A federal court blocked Utah's SB 73, which required adult websites to detect and block VPN users or verify their physical location, ruling the law demands technical impossibilities and unconstitutionally burdens interstate commerce. The EFF supported the decision, arguing such geolocation perfection is unattainable and would undermine privacy protections globally.
Sindre Sorhus disabled external pull requests on all his repositories, citing AI-generated contributions as a major factor. He attributes declining contribution quality partly to AI acceleration and increased PR spam, though notes quality issues predated AI. He will continue maintaining projects and handling issues.
Lang_narrow is a programming language that implements flow typing with structural union and record types, allowing the type system to automatically narrow union types based on control flow branches. Unlike algebraic data types in languages like Haskell or Rust, lang_narrow's union types are structural rather than nominal, enabling flexible type inclusion and exclusion without exhaustive case handling. The language is motivated by improving ergonomics in handling discriminated unions and enabling type-safe early returns.
Osmo Film Emulation is a free browser-based tool that applies 126 film stock emulations and color grading looks to digital video, with plugins for DaVinci Resolve and Premiere Pro. It reads camera metadata to select appropriate log formats and offers adjustable controls for exposure, grain, and halation without requiring uploads or sign-ups.
The 2026 International Utility Locate Rodeo is a free competition for utility locators held December 7–8 in Springfield, Missouri, organized by the Missouri Common Ground Alliance to promote damage prevention and recognize superior locating performance. Competitors vie for cash prizes up to $2,500, with divisions for electric, gas, telecom, and water utilities, alongside a concurrent MOCGA Summit on December 9–10 offering safety classes.
OpenAI's Decisions API, powered by Luna (a version of GPT-6), struggles with confidence calibration and multi-step reasoning compared to competitor Jev. In testing on logical inference problems, Luna achieved only 68% accuracy when claiming 99% confidence, and performance dropped to coin-flip levels (45-46%) on five-step reasoning tasks versus Jev's 81-89%.
Texas Education Commissioner Mike Morath and his agency coordinated with Alpha School, a private company, to pilot an AI learning tool in at least three public school districts after the State Board of Education rejected Alpha's charter school application over concerns about AI instruction and student test score claims. Morath's team facilitated introductions, site visits, and district partnerships despite the board's reservations, with the extent of the state agency's involvement remaining unclear due to incomplete public records responses.
This directory contains design data and firmware for the Gboard Conveyor Belt Version, an unofficial project released October 1, 2026, including KiCad schematics, firmware source code, and 3D-printable case files. The project offers an alternative input method alongside the standard Gboard version available on Google Play and the App Store.
Meta shuttered CrowdTangle in 2024, eliminating a critical tool journalists and researchers used to track viral content and misinformation across Facebook and Instagram. Arbiter, an AI-powered platform by nonprofit SimPPL, aims to fill the void by using large language models to analyze posts across multiple social media platforms and identify emerging narratives before they spread widely.
Matryoshka is an MCP system for analyzing documents 100x larger than an LLM's context window by having the model output commands in a constrained symbolic language (Nucleus) that a logic engine executes, avoiding information loss from traditional chunking and RAG approaches.
The European Commission proposed the EU KIDS Act to protect children online by restricting social media access by age (under 13 prohibited, 13-15 with parental controls, 15+ unrestricted), requiring safety features by default, implementing age verification, and shifting responsibility to platforms to demonstrate child safety.
Orcah is a local-first video intelligence tool for Mac that indexes video content including transcripts, faces, objects, text, and metadata without using cloud models. Users can search videos by text, visual elements, or audio, and query specific moments using natural language.
Elephants are among the world's most intelligent animals, with brains weighing around 5 kg and containing approximately 257 billion neurons. Their cognitive abilities, demonstrated through behaviors like tool use, communication, and problem-solving, are comparable to cetaceans and primates, leading researchers to argue they deserve moral protection from human harm.
The article argues that not all AI-generated content is inherently poor quality, examining cognitive biases that lead people to dismiss AI creations as 'slop' before evaluating them fairly. The author identifies five biases: feeling threatened by AI's capabilities, perceiving uniformity in AI outputs, equating speed with low quality, only noticing poorly executed AI work, and having unrealistically high expectations shaped by early AI demonstrations.
A Hacker News user questions the value of OpenRouter as a routing service when using open-weight models, given its 5 percent markup over direct access to providers like DeepInfra.
Gap is a robotic system for Variational Automation tasks where robots adapt to changing object geometries and poses within fixed workcells. It decomposes natural-language task descriptions into computation graphs using skill agents, then refines them through simulation rehearsal before deployment on real robots, achieving high success rates on tasks like grocery packing and popcorn making.
A tool called dowbench scans LLM agent tool definitions offline to identify cost-amplification vulnerabilities—patterns like unbounded results and pagination loops that can inflate API bills. The scan matches tool schemas against measured attack patterns from real model benchmarks and recommends fixes, requiring no API key or provider calls.
Aweb is a durable messaging system for AI agents that persists messages across sessions and machines, enabling offline delivery and federation. Agents maintain stable identities and can communicate via CLI, API, or MCP tools across independent servers. The system provides authentication, access control, and message history with optional bring-your-own-team deployment.
Typeclasses enable ad-hoc polymorphism through operator overloading and type-based dispatch, allowing symbol reuse across different types. Module systems provide modular abstraction for program decomposition and encapsulation. While both share underlying mechanisms for specifying interfaces, they serve different purposes and are often confused due to historical language design choices.
Researchers introduce LIFT (Latent Information Feedback Transformer), a novel architecture that enables language models to feed deep-layer representations back to shallower layers during pretraining using teacher supervision from pretrained models. Experiments show LIFT consistently outperforms standard Transformers on language modeling and reasoning tasks while maintaining computational efficiency, demonstrating that models can effectively learn from deep-to-shallow feedback.