From Hacker News Front Page · 40 stories
Nvidia to Extend CUDA Support to RISC-V, Outlining Server-Grade Requirements
Nvidia plans to extend its CUDA platform to support RISC-V CPUs, enabling RISC-V processors to integrate with GPU compute applications. This expansion requires RISC-V CPUs to meet specific server-grade criteria, including adherence to server SoC and platform specifications, ACPI support, and PCIe coherency.
Shipyard to Cease IPFS Maintenance and Operations by September 2026 Due to Funding Cut
Shipyard, a key contributor to the InterPlanetary File System (IPFS) ecosystem, announced it will wind down all IPFS-related engineering, maintenance, and infrastructure operations by September 30, 2026, after Protocol Labs decided not to renew its funding. This cessation will leave numerous core IPFS projects and public infrastructure without dedicated maintainers, impacting the future development and stability of the decentralized web technology.
GPT-5.6 Sol input token pricing reduced by 20%, output by 33%
The GPT-5.6 Sol model now costs $4 per million input tokens and $20 per million output tokens, representing a 20% reduction in input pricing and a 33% reduction in output pricing. This pricing change makes the frontier model in the GPT-5.6 family more accessible for developers and applications, impacting the cost-effectiveness of integrating advanced AI capabilities.
Security Researcher Discovers Vulnerabilities in Electric Scooter Backend
A security researcher identified multiple vulnerabilities in an electric scooter company's backend systems, including an exposed administrative panel and potential for unauthorized scooter unlocking. This highlights common security oversights in IoT deployments and the importance of thorough access control.
Omacom Foundation Launches with $8 Million to Support Omarchy Quattro Open-Source Project
The Omacom Foundation has been established with $8 million in funding from eight founding patrons, including CEOs of major tech companies. The foundation will support the open-source Omarchy Quattro project, aiming to advance its development and broader availability.
Study finds pilots and flight attendants have highest cancer mortality rates from radiation
A new study published in JAMA Internal Medicine indicates that flight attendants and pilots have the highest mortality rates from radiation-related cancers among over 500 occupations. This research suggests that occupational radiation exposure, rather than lifestyle factors, contributes to these elevated cancer risks due to increased exposure to cosmic radiation at high altitudes.
Kino Web Server for Ruby 4.0+ Uses Ractors for Parallel Performance and Reduced Memory
Kino is a new web server for Ruby 4.0+ that utilizes Ractors to enable parallel execution of Ruby code across multiple CPU cores within a single process, addressing the Global Virtual Lock (GVL) limitation. This approach results in significant performance gains and reduced memory consumption compared to traditional forking servers like Puma.
Codex on AWS Bedrock Lacks Explicit Cache Controls, Causing High GPT-5.6 Sol Costs
A bug in the native Amazon Bedrock provider for Codex CLI prevents explicit prompt caching for GPT-5.6 Sol, leading to significantly higher costs due to excessive cache-write tokens. This issue impacts agentic coding workloads where stable instructions are followed by changing content, which should benefit from caching.
Law School Prohibits AI Use for Academic Work to Preserve Core Legal Skills
A law school has implemented a policy, effective Summer 2026, prohibiting students from using AI for conceptualizing, outlining, drafting, revising, translating, or editing any work submitted for credit. This policy aims to ensure students develop essential cognitive skills for legal practice and uphold ethical obligations, while allowing instructors to deviate for courses specifically designed to teach AI fluency.
Scientists Release Largest 2D Map of the Universe with 5.6 Trillion Pixels
The DESI Legacy Imaging Surveys team released the largest 2D color map of the universe, containing nearly 4 billion celestial objects and covering 75% of the sky. This publicly available data will aid researchers and citizen scientists in studying phenomena like dark matter, dark energy, gravitational lenses, and supernovae.
Linux 7.2 Kernel Released with Cache-Aware Scheduling and Raspberry Pi GPU Power Management
Linux kernel version 7.2 has been released, introducing new features like cache-aware scheduling, improvements to MGLRU, and runtime power management for Raspberry Pi 4 and 5 GPUs. This update continues a trend of busy development cycles, bringing optimizations and new capabilities to the Linux ecosystem.
Slack introduces 'Code channels' for AI-assisted software development collaboration
Slack has launched "Code channels," a new feature designed to integrate AI development directly into the platform, enabling multi-player collaboration between humans and AI agents on software projects. This allows teams to work with coding agents like Claude Tag or ChatGPT within Slack to fix bugs, update web pages, or build new features.
Bun 1.4 Released with Increased Node.js Compatibility and Performance Improvements
Bun 1.4 has been released, adding over 1,500 Node.js test suite compatibility improvements, fixing more than 2,900 issues, and introducing new APIs. The update also includes performance enhancements such as reduced CPU and memory usage, and faster startup times on Linux.
Xorg-Server 26.0.99.901 Released as First Release Candidate for 26.1.0
Xorg-Server 26.0.99.901 has been released as the first release candidate for the upcoming 26.1.0 version, including updates for Xorg, Xephyr, Xnest, Xvfb, Xwin, and Xquartz. This release introduces significant changes like the removal of the autoconf/automake build system in favor of Meson, and adds support for DPMSInfoNotify events and XFixes 6.1.
Google Research Develops PhotoScan for Insulin Resistance Prediction from Smartphone Imagery
Google Research has developed PhotoScan, a deep learning method that estimates body composition from smartphone photos to predict insulin resistance with accuracy comparable to DXA scans in a clinical research setting. This development offers a more accessible and less invasive alternative to traditional methods for early detection of metabolic disease drivers.
CSS Working Group Resolves to Add Class Prefix Selector to CSS Selectors Level 5
The CSS Working Group resolved to add a new Class Prefix Selector (`.prefix-*`) to the CSS Selectors Level 5 specification, allowing developers to target multiple classes sharing a common prefix more efficiently. This feature simplifies styling for utility classes and design systems by reducing the need for multiple class listings or brittle attribute selectors.
XWayland 26.1-rc1 Released with Clipboard Bridge and Multi-seat Support
XWayland 26.1-rc1 has been released, marking the first release candidate for the standalone XWayland 26.1.0, two years after the previous 24.1 branch. This update introduces a clipboard/primary selection bridge for rootful XWayland, multi-seat support via Xi2, and removes EGLStream support, improving integration and functionality within Wayland environments.
New adaptation method ports task knowledge across frozen models with high accuracy
A new method allows task adaptations to be learned once in a base-agnostic form and then ported to new frozen models by refitting a thin per-base alignment. This approach recovers approximately 98% of per-task LoRA's performance on unseen models within the same family and 94% across different families.
Claude Opus 4.8 and 5.0 criticized for incoherent and toxic language style
Users report that Claude Opus 4.8 and 5.0 models exhibit a verbose, jargon-heavy, and over-stylized writing style, making responses difficult to understand. This issue, widely discussed on Reddit, affects the chat/web UI, Claude Code, and Fable 5, leading some users to consider alternative AI providers.
Remote Workers Report Highest Well-Being in Study of 7,700 Employees
A study of 7,704 employees at a large healthcare organization found that fully remote workers reported the highest levels of well-being, while onsite workers reported the lowest. The research also indicated that remote workers felt connected to colleagues and workplace culture, challenging common employer concerns about remote work isolation.
MicroGPT in pure C achieves 10M tokens/sec inference on Apple M5
A character-level transformer, MicroGPT, implemented in pure C with no external dependencies beyond libc, achieved over 10 million tokens per second inference speed on an Apple M5 chip. This implementation demonstrates efficient GPT training and inference in a minimal C environment, highlighting performance capabilities on modern hardware.
Mojo 1.0 Language and Tooling Now Open Source Under Apache 2.0 License
Modular has open-sourced the Mojo 1.0 language, compiler, and tooling under an Apache 2.0 license, making the AI programming language fully available to the public. This move aims to expand Mojo's adoption and allow developers to extend the language and build applications across various platforms, including new native Windows support.
Children's Lung Function Improves in London's Ultra Low Emission Zone Study
A study found that children's lung growth, previously stunted by pollution, accelerated significantly after the introduction of London's Ultra Low Emission Zone (ULEZ) in 2019. This research provides strong evidence that local clean air zones can mitigate the harm caused by childhood pollution.
Solo Enables Static Musl Binaries to Load Glibc-Linked GPU Drivers on Linux
Solo is a new tool that allows fully static musl-linked Linux binaries to load glibc-linked shared objects, specifically GPU drivers, at runtime. This development addresses a long-standing challenge for static binaries needing hardware acceleration, offering a deployment method without containers or a second libc in the process.
Turbovec: A Rust Vector Index with Python Bindings Based on Google's TurboQuant Algorithm
Turbovec is a new Rust vector index with Python bindings that implements Google Research's TurboQuant algorithm, offering significant memory reduction and faster search performance compared to FAISS. This development provides an alternative for vector search in applications prioritizing privacy, memory efficiency, or low latency, particularly for Retrieval Augmented Generation (RAG) stacks.
Nova3D Generates Programmable 3D Assets as Executable Blender Source Code
Researchers developed Nova3D, a system that generates 3D assets as executable Blender source code, treating the compiled mesh as an artifact rather than the asset itself. This approach allows for semantic handles at generation time, enabling downstream systems to inspect, measure, edit, and animate the generated 3D objects.
Linux 7.3 to include VRAM management improvements for better gaming performance
Kernel patches designed to improve VRAM management for games have been merged upstream and are queued for inclusion in Linux 7.3. These changes aim to mitigate performance degradation when games exceed available VRAM by optimizing how memory is moved between GPU and CPU RAM.
California Approves Nation's First Tire Efficiency Standards to Save Drivers $1B Annually
California has approved the nation's first tire efficiency standards, which will be phased in starting in 2029, to ensure replacement tires are as energy efficient as original equipment tires. This regulation is projected to save California drivers approximately $1 billion per year in refueling costs and reduce carbon dioxide emissions by 2 million tons annually.
Roboflow Launches Playground for Side-by-Side Comparison of 30+ Computer Vision Models
Roboflow introduced Playground, a new tool that allows users to compare the outputs of over 30 zero-shot computer vision models side-by-side using the same image and prompt. This platform removes the need for individual API provisioning or infrastructure setup, simplifying the evaluation process for various vision tasks like object detection, classification, and OCR.
Researchers develop 'Red Queen Gödel Machine' for co-evolving AI agents and evaluators
Researchers, including collaborators from NVIDIA and Flower Labs, have introduced a new method called the 'Red Queen Gödel Machine' that allows self-improving AI agents and their evaluators to evolve together. This approach addresses the problem of AI agents hitting an "evaluation ceiling" where improvement stalls due to fixed benchmarks, potentially enabling continuous self-improvement and reducing computational costs.
Buf releases first fully-featured LSP server for Protobuf development
Buf has launched the first production-grade Language Server Protocol (LSP) server for Protobuf, enabling modern IDE features like go to definition and code completion. This release integrates Protobuf with popular editors such as VSCode and Neovim, providing enhanced developer experience for the schema language.
MathCode AI Assistant Converts Math Problems to Lean 4 Theorems and Proofs
MathCode is a new terminal AI coding assistant that formalizes mathematical problems into Lean 4 theorems and attempts to prove them. This tool integrates a persistent Lean REPL, reusable theorem libraries, and agentic proving, which could accelerate mathematical formalization and verification processes.
NIH to discontinue K12 grant for early-career clinical researchers
The National Institutes of Health (NIH) is ending its K12 grant program, which has supported early-career clinical researchers for over two decades. This decision will impact the training and career paths of aspiring clinical scientists, particularly those from underrepresented backgrounds.
Abdominal Fat Predicts Heart Disease Risk Better Than BMI, Study Finds
A study published in JACC indicates that waist circumference (WC) and waist-to-hip ratio (WHR) are better indicators of cardiovascular disease risk than Body Mass Index (BMI) alone. The research suggests that relying solely on BMI can misclassify cardiovascular disease risk, as central adiposity is more strongly linked to adverse cardiovascular outcomes.
Deep cervical lymphatic-venous anastomosis (dcLVA) surgery for Alzheimer's enters global trials
A surgical procedure called deep cervical lymphatic-venous anastomosis (dcLVA), which aims to improve brain waste drainage, is entering trials worldwide after gaining traction in China. This technique involves connecting lymphatic vessels in the neck to veins, theoretically allowing waste proteins to flow more easily into the bloodstream, and could offer a new approach to treating Alzheimer's disease.
New Method Uses 2D Gaussian Splatting for Bézier Spline Line Art Vectorization
Researchers from Disney and ETH Zurich introduced a new method for line art vectorization that uses 2D Gaussian splatting to convert raster images into Bézier splines. This approach improves stroke extraction by leveraging depth prediction and semantic feature models, and enables efficient fitting of strokes with joint optimization of control points and brush texture.
GCC 16 Guarantees Trampoline-Free Nested Functions Without Parent Context Access
GCC 16 guarantees that nested functions not accessing parent variables will not require run-time trampolines, enhancing security by removing the need for an executable stack. Additionally, a patch implementing new built-in functions for wide pointers in nested functions was merged into GCC's development branch, aiming to improve how nested functions handle parent context access without trampolines.
IETF TLS Working Group Opposes NSA-backed Proposal to Remove ECC from Hybrid ML-KEM
The Internet Engineering Task Force (IETF) TLS working group has largely opposed a specification, supported by the NSA, that would remove Elliptic Curve Cryptography (ECC) from hybrid ECC+ML-KEM in TLS. This opposition highlights concerns about the security implications of relying solely on ML-KEM without the added protection of ECC, particularly given the potential for future mathematical breaks in ML-KEM.
RustDesk Adds Unattended Remote Access for Wayland on Debian/Ubuntu x86_64
RustDesk now supports true unattended remote access on Wayland for x86_64 Debian/Ubuntu-based systems, including multi-monitor setups. This development addresses a long-standing challenge in Linux remote desktop solutions, as many competitors still lack robust Wayland support.
Toast 1 Search Agent Released, Outperforms GPT-5.6 Sol and Claude Opus 5 in Cost and Speed
Toast 1, a specialized search agent, has been released, offering frontier search quality at a lower cost and faster speed than existing models like Claude Opus 5 and GPT-5.6 Sol. This development is significant because it provides a more economical way for AI models to perform knowledge work by specializing the search loop, freeing up generalist models for reasoning and final answer generation.