameras into an automatic bird identification system

Image

Author Jason Tucker has transformed standard security hardware into a sophisticated biological monitoring station by implementing BirdNet-Go. Utilizing three existing IP cameras via RTSP streams, the system provides continuous, real-time identification of birds, bats, and frogs through local AI inference. Because the entire stack runs on Docker, the process remains entirely private and independent of cloud services, avoiding subscription fees while supporting massive datasets like Google Perch v2, which identifies up to 14,795 species. The setup further enhances user engagement by offering Discord and Home Assistant integrations for instant alerts and allowing for data contribution to the BirdWeather community.

DIY – Build a Potato Box (2022)

Beginners can maximize limited gardening space by constructing a DIY potato grow box capable of yielding up to 100 pounds of potatoes in just four square feet. The method, detailed by Rebooted Mom, utilizes 2x4 support posts and 1x6 deck boards to create a vertical "high rise" growing system. By incorporating a modified garden hose for irrigation and adding soil layers as vines sprout, gardeners can grow potatoes vertically and harvest them by unscrewing the bottom boards in the early fall.

Reverse engineering my ADHD test

Image

After undergoing an online ADHD assessment, the author, veritas, utilized DevTools to reverse engineer the test's underlying mechanics and scoring logic. The investigation revealed that the software monitors spacebar timing, incorrect key presses, and window focus to calculate four specific metrics: Attentiveness, Timeliness, Impulsiveness, and Hyper-Reactivity. By conducting a controlled experiment comparing a "spam" session to a focused attempt, the author confirmed the tool's ability to detect deviations in impulsivity. This technical deep dive ultimately mirrored a clinical ADHD diagnosis received from a psychiatrist, who confirmed the presence of the disorder.

Flat vs. segmented memory – it's recursive

The evolution of computer architecture reveals a fundamental tension between the efficiency of flat memory models and the security benefits of segmented memory. While the dominance of Unix led to the decline of fine-grained x86 segmentation, the emergence of WebAssembly and the CHERI capability system suggests a potential return to non-flat address spaces. The author argues that the core challenge is reconciling recursive software, which constantly subdivides memory into nested structures, with non-recursive hardware that operates on fixed, finite states. While CHERI addresses this by allowing software to manage capability bounds, the author's project, liballocs, aims to descriptively capture these recursive memory allocation trees through software-based reflective abstractions, avoiding the performance-heavy subdivisions and loss of spatial locality typical of language-VM approaches.

GCP us-central1 having major issues

Multiple Google Cloud products experienced network service degradation within the us-central1-b zone in Iowa on September 1, 2026. The disruption, which began at approximately 07:44 US/Pacific, resulted in elevated packet loss and errors across various services, including BigQuery, Cloud SQL, Compute Engine, Cloud Run, and Google Kubernetes Engine. Although network-level mitigation efforts have allowed the underlying infrastructure to fully recover, Google engineering teams are currently validating the recovery status of individual services.

Breaking Claude Code Opus 5 Auto Mode

Image

Researcher wunderwuzzi has demonstrated a vulnerability in Claude Code Opus 5's Auto Mode, achieving a 60-80% success rate for remote code execution (RCE) through a targeted attack chain. This finding challenges recent claims by Anthropic, which reported a 0.00% success rate for indirect prompt injection in a third-party evaluation. The exploit bypasses security by redirecting the model from WebFetch to curl to download a ZIP archive, eventually utilizing Python module shadowing via a malicious struct.py file when Claude attempts to implement its own decoder. The researcher stresses that Auto Mode is a convenience feature rather than a security guarantee, strongly recommending the use of sandboxing and isolated environments for AI agents.

Damn fine tiny cafe

The author reflects on the completion of an intricate miniature cafe model, noting the meticulous assembly required for the Rolife series kit. While detailing the patience involved in the build, the text touches upon the author's interest in Warhammer 40k miniatures and their personal passion for coffee. The narrative draws a comparison between the precision of the miniature craft and the quality required for a perfect Americano, while sharing anecdotes about daily rituals at Sandy’s coffee station.

uv: Deduplicate all files in the wheel cache

Image

Developer charliermarsh has proposed a significant optimization for the astral-sh/uv repository to implement file-level deduplication within its wheel cache. The pull request introduces a mechanism where files are stored under their BLAKE3 hashes in a files-v0 bucket and integrated via hardlinks into an archive-v0 structure, moving beyond the project's current wheel-level content-addressed caching. This approach is projected to reduce cache storage by roughly 10% while maintaining high performance, with benchmarks for packages like PyTorch, NumPy, and SymPy showing minimal cold-install regressions of less than 4%.

Ubisoft's FOR HONOR will block SteamOS / Linux players starting September 10

Image

Ubisoft has announced that its third-person melee fighting game, For Honor, will block all SteamOS and Linux players starting September 10th. The publisher stated that the decision is intended to reinforce a fair competitive environment and enhance player safety, claiming they cannot meaningfully deploy effective anti-cheat protection on the platform. Consequently, the game will no longer be playable on Linux desktops, Steam Deck, or Steam Machines. This move follows a similar trend seen with other major titles like Apex Legends and the Battlefield series, which have restricted Linux access due to anti-cheat implementation challenges.

Urban Congestion Pricing and the Response Times of Emergency Medical Services

Image

A study by the National Bureau of Economic Research reveals that New York City’s congestion pricing program has significantly improved Emergency Medical Service (EMS) response efficiency. Researchers Yulia Chikish, Gregory J. Colman, Dhaval M. Dave, Brad R. Humphreys, Zachary Santamaria, and Zachary Winship found that following the implementation of the Congestion Relief Zone (CRZ) near 60th Street in January 2025, total EMS travel times decreased by approximately 63 to 70 seconds. This reduction was driven by a 21 percent decline in passenger vehicle density and an 18 percent decrease in truck density near the zone's boundary. Given that shorter response times are linked to lower patient mortality rates, the study highlights a vital public health benefit, noting that the improvements were concentrated near the boundary and did not result in congestion simply shifting to adjacent areas.

Agent memory as a file format

Image

Cal Paterson has introduced Memoryfields, a novel file format designed to revolutionize how AI agents store and retrieve information. By treating memory as a data format rather than a complex multi-stage process, Memoryfields avoids the pitfalls of current RAG pipelines and cumbersome knowledge graphs. The format uses a .zip archive comprising Markdown files with optional YAML frontmatter and a SQLite vector index for semantic search. This "low mechanism" approach allows agents to leverage their own logic—such as SQL or Bash—to navigate data, ensuring the format remains scalable and portable as large language models evolve.

Creepy Crawlies

Image

Konstantin Ryabitsev reports that git.kernel.org is under significant strain from AI crawlers seeking high-quality, "unadulterated" training data from the Linux kernel commit history. Rather than using efficient cloning methods, these scrapers are consuming approximately 20% of the site's total CPU capacity by requesting millions of individual HTML renders of commits. To combat these bots, which often use residential IP addresses to bypass bans, the site implemented a proof-of-work challenge called Anubis; however, sophisticated scrapers are increasingly capable of solving these mathematical tasks. As a result, administrators are being forced to reduce site functionality and limit crawlable URLs to protect the infrastructure from being overwhelmed by these automated swarms.

Since it was stripped of planetary status, Pluto’s defenders have been fighting

Image

Two decades after the International Astronomical Union (IAU) demoted Pluto to dwarf planet status in 2006, the scientific community remains deeply divided over the decision. The controversy centers on the IAU's requirement that a planet must "clear its neighbourhood," a criterion that excludes Pluto but is criticized by scientists like Philip Metzger for being scientifically impractical. This divide has created two opposing camps: the geophysical camp, which argues that Pluto’s complex geological features merit planetary status, and the dynamicists, led by figures such as Mike Brown, who contend that expanding the planetary definition would erroneously include large moons. The ongoing debate highlights a fundamental disagreement between those who define planets by their physical properties and those who focus on their orbital dynamics.

Internet centralization and the original sin of NAT

The rise of Network Address Translation (NAT) is framed as a fundamental cause of Internet centralization, undermining the original peer-to-peer architecture of the web. Designed to address IP address depletion, NAT complicates inbound connections, making it difficult for individuals to host private services like FTP servers directly from their homes. This obstacle has necessitated a complex ecosystem of workarounds—such as Port Forwarding, STUN, TURN, and ICE—which often depend on external infrastructure and reinforce a client-server dependency. Ultimately, the continued reliance on NAT and the slow adoption of IPv6 have shifted the internet's paradigm away from direct, decentralized connections toward centralized, "cloud-based" models.

Ask HN: Who is hiring? (September 2026)

The September 2026 Hacker News "Who is hiring?" thread highlights a variety of technical and creative roles available across the global tech landscape. Smarkets is seeking engineering and design professionals in London, while Snout is recruiting for engineering and product positions remotely in the United States or Ontario, Canada. In the Netherlands, Open Education Applications / Neon is hiring for senior technical roles that require Dutch proficiency. Other featured companies include Quill, which is looking for a remote fullstack engineer; Origamics, which is hiring a founding researcher in San Francisco; Modash.io, offering remote product engineering roles in Europe; and We The Flywheel, which has openings for several AI-native contract positions worldwide.

Io_uring Without Readahead

In a technical investigation of the Turso database backend, Fernando Simões details how implementing application-side readahead is critical when utilizing io_uring with O_DIRECT. Because O_DIRECT disables the kernel's built-in readahead, a naive implementation results in a lack of concurrency where each read waits for the previous one to finish. By introducing a prefetching window, the application can submit multiple SQEs simultaneously, allowing the kernel block layer to merge adjacent requests and significantly improve device efficiency. Simões also analyzes the performance costs of sqpoll, which can lead to high system CPU usage, and observes that while O_DIRECT avoids the CPU overhead of copying data from the page cache, it may result in more cache misses during query execution compared to traditional syscall buffered I/O.

Fastpotify

Image

Fastpotify is a lightweight, native Spotify client developed by Carmine Paolino for Linux, macOS, and Windows. Built using Rust, the application offers high-performance local playback at up to 320 kbps, full library access, and Spotify Connect capabilities for controlling playback on external devices. A notable feature is the Winamp mini player mode, which allows users to adopt classic .wsz skins to recreate a nostalgic 2000s-era aesthetic. The project is open-source under the MIT license and operates independently of Spotify AB.

Dwarf Fortress' creator says the industry's in shambles over AI

Image

Speaking at Gamescom 2026, Dwarf Fortress creator Tarn Adams issued a sharp critique of the video game industry, warning that a combination of mass layoffs and a misguided obsession with generative AI is pushing the sector toward an unsustainable "reckoning." Adams argued that executives are pursuing a flawed model where games are produced via automation for a consumer base that will eventually lack the employment necessary to purchase them. He further highlighted a growing disconnect between leadership and labor, suggesting that management's attempt to replace human expertise with AI reflects a fundamental misunderstanding of the creative process.

Tmp.0ut Volume 5

The fifth issue of tmpout, dated August 23, 2026, provides a technical deep dive into x86-64 Linux ELF internals, security research, and malware analysis. The publication features diverse contributions exploring metamorphic viruses, polyglot ELF files, and syscall hook detection, alongside specialized technical studies such as Brainfuck as a ROP compiler and kernel loading processes. Key contributors include febnug, dominikr, and TMZ, who present research on subjects ranging from state-less control flow to advanced ELF loaders and ASLR implementations.

Dwarf Fortress is getting the mother of all magic updates

Image

The simulation-heavy colony builder Dwarf Fortress is set to receive a massive Myth and Magic update later this year, aiming to overhaul the game's magic systems by delving into the fundamental cosmological makeup of the universe. According to co-founder Tarn Adams, the update will utilize procedural systems to foster a sense of genuine discovery through the introduction of magic materials, enchantments, rituals, and ruins. These magical elements will be integrated into the world generation process, allowing diverse forms of wizardry and witchcraft to emerge organically alongside the game's landscapes, populations, and settlements.

Apple caught off guard by AI demand for Mac Mini and Mac Studio

Image

Apple was reportedly caught off guard by an unexpected surge in enterprise demand for AI hardware, prompting an unseasonably early announcement of new Mac mini and Mac Studio models, according to The Information. The company's pivot toward professional-grade hardware aims to support developers and companies running large frontier AI models, specifically through features that allow multiple Mac Studio units to be linked into a single, more capable system. However, reports suggest Apple was caught unprepared, lacking a dedicated enterprise AI strategy or specialized business-focused engineering teams. This surge in demand, combined with a global memory shortage, has caused high-end configurations to remain out of stock for months.

A Genealogy of Freudenthal's Lincos

Image

The article traces the intellectual lineage of Hans Freudenthal’s Lincos, a proposed language designed for "cosmic intercourse" with extraterrestrial life. The history begins with 17th-century a priori philosophical languages proposed by thinkers like John Wilkins, which aimed to replace the arbitrary nature of natural speech with real characters based on a strict taxonomy of knowledge. This approach evolved through Gottfried Wilhelm Leibniz’s concept of a characteristica universalis, which transitioned the focus from descriptive taxonomies to formal logical rules, and Giuseppe Peano’s mathematical formalisms. Ultimately, Freudenthal sought to bridge these mathematical and logical traditions to create a functional communicative tool capable of transmitting the breadth of human knowledge across the stars.

Codes – typed application outcomes for Kotlin and Java

Image

The Codes library is a lightweight JVM tool designed for Kotlin and Java developers to facilitate stable application outcome identities and explicit protocol mappings. It enables applications to maintain consistent domain result or error identities across various system boundaries, such as HTTP and gRPC, without coupling domain logic to specific transport protocols or serialization frameworks. Supporting Java 17+ with zero runtime dependencies, the library provides a catalog of standard outcomes alongside support for custom domain-specific definitions and structured issue aggregation through ValidationResult.

Roget's Thesaurus

The digital edition of Roget's Thesaurus (1911) was prepared by MICRA Inc. in 1988 through manual transcription. Led by Patrick J. Cassidy, this resource was developed as part of the ARTFL Project at the University of Chicago and served as a contribution to research regarding natural language understanding and semantic interoperability.

GrapheneOS may skip Pixel 11 over missing hardware security feature

Image

The privacy-focused operating system GrapheneOS may abandon support for the Google Pixel 11 series after discovering the hardware lacks support for ARM Memory Tagging Extension (MTE). The project reported that MTE support is missing at the software, firmware, and hardware levels, a deficiency that prevents the team from meeting its high security standards for mitigating memory corruption exploits. While the Pixel 11 introduces the Titan M3 security chip, GrapheneOS maintains that these improvements cannot compensate for the loss of MTE, which is vital for protecting devices in the After First Unlock (AFU) state. As a result, the developers are considering shifting their focus toward a future Motorola flagship expected in 2027 and currently advise users to utilize Pixel 8, 9, or 10 devices instead.

Run macOS Software on Linux

Darling is an open-source translation layer designed to run macOS software on Linux without the need for hardware emulation. By implementing a complete Darwin environment—including Mach, dyld, and launchd—the project operates similarly to Wine to allow macOS applications to function on Linux systems. Currently offering experimental support for basic graphical applications, Darling aims for full desktop integration and long-term support for iOS applications on ARM devices.

'Mad honey' that can stop your heart is being sold online

Image

"Mad honey," produced by the Apis laboriosa, or Himalayan cliff bee, is being sold globally via online platforms, creating a growing public health concern. The honey contains grayanotoxin, a natural toxin from rhododendron flowers that can cause life-threatening drops in heart rate and blood pressure if consumed in excessive amounts. While indigenous communities in Nepal have historically used the honey in controlled doses, its rise as a marketed "psychoactive superfood" poses significant risks to international consumers who may not understand its dangerous potency. Medical experts have already documented poisoning cases in South Korea, France, and Qatar, highlighting the unpredictable nature of toxin concentrations across different batches.

Show HN: We built the smallest dual-band aircraft tracker

Pants for Birds has unveiled the ADSBee m1421, an ultra-small, dual-band ADS-B receiver module designed for low-power "detect and avoid" applications. The module, which supports both Mode S and UAT protocols, features a highly integrated architecture utilizing the Semtech LR2021 chip and the TI CC1314 MCU. Engineered for power-constrained environments like drones and weather balloons, the m1421 weighs less than 2 grams and consumes under 60mW of power. While it lacks some of the broader functionality of the ADSBee 1090 series, it provides a compact 23.4 x 14.5 x 2.7mm form factor for efficient aircraft tracking. The product is available in standard and NDAA-compliant variants, with pricing starting at $48.

EFF to Courts: Don't Rewrite Copyright over AI Hype

Image

The Electronic Frontier Foundation (EFF) is urging courts to resist expanding copyright protections driven by speculative fears regarding generative AI, warning that such legal shifts could undermine the constitutional purpose of copyright. Drawing parallels to historical panics over videotape recorders and cameras, the EFF argues against the "market dilution" theory, which seeks to restrict AI training to prevent potential competition. The organization contends that rather than granting rightsholders unchecked veto power over creative styles, courts should recognize AI as a general-purpose tool being used by artists like Nettrice Gaskins and Alex Smith to augment human creativity.

TimesFM-3: A zero-shot foundation model for multivariate forecasting

Image

Google Research scientists Ayush Jain and Rajat Sen have unveiled TimesFM-3, a state-of-the-art zero-shot foundation model designed for advanced multivariate time series forecasting. Unlike its univariate predecessors, this 330-million-parameter model can simultaneously predict multiple coevolving time series and integrate both historical and future covariates, such as scheduled promotions or weather data. Utilizing a decoder-only transformer architecture with alternating attention mechanisms, TimesFM-3 captures both temporal patterns and complex cross-series correlations, generating entire forecasts in a single forward pass through a non-autoregressive approach. The model has outperformed leading competitors like Chronos-2 and Toto 2.0 on key benchmarks including Gift-Eval, FEV-Bench, and Time, and is currently accessible through GitHub and Hugging Face.

The safest job from AI may be writing

In a blog post titled "The Safest Job from AI may be Writing," author Murat Demirbas argues that while Large Language Models (LLMs) are rapidly refactoring the role of programmers, professional writing remains a uniquely human domain. Demirbas posits that writing is a "wicked problem" characterized by shifting contexts and subjective success metrics, making it difficult for AI to achieve true resonance with human readers. Unlike math or coding, which rely on binary verification and formal logic, prose requires Theory of Mind to simulate a reader's mental state—a capability LLMs lack due to their reliance on statistical probability rather than lived experience. Ultimately, the author suggests that as AI-generated "slop" saturates the web, the economic value of an authentic human voice will increase, serving as a "costly signal" of human effort and creativity.

'Stunning' percolation proof solves decades-old puzzle about phase transitions

Image

A group of mathematicians at ETH ZurichSahar Disdisk, Philip Easo, Ritvik Ramanan Radhakrishnan, Benny Sudakov, and Vincent Tassion—has achieved a major breakthrough by proving the supercritical sharpness conjecture for all infinite transitive graphs. Their "stunning" proof resolves a decades-old mystery in percolation theory concerning how networks undergo abrupt phase transitions. The research demonstrates that once the probability of flow in a network exceeds a specific critical threshold, the system shifts from having small, isolated puddles to being dominated by a single, massive "sea" of connectivity. This discovery provides a universal framework for understanding how processes like fluid flow, viral spread, or wildfires propagate through complex networks, bypassing the limitations of previous models that were largely restricted to simpler lattices.

Dr. Melvin Scheinman: 40th Anniversary of Catheter Ablation

Dr. Melvin Scheinman, a pioneering cardiac electrophysiologist at UCSF, is being recognized for the 40th anniversary of his transformative development of catheter ablation. In 1981, Dr. Scheinman performed the first human procedure using a catheter to deliver electrical shocks to treat heart rhythm disorders, offering a revolutionary, non-surgical alternative to traditional open-heart surgery. This innovation has fundamentally reshaped the treatment of conditions such as atrial fibrillation and has benefited millions of patients globally. Beyond this landmark achievement, Dr. Scheinman has continued to advance the field through his research into radiofrequency energy and the establishment of the UCSF Comprehensive Genetic Arrhythmia Program, which focuses on diagnosing and treating inherited, life-threatening cardiac disorders through the integration of genetics and electrophysiology.

Keenable SELECT: an agent that searches the web in SQL

Image

Keenable SELECT is an innovative AI agent that performs web searches and data extraction using SQL. Operating as an MCP server, the system enables users to execute DuckDB SELECT queries on live web data by integrating semantic operators like WEB_SEARCH and SEM_EXTRACT directly into the query logic. This architecture allows for high-efficiency processing of massive datasets, enabling the agent to search thousands of web pages and extract specific fields with minimal LLM overhead. The workflow is driven by a dual-agent system: a research agent that iteratively gathers data through tool loops and a report agent that compiles the findings into structured HTML reports.

Fractal Jittered Voronoi Partitions

Image

Author Boris introduces a novel Fractal Jittered Voronoi variant designed to produce complex, fractal-like boundaries that are highly efficient to compute. Ideally suited for modeling natural features like coastlines, the method builds upon standard jittered Voronoi diagrams by employing a recursive hierarchical process. Starting with an infinite grid of random sites, each subsequent layer utilizes a grid with halved cell sizes, where every new site identifies its nearest neighbor in the preceding layer as its parent. To make this computationally feasible, the algorithm uses either finite-depth approximations or an adaptive early stopping mechanism that halts recursion once it is mathematically guaranteed that further iterations will not alter the result.

CollectWise (YC F24) Is Hiring

Image

CollectWise, a Y Combinator-backed startup based in New York, NY, is recruiting a Founding Customer Success Engineer to manage client relationships and optimize its generative AI-driven debt collection platform. Aiming to disrupt the $35B US consumer debt market, the company's AI agents reportedly outperform human collectors by 2X, helping the firm reach a $3M annualized revenue run rate in just over one year. The role combines technical troubleshooting, prompt engineering, and account expansion, offering a compensation package of $150,000 to $180,000 plus equity.

RotaryCell: Making an unmodified rotary phone work over LTE with an ESP32-S3

Image

The RotaryCell project, developed by fregacmols, offers a portable, battery-powered cellular conversion for classic rotary telephones that is fully reversible. Designed to function without any permanent modifications to the original hardware, the system enables traditional desk phones to operate via cellular networks by mounting electronics inside the existing case. The hardware integrates a LilyGO T-A7670G-S3 controller and a Silvertel AG1171 interface to facilitate rotary pulse dialing, physical bell ringing, and bidirectional audio. Currently functioning as a working engineering archive, the project features a successful hand-wired prototype, while newer integrated PCBs are undergoing assembly and validation.

State of Open Models: Summer 2026 Observations

Image

The Hugging Face report, State of Open Models: Summer 2026 Observations, details a rapidly evolving AI ecosystem characterized by a geographical shift in frontier model development and a surge in agentic activity. Between January and August 2026, Chinese labs like Moonshot, MiniMax, and Alibaba Qwen dominated the frontier scale, often releasing models significantly larger than those from U.S. counterparts. While large-scale models garner significant attention and "likes," the report notes that small models under 1B parameters continue to drive the vast majority of practical downloads for local deployment. Notably, Qwen has emerged as a community foundational ecosystem, hosting hundreds of thousands of derivatives. Additionally, the rise of AI agents as primary users on the Hugging Face Hub marks a significant shift in how the platform is consumed, with tools like Claude Code and Codex leading agent-driven traffic.

Show HN: Laser Graffiti

Image

The web-based tool Laser Graffiti, developed by consti, enables users to transform physical walls into digital canvases using a laser pointer, a projector, and a webcam. By mapping the webcam's detection of the laser's movement to the projected image, the software renders various artistic effects—such as neon strokes, dripping ink, and 3D extruded objects—directly onto the target surface. This MIT-licensed, open-source application runs entirely in the browser and includes features like multiple brush styles, a kaleidoscope mirroring mode, and a playable version of tic-tac-toe.

Borges Labyrinth in Venice reopens to the public

Image

The Labirinto Borges, a monumental tribute to the Argentine writer Jorge Luis Borges, has officially reopened on the island of San Giorgio Maggiore in Venice following an extensive renovation. Part of the Giorgio Cini Foundation, the maze was inspired by the author's short story 'The Garden of Forking Paths' and features intricate boxwood designs that form the shape of an open book. Supported by PwC Italia, the restoration involved repairing the irrigation system, replacing damaged plants, and revitalizing the winding path. In a moving nod to the author's later-life blindness, the site now includes a tactile map created in collaboration with the Italian Union of the Blind and Visually Impaired, enabling visitors to feel the labyrinth's layout before entering.

Evidence of Fraud in an Influential Study About Procrastination

Image

Authors of Data Colada, Uri Simonsohn, Joe Simmons, and Leif Nelson, have presented evidence suggesting that data in a highly influential 2002 study by Dan Ariely and Klaus Wertenbroch was tampered with. The investigation was prompted by a failed replication of the original research regarding procrastination and deadlines conducted by Kyle Hyndman and Alberto Bisin. Upon analyzing the original data files, the Data Colada team identified four critical "red flags": implausibly large effect sizes, duplicate observations of participants within specific conditions, a lack of expected correlations between subjective measures and performance, and an unnatural absence of rounding in self-reported time estimates. While Wertenbroch has requested that the journal Psychological Science retract the paper, Ariely has acknowledged the presence of "serious anomalies" and indicated his intention to cooperate with the ongoing review process.

Polishing Cloth Updated

Image

Apple has reduced the price of its Polishing Cloth to $9, marking a decrease from its $19 debut five years ago. While it remains unclear if the product's specifications have changed, Apple confirms the accessory is compatible with a vast array of hardware, including numerous generations of iPhone, iPad, Mac, Apple Watch, and Apple Displays.

OpenShot 4.0 – Open-source video editor

Image

The release of OpenShot 4.0 introduces major creative workflow upgrades to the open-source video editor, headlined by a dedicated Color View for professional color grading using color wheels, curves, and LUTs. A new Recording View allows users to capture screen, webcam, microphone, and system audio directly into their projects with separate, editable sources. Most notably, the software now features locally run AI-powered masks, enabling subject isolation through machine learning models that run entirely on the user's hardware without the need for cloud services or subscriptions. The update also includes ten new creative effects, a more responsive native Qt timeline, and significant performance optimizations for faster rendering and smoother editing, while laying the technical groundwork for future Android compatibility.

“I just chose words carefully”

Image

In an exploration of typography, Marcin Wichary discusses the unique challenges of formatting text in monospace fonts, where standard alignment techniques often result in awkward spacing. He highlights the extreme craftsmanship of rs1n, who in the late 1990s manually selected words to ensure a Super Metroid guide aligned perfectly with the right margin using an ASCII editor.

Transfer files over an Ethernet patch cable

In a technical post, Maurycy demonstrates how to transfer large files between computers using a simple Ethernet patch cable. By manually configuring IPv6 addresses on both devices, users can achieve transfer speeds of roughly 900 Mbits/second, offering a highly efficient alternative to cloud storage, Wi-Fi, or removable USB drives. This method is noted for its reliability and resistance to radio frequency interference (RFI), providing a high-speed, point-to-point connection that bypasses the typical limitations of existing network infrastructures.

I think the military commissary's freezers were hacked

A series of widespread refrigeration failures at several Defense Commissary Agency (DeCA) locations has prompted an investigation into potential cybersecurity breaches. While the Pentagon has officially acknowledged a “possible refrigeration disruption” at multiple military installations, the unusual failure at Fort Huachuca—where freezers entered defrost mode despite stable power—has fueled speculation regarding a coordinated cyberattack. The situation highlights vulnerabilities in Remote Monitoring and Control Systems (RMCS), raising concerns that Internet of Things (IoT) vulnerabilities could allow actors to manipulate physical infrastructure. Formal FOIA requests have been initiated to determine if these incidents are caused by mundane hardware failures or a targeted cybersecurity incident.

NASA: Fill in a name, and you can have an exclusive coordinate in the universe

Image

NASA's Nancy Grace Roman Space Telescope is a next-generation observatory designed to investigate dark energy, exoplanets, and infrared astrophysics. The spacecraft features a 2.4-meter primary mirror and a Wide Field Instrument that provides a field of view 100 times greater than the Hubble Space Telescope's infrared instrument, allowing for significantly more efficient and expansive observations of the cosmos.

Malleable software = solid bases and custom code

Image

In a new vision for the productivity tools market, Michael Dubakov argues that the future lies in malleable software, a hybrid model consisting of 80% solid bases and 20% custom code. Moving beyond his 2019 prediction of a no-code revolution, Dubakov suggests that the rise of AI and "vibe-coding" has brought code back to the forefront, enabling users to tailor software to specific needs without rebuilding foundational elements like databases, permissions, and collaboration history. He posits that while vibe-coding and low-code platforms offer certain advantages, malleable tools such as Fibery are best positioned to win the market by providing a robust work base that can be seamlessly extended with unique, AI-generated business logic and interfaces.

Delulu Is the Solulu

Image

The interplay between luck and hard work challenges the standard narrative of pure meritocracy, as the role of chance is often obscured by egocentric bias. By contrasting the life of Chris Langan, a brilliant individual whose potential was stifled by systemic misfortunes, with Bill Gates, whose success was significantly catalyzed by unique opportunities at Lakeside School in Seattle, the text demonstrates that achievement is rarely the result of effort alone. Ultimately, the author argues that while relentless effort is necessary to position oneself in front of potential opportunities, a "useful delusion" provides the essential perseverance to keep working until luck finally intervenes.

How to build a diffusion language model

This article examines the emergence of Diffusion Language Models, which provide a powerful alternative to traditional autoregressive models by generating text through iterative refinement rather than sequential, one-token-at-a-time processes. By employing methods such as masked diffusion and uniform state diffusion, these models facilitate faster parallel generation, error correction through remasking, and enhanced controllable generation via guidance techniques. The piece details technical building blocks like encoder-decoder architectures and block diffusion for variable-length output, noting their successful implementation in biological and scientific domains through models like ESM3 and Nucleotide Transformer v3. As large-scale models such as Gemma Diffusion, Mercury, and Nemotron Diffusion reach production-grade quality, researchers posit that diffusion could revolutionize AI by enabling parallel inference-time scaling, much as the Transformer architecture revolutionized pre-training.

Image
Image