From Hacker News Front Page · 40 stories
Live Migration for bhyve Guests on FreeBSD 15.1 Reaches First Working State
Live migration for bhyve guests on FreeBSD 15.1 is now in a first working state, allowing running amd64 guests to move between hosts over TCP with minimal interruption. This development builds on bhyve's experimental snapshot support and introduces dirty-page tracking in the vmm kernel module, along with a migration protocol in bhyve and bhyvectl. This capability is important for enterprise virtualization, enabling host maintenance and load balancing with less disruption, and is central to projects like Sylve and BoxyBSD.
Experiments with LeWM JEPA world model show planning improvements without retraining
A researcher conducted 600 experiments on LeWM, a JEPA world model, to evaluate its predictive capabilities and planning methods. The experiments revealed that LeWM's predictions are comparable to perfect physics for planning on trained paths, and new planning methods significantly improved its success rate on long-horizon tasks without retraining the model.
Anthropic Investigates Authentication Issues Across Claude.ai and Related Services
Anthropic is investigating an issue affecting user authentication on Claude.ai, Claude Code, and platform.claude.com. This incident impacts access to multiple Claude services, including Claude Console and Claude Cowork.
RealtimeVAD detects end-of-speech up to 110ms faster than Silero VAD
RealtimeVAD is a new voice activity detection (VAD) tool that identifies the end of speech significantly faster than the Silero VAD, by combining Silero's speech detection with a real-time audio level comparison. This earlier detection can reduce latency in voice assistant interactions, as subsequent processes like speech recognition and LLM calls can begin sooner.
AI-generated apps overwhelm Apple's App Review, delaying synth app releases
Apple's App Review process is experiencing significant delays, with some app submissions taking weeks instead of days, due to a surge in new iOS apps, many of which are AI-generated. This backlog is impacting developers, particularly those in the music app sector, who face delayed releases and difficulties with cross-platform synchronization.
Data Centers Shift Infrastructure Costs to Lower-Income and Rural Residents
New Cornell research indicates that data centers are disproportionately shifting the costs of energy infrastructure upgrades to residents, particularly those in lower-income and rural areas. Data centers paid only 38% of the $8.25 billion infrastructure bill in 2025 related to their growth, leaving residents to cover the remaining 62%, with some low-income households facing energy cost increases up to 28% of their income.
Astronomers discover new dwarf planet candidate 2026 ED138 with 281-year orbit
Astronomers Matthew Holman and Kevin Napier, using data from the Vera C. Rubin Observatory, identified a new celestial object, 2026 ED138. This object is 46 times farther from the sun than Earth and has an orbital period of 281.1 years, potentially qualifying it as a new dwarf planet.
Google Antigravity Introduces Gemini 4 Hydrogen and Orchestrator for AI Model Management
Google Antigravity has launched Gemini 4 Hydrogen and a new Orchestrator feature. Orchestrator dynamically switches between Gemini 4 Carbon and Gemini 4 Hydrogen to optimize for quality, efficiency, and latency in AI tasks, achieving near-Carbon success rates with reduced quota consumption and faster completion times.
GitHub Actions, Pages, and Webhooks Experienced Degraded Performance
GitHub reported and resolved an incident causing degraded performance for Actions workflows, Pages deployments, and webhook events. The issue also affected code search and live updates, but services have since recovered.
Apple Researchers Propose Neural Network Approach for Faster RAG Retrieval
Apple researchers presented a paper at ICML 2026 detailing a new method for Maximum Inner Product Search (MIPS) in Retrieval-Augmented Generation (RAG) systems. Their approach uses small neural networks, trained offline, to route queries more efficiently to their best matches, improving retrieval speed and accuracy. This method addresses the computational cost of MIPS with large datasets and the differences between query and key vectors in RAG.
Multiple Open-Source Projects Ban AI-Generated Contributions
A growing number of open-source projects, including Immich, Meilisearch, and Servo, are explicitly prohibiting contributions generated by large language models (LLMs) and other generative AI tools. These bans cover various forms of content, such as code, documentation, pull request descriptions, and issue comments, reflecting concerns about AI-generated content in their repositories.
Ginux Transitions from Standalone Kernel to Configurable Userspace Kernel
Ginux, a project by Greg Kurtzer, has shifted from being an independent Rust kernel to operating as a configurable userspace kernel. This change allows Ginux to run as an ordinary Linux process, providing isolated Linux compatibility instances with a defined, versioned contract between managed workloads and the host kernel.
U.S. Imposes New Rules on Synthetic DNA Ordering to Deter Bioterrorism
The U.S. government has introduced new regulations for companies that synthesize DNA, requiring them to screen all orders for sequences that could be used to create dangerous pathogens. This measure aims to prevent the misuse of synthetic DNA technology for bioterrorism, expanding on previous voluntary guidelines.
LLMs Solve and Optimize Puzzles in Zachtronics' TIS-100 Programming Game
Sixteen large language models (LLMs) were tested on Zachtronics' TIS-100, a low-level assembly programming game, to assess their ability to solve and optimize complex programming puzzles. Opus 5.5 and GPT-6 Astra demonstrated the highest performance, solving all 55 puzzles and discovering new optimization frontiers. This experiment highlights LLMs' capabilities in code generation and optimization within constrained, interactive environments.
Tendrils Raises $30M to Develop Parallel-First CPU Architecture and Vine Language
Tendrils secured $30 million in funding, led by Benchmark, to develop a new CPU architecture designed for parallel processing. This initiative aims to move beyond legacy sequential CPU designs by creating processors with thousands of cores and an accompanying open-source language called Vine.
Apple Developing 'Harbor' Mac App for Local AI Model Inference and Clustering
Apple is developing a new Mac application named Harbor, which allows users to run open-weight AI models locally on Apple silicon devices. The app can pool unified memory across multiple Thunderbolt 5-connected Macs to handle larger models and higher concurrency, positioning Macs as private AI inference servers.
Mozilla Study Measures Browser Fingerprinting Effectiveness, Firefox 143 Reduces Uniqueness
Mozilla conducted a large-scale study on browser fingerprinting using data from 25 million Firefox users to measure how identifiable browsers are and identify unique characteristics. The study informed the development of fingerprinting protections in Firefox 143, which reduced the percentage of uniquely identifiable users from 75% to 46%. This research provides real-world data on fingerprinting effectiveness and demonstrates a significant reduction in user identifiability through browser-level mitigations.
Modern Web Guidance CLI helps coding agents use current web platform features
Modern Web Guidance is a new CLI tool designed to update coding agents' knowledge from legacy JavaScript patterns to modern web primitives. It provides expert-curated guidelines on current browser APIs, performance, accessibility, and responsible fallbacks, addressing the issue of LLMs generating outdated code due to training data bias.
Claude launches Whiteboards and Sheets in beta, rebrands Dashboards to Data
Claude has released Claude Whiteboards and Claude Sheets in beta, and rebranded Claude Dashboards to Claude Data, making it generally available. These new tools allow users to brainstorm, create spreadsheets, and analyze data directly within the Claude environment, facilitating collaborative work with the AI.
Blender 5.3 Introduces Gaussian Splat Support, NVIDIA DLSS, and Project System
Blender 5.3 has been released, adding support for Gaussian Splatting, integrating NVIDIA DLSS for viewport denoising, and introducing a new project system. These updates provide new rendering capabilities, improve real-time workflow, and offer better organization for complex productions.
Ableton Live 13 Introduces New Audio/MIDI Workflows and Sound Design Tools
Ableton Live 13 has been released, adding new audio and MIDI workflows, sound design tools, and mixing effects. This update includes features like a built-in Step Sequencer, ARA 2 support, and expanded mixer controls, providing new capabilities for music production.
.NET 11 Runtime Updates Minimum Hardware Requirements for x86/x64 and Arm64
.NET 11 has updated its minimum hardware requirements for both x86/x64 and Arm64 architectures, requiring more modern instruction sets. This change means .NET 11 will not run on older CPUs that lack these instruction sets, and ReadyToRun compilation targets have also been updated to utilize newer hardware capabilities.
Queens Public Library Suffers Ransomware Attack, Suspending E-Borrowing Services
The Queens Public Library (QPL) is experiencing a ransomware attack that has suspended e-borrowing services and impacted some virtual and in-person programs. Library locations remain open for borrowing and returning physical materials, but renewals are unavailable as the team works to restore network and systems access.
Amazon Replaces Claude Sonnet 4.6 with Strands Decider for Internal Ticket Labeling
Amazon and AWS teams replaced Claude Sonnet 4.6 with their internal Strands Decider model for labeling tickets in an internal system. This change reduced token costs by 98% while maintaining performance, projected to save millions of dollars annually.
CoreWeave bond yield exceeds 16% amid rising AI credit stress
The yield on CoreWeave's 2032 bond has risen above 16%, and its credit default swaps (CDS) spiked, indicating increased investor concern over AI data center financing. This stress in the credit market is partly attributed to potential regulatory hurdles for data centers under new political leadership in key states. The situation suggests a potential increase in financing costs for AI infrastructure development.
AI Agents Formalize CPython Internals in Lean, Uncovering 418 Bugs
AI agents, specifically Opus 5.5, have translated 613,318 lines of CPython's internals into Lean, a language for machine-checked proofs. This process involved rewriting functions line-by-line, testing them against real Python, and having a second AI agent verify the work, resulting in 4,071 functions passing verification. The formalization effort identified 418 bugs in CPython, with approximately 55 capable of causing interpreter crashes or memory corruption.
Claude Fable 5.2 Achieves Highest Score on Artificial Analysis Intelligence Index
Claude Fable 5.2 (Max, Default Fallback) scored 65 on the Artificial Analysis Intelligence Index, making it the highest-scoring model to date. This places it significantly above the median intelligence for comparable models, though it is also noted for its high cost and slow performance per task.
E-means algorithm achieves 10x faster k-means clustering with lower loss
A new k-means clustering algorithm, e-means, demonstrates significantly faster performance and lower test loss compared to existing methods. E-means achieves a 10-45x wall-clock speedup on GPU and 8-15x on CPU, while also reaching a lower test loss in fewer epochs. This is achieved through mini-batch updates with exponential moving averages, automatic hyperparameter tuning, and improved initialization strategies.
DeepX Tapes Out DX-M2 AI Accelerator for Edge Devices, Claims 100B Parameter Inference at Under 5W
South Korean startup DeepX has taped out its second-generation AI accelerator, the DX-M2, which integrates Samsung's custom processor-in-memory (PIM) technology. The DX-M2 is designed for generative AI inference on edge devices like robots and drones, supporting models up to 100 billion parameters while consuming less than 5 watts. This development aims to make large AI models practical for power-constrained edge applications by significantly improving memory bandwidth and power efficiency.
Infineon Processes First Functional Power Devices on 300mm Silicon Carbide Wafers
Infineon Technologies has successfully processed functional power devices on 300-millimeter silicon carbide (SiC) wafers, marking an industry first. This achievement, utilizing existing silicon manufacturing equipment, lays the groundwork for more efficient SiC chip production, as 300mm wafers yield 2.3 times more chips than 200mm wafers.
Payments Canada Launches Real-Time Rail (RTR) Payment System for Instant Settlements
Payments Canada has launched the Real-Time Rail (RTR), a new national payment system enabling instant clearing and settlement of payments 24/7 for Canadian businesses and consumers. This infrastructure provides real-time cash-flow visibility and data-rich messaging, modernizing Canada's financial system and supporting future economic growth.
SpaceX Plans Starship Flight 15 to Catch Both Booster and Ship on November 13
SpaceX announced Starship Flight 15 is scheduled for November 13, aiming to catch both the Super Heavy booster and the Starship spacecraft. This mission marks an attempt to achieve full reusability for the Starship system, a key step in its development.
Outlook.com Quietly Rejects Email from Domains Without DMARC Records
Since November 5, 2026, Microsoft's consumer email services (Outlook.com, Hotmail, Live, MSN) have been rejecting some emails from domains that do not publish a DMARC record. This unannounced change affects even low-volume senders, causing intermittent email bounces for communications with these Microsoft addresses. The issue can be resolved by adding a simple DMARC TXT record to the domain's DNS.
Police Use Ad Tech to Track Users Through Popular Apps
Law enforcement agencies are utilizing data from advertising technology embedded in popular mobile applications to track individuals. This practice allows police to access location and other personal data without traditional warrants, raising privacy concerns.
Llama-modchip Enables 284B DeepSeek-V4-Flash on 64GB RAM Without GPU
Llama-modchip, a set of kernel patches and tools, allows the 284B DeepSeek-V4-Flash Mixture-of-Experts (MoE) model to run entirely on 64GB system RAM without a dedicated GPU. It achieves this by reducing the model's checkpoint size to 56.56 GiB and includes CPU cache and SIMD kernel patches for improved performance. This development makes large language models more accessible on consumer-grade hardware.
GitHub launches new pull request overview page in public preview
GitHub introduced a new "Overview" page for pull requests in public preview, which consolidates merge status and actions at the top of the page. This update aims to reduce scrolling and help developers quickly identify blocking issues and next steps in long pull requests. The new page also includes features to narrow conversations and sort comments.
Microsoft Research's AI Frontiers Group Reorganized into Microsoft AI
Microsoft Research's AI Frontiers (AIF) sub-organization, formed in 2023 to coordinate AI research, has been reorganized and integrated into Microsoft AI. This change aims to consolidate AI efforts, though it merges a smaller, focused group with a larger existing organization.
Video Rebirth releases HyperFlow H3 Flash, a 14B parameter video+audio diffusion model
Video Rebirth has released HyperFlow H3 Flash, a compressed version of its MiniMax-H3 model, which generates video and synchronized stereo audio. This new model is significantly smaller at 13.7B-13.9B parameters, down from ~31B, and can generate 5.2-15 second clips in 4-8 steps, enabling faster inference and broader hardware compatibility.
macOS 13+ Includes Undocumented Daemons for Personal Knowledge Graph Construction
Every Mac running macOS 13 or later contains undocumented daemons that build a personal knowledge graph about the user and their contacts. This graph cross-references data from various applications and behavioral signals to infer relationships and activities. The system is reportedly queryable by LLMs and lacks official documentation regarding its full scope and integration.
Asahi Linux Progresses with Linux 7.3, M3 Support Now Generally Available
Asahi Linux has released its Linux 7.3 progress report, detailing continued upstreaming efforts for Apple Silicon drivers like macaudio and AVD. M3 series device support is now generally available without an EXPERT flag, following successful early adoption.