Alvaro Lopez Ortega / 2025-10-26 Briefing

Created Sun, 26 Oct 2025 19:08:36 +0000 Modified Mon, 27 Oct 2025 02:08:45 +0000
6266 Words

Today’s tech news highlights include the U.S. and China finalizing the TikTok deal with U.S. control and Oracle security, California’s $15.7 billion investment boosting renewable energy resilience, and Amazon testing AI-powered smart glasses to improve delivery safety and efficiency. Meanwhile, the DynamoDB DNS race condition caused a major AWS outage with billions in damages.

▶️ Internet Infrastructure

Libera.Chat’s legal assessment indicates it likely does not meet UK links criteria under the OSA, with minimal UK user base and no targeted UK market, reducing enforcement risk.

  • Libera.Chat’s legal firm concluded it likely lacks sufficient links to the UK for the Online Safety Act (OSA) to apply
  • The organization provides a regulated U2U service with broad scope, including IRC chat, which counts as a regulated service under OSA
  • UK links are assessed based on significant UK user base, target market, or potential harm; Libera.Chat has minimal UK user presence and no targeted UK market

How Linux Kernel Loads ELF Binaries During Execution

The article details how the Linux kernel loads ELF binaries during execve(), including parsing headers, mapping segments, handling dynamic linking, and setting up the user stack with arguments and auxiliary vector.

  • The article explains the Linux kernel’s process for executing ELF (Executable and Linkable Format) binaries, the primary binary format on modern Linux systems.
  • Loading an ELF binary involves reading the ELF header, processing program headers (PT_LOAD, PT_INTERP, PT_GNU_STACK), and setting up memory segments, stack, and auxiliary vector.
  • The load_elf_binary() function manages ELF loading, including clearing old execution state, mapping segments, setting credentials, and preparing the user stack with arguments, environment, and auxiliary data.

NetBSD Foundation Seeks $50K Funding to Boost RISC-V and Wi-Fi Development

NetBSD Foundation seeks additional funding to reach $50,000 goal by year-end, supporting RISC-V expansion and wireless stack updates to enhance NetBSD’s hardware compatibility and security.

  • NetBSD Foundation has raised $10,738 of a $50,000 annual goal as of October 26, 2025
  • The foundation aims to close a $39,262 funding gap before year-end
  • Donations support RISC-V hardware support and Wi-Fi stack modernization

U.S. and China to Finalize TikTok Deal with U.S. Control and Oracle Security

U.S. and China are set to finalize the TikTok deal on Thursday in Korea, with all details agreed, involving U.S. control over TikTok’s core operations and Oracle managing security, following a September framework agreement.

  • Trump and Xi are scheduled to finalize the TikTok deal on Thursday, according to Treasury Secretary Scott Bessent.
  • The U.S. and China reached a “final deal” during discussions in Madrid, with all details reportedly finalized for the leaders to consummate in Korea.
  • The deal follows a framework agreement from September, with Trump signing an executive order to facilitate the transaction, involving U.S. control over TikTok’s recommendation algorithm, source code, and content moderation, with Oracle responsible for security operations.

Synadia and TigerBeetle pledge $512K to support Zig language development

Synadia and TigerBeetle committed $512,000 to the Zig Software Foundation, emphasizing Zig’s safety, simplicity, and suitability for TigerBeetle’s high-performance, correctness-critical database development.

  • Synadia and TigerBeetle pledged $512,000 to the Zig Software Foundation over two years
  • The pledge supports Zig language development, leadership, and community building
  • TigerBeetle was developed in Zig starting in 2018, chosen over C and Rust for safety, simplicity, and performance

Chipmakers Use RSUs to Retain Talent Amid AI Boom

Chipmakers Nvidia, AMD, and Broadcom leverage long-term RSU vesting to retain talent during AI growth, with some employees earning over $2 million, significantly increasing retention and wealth.

  • Nvidia, AMD, and Broadcom use stock-based pay, including RSUs, to retain employees amid AI-driven valuation surges.
  • Stock payouts vest over multiple years, with some employees earning over $2 million from RSUs, incentivizing long-term retention.
  • Nvidia’s stock value increased significantly since 2023, with some RSUs valued over six times their original worth, contributing to reduced turnover rates.

California’s $15.7B Battery Storage Boost Eliminates Flex Alerts and Supports Renewables

California’s $15.7 billion investment in battery storage since 2020 has enabled the grid to operate without Flex Alerts since 2022, replacing aging gas plants and supporting 60% renewable energy.

  • California’s battery energy storage capacity increased over 3,000% from 2020 to 2025, reaching more than 15,700 MW.
  • The state has not issued a Flex Alert since 2022, credited to expanded battery storage and grid resilience efforts.
  • CAISO overseen over 26,000 MW of new capacity since 2020, reducing reliance on fossil fuels and preventing blackouts.

Apple Maps to Introduce Paid Ads by 2026 Amid Consumer Concerns

Apple plans to add paid ads in Maps by 2026, enabling businesses to pay for enhanced visibility, using AI for relevant results, amid potential consumer concerns.

  • Apple Maps may introduce integrated ads as early as 2026, according to Bloomberg’s Mark Gurman
  • The project will allow businesses like restaurants to pay for prominent placement in search results
  • Similar to App Store Search Ads, leveraging AI for relevant results, with potential consumer backlash

Synadia and TigerBeetle donate $512K to Zig Foundation for high-performance software

Synadia and TigerBeetle committed $512,000 to the Zig Software Foundation to promote Zig’s role in high-performance, reliable systems software development, aligning with their focus on correctness and simplicity.

  • Synadia and TigerBeetle pledged a combined $512,000 USD to the Zig Software Foundation over two years
  • The donation supports Zig’s development, emphasizing performance, reliability, and maintainability
  • Zig is designed for systems, embedded, and high-performance applications, led by Andrew Kelley

Amazon Tests AI Smart Glasses to Enhance Delivery Efficiency and Safety

Amazon is deploying AI-enabled smart glasses with cameras and computer vision to improve last-mile delivery accuracy, safety, and efficiency, supported by its $16.7 billion Delivery Service Partner program.

  • Amazon is testing AI-powered smart glasses with mounted cameras, computer vision, and “AI-powered sensing” to assist delivery drivers.
  • The glasses project turn-by-turn directions and delivery instructions directly into the driver’s view, supporting a hands-free experience.
  • Features include automatic activation upon parking, parcel identification, navigation assistance, and a controller with battery, emergency contact, and support for prescription and transitional lenses.
  • Future updates may detect misdelivered parcels, identify loose pets or hazards, and adapt to low-light conditions.
  • Developed under Amazon’s $16.7 billion Delivery Service Partner program, with feedback from hundreds of drivers, aiming to improve safety and delivery efficiency.
  • Critics highlight concerns over wearable cameras and continuous tracking, questioning privacy and surveillance implications.

DynamoDB DNS Race Condition Triggers Major AWS Outage and Billions in Damage

A race condition in DynamoDB’s DNS system caused an empty DNS record, cascading into widespread AWS service outages for hours, with damages possibly reaching hundreds of billions of dollars.

  • A race condition in DynamoDB’s DNS management system caused an empty DNS record for the regional endpoint on October 19, 2025, at 11:48 PM PDT
  • The fault led to DNS failures, impacting DynamoDB API access, EC2 instance launches, network configuration, and dependent AWS services
  • The incident resulted in a day-long outage affecting major websites, government services, and estimated damages potentially reaching hundreds of billions of dollars

AWS Launches Automated Incident Reports for Rapid CloudWatch Analysis

AWS introduced an automated incident report generator for CloudWatch, enabling rapid, comprehensive post-incident analysis by correlating telemetry, configurations, and investigation data.

  • AWS announced an interactive incident report generation feature for CloudWatch on October 22, 2025
  • The service automatically collects telemetry data, input, and actions during investigations, producing detailed reports within minutes
  • Reports include summaries, event timelines, impact assessments, and recommendations to improve operational response

Microsoft Explores Copilot AI for On-Premises Exchange Server

Microsoft is investigating integrating Copilot AI into Exchange Server on-premises through a survey, emphasizing data privacy, compliance, and offline operation, despite no formal plans.

  • Microsoft is exploring adding Copilot AI assistance to on-premises Exchange Server, via a survey asking about data sharing and compliance requirements.
  • The survey includes questions on capabilities like email summarization, server health monitoring, and non-negotiable requirements such as regulatory compliance, data boundaries, and disconnection from the internet.
  • While not officially planned, the direction indicates potential future integration, raising concerns among administrators about data privacy and on-premises control.

Alaska Airlines Grounds Fleet After Major Data Center Outage

Alaska Airlines experienced a major IT outage at its primary datacenter on October 23, causing a fleet ground stop, over 360 cancellations, and passenger disruptions, with no cybersecurity link.

  • Alaska Airlines grounded its fleet on October 23 at 3:30 pm PT due to a failure at its primary datacenter, with no indication of a cybersecurity event.
  • Over 229 flights were canceled during a system-wide ground stop lasting until 11:30 pm PT; the airline reported more than 360 cancellations as of the update.
  • The outage impacted key operational systems, with no safety compromise; this is the second outage since July, following a previous incident (July grounding).

Microsoft Releases Emergency Patch for Critical WSUS Vulnerability

Microsoft issued a surprise out-of-band update to fix a critical remote code execution flaw (CVE-2025-59287) in WSUS on Windows Server 2012-2025, highlighting WSUS’s declining support status.

  • Microsoft released an out-of-band patch addressing CVE-2025-59287, a critical remote code execution vulnerability in Windows Server Update Services (WSUS).
  • The flaw affects Windows Server versions 2012 through 2025, stemming from insecure deserialization of untrusted data; a public proof-of-concept exists.
  • The update is cumulative, includes October patches if not installed, and requires a reboot; affected servers with WSUS role enabled are vulnerable.

Smart AI Network Design to Boost Sustainability and Capacity by 2033

Building sustainable AI starts with smart network design, integrating AI into infrastructure to optimize energy use, support 60% of network traffic by 2033, and enable local edge processing.

  • AI server and storage infrastructure spending projected to reach $200bn by 2028, mainly supported by service providers (IDC, August 2024)
  • Data centers accounted for 1.5% of global electricity demand in 2024, expected to rise to 3% by 2039; data transmission networks have similar energy use
  • ZTE emphasizes integrating AI into network infrastructure, including AI-enhanced optical, RAN, and core networks, supporting capacities up to 1.6T per wavelength and nearly 100T single fiber capacity

▶️ Open Source

Clojure Land Unites Open-Source Projects Spanning Web, AI, Graphics and More

Clojure Land consolidates open-source projects like Pedestal, Tesser, and Tapestry, supporting domains such as web development, AI, graphics, quantum computing, and distributed systems.

  • Clojure Land hosts multiple open-source projects including Pavlov (38 stars), ECA (446 stars), clojure2d (570 stars), Joyride (551 stars), Replicant (437 stars), Legba (32 stars), Typed Clojure (561 stars), hashp (474 stars), Component (2125 stars), calva-backseat-driver (36 stars), QClojure (37 stars), lacinia-pedestal (203 stars), Tesser (884 stars), ECA VS Code (4 stars), Pedestal (2750 stars), data.json (566 stars), clojure-cli-config (552 stars), and Tapestry (247 stars), and Daphne (45 stars).
  • Projects cover diverse domains: behavioral programming, AI pair programming, graphics, VS Code integration, web frameworks, JSON parsing, CLI configuration, asynchronous programming, quantum computing, GraphQL, distributed systems, probabilistic programming, and more.
  • Notable tools include Pedestal for web applications, Tesser for parallel reducers, and Daphne for probabilistic programming.

Org-social.el 2.3+ Enhances Feed Downloads with Concurrency and Range Requests

Org-social.el 2.3+ efficiently handles millions of lines in social.org files by parallel processing, HTTP Range fetching, and UI pagination, enabling scalable, bandwidth-optimized feed management.

  • Org-social.el 2.3+ uses concurrent queue processing and HTTP Range requests to optimize feed downloads
  • Implements a three-layer approach: parallel downloads, partial file fetching, and UI pagination
  • Maintains compatibility with servers like Apache, Nginx, Caddy, GitHub, but falls back on full downloads for Cloudflare and Codeberg
  • Downloads only recent posts within 14 days, with typical partial fetch size around 8 KB regardless of total feed size
  • Supports configurable parameters: org-social-max-concurrent-downloads (default 20) and org-social-max-post-age-days (default 14)
  • Performance benchmarks show feeds with over 100 posts download approximately 8 KB, independent of total size

Evaluating the Limitations and Successes of Top Generative Image Models

The article evaluates 14 generative image models on complex prompts, highlighting varying success rates, with Midjourney v7 and OpenAI 4o excelling in specific tasks, demonstrating current limitations in adherence and technical accuracy.

  • The article compares state-of-the-art generative image models across multiple prompts and challenges with emphasis on adherence.
  • Models tested include OpenAI 4o, Gemini 2.5 Flash, Imagen 4, Seedream 4, Flux.1 [dev], Midjourney v7, among others.
  • Performance varies significantly; some models succeed in specific tasks (e.g., Midjourney v7 and OpenAI 4o in 9 and 1 attempts respectively for nine-pointed star), while others fail (e.g., zero success in generating a prime number D20 die).

PCB Libraries for USB C Connectors Enable Direct PCB Integration

Repository offers KiCAD and EasyEDA libraries for PCB-edge USB C connectors (10 and 14 pin), enabling PCB integration as a USB C port; last update was October 25, 2025.

  • Repository provides KiCAD and EasyEDA libraries for PCB-based USB C connectors with 10 and 14 pin versions
  • Uses PCB itself as a USB C connector, enabling direct integration into PCB design
  • No releases or packages published; repository last updated on October 25, 2025

Oxdraw: Rust-Based Draggable Mermaid Diagramming Tool with Web Interface

oxdraw is a Rust-based diagramming tool that enables creating, editing, and styling Mermaid diagrams via CLI and a web interface, supporting draggable nodes, multi-line labels, and style overrides.

  • oxdraw is a diagram as code tool written in Rust with draggable editing capabilities, available on GitHub.
  • Combines Mermaid syntax for diagram creation with a web interface for fine-tuning positions, connector paths, and styles, with changes persisted as comments in the source.
  • Features include CLI commands (cargo install oxdraw, oxdraw --input, oxdraw --edit) and frontend controls for node/edge styling, node movement, and route adjustments, supporting multi-line labels with HTML breaks.

FSF Examines Licensing Challenges of LLM-Generated Code

The FSF is investigating legal and licensing issues surrounding LLM-created code, emphasizing transparency, copyrightability, and licensing implications, without current plans for license updates.

  • The FSF’s Licensing and Compliance Lab is exploring the intersection of free software licensing and large language models (LLMs), including potential guidance and survey of projects.
  • No immediate plans for a GPLv4; instead, the FSF considers adjustments to the Free Software Definition regarding LLM-generated code.
  • Concerns include non-free training data, copyrightability of LLM output, licensing terms, and the need for metadata collection (model info, prompts, restrictions) when accepting LLM-generated code.

NewPipe Celebrates 10 Years of Community-Driven Innovation and Resilience

NewPipe turned 10 in October 2025, with ongoing refactoring to modernize the app using Kotlin and Compose, while confronting legal and technical challenges from Google and maintaining community-driven development.

  • NewPipe celebrated its 10th anniversary in October 2025
  • The project was founded during the second COVID-19 pandemic wave, with ongoing development including a major refactor and new extractor design
  • The team established NewPipe e.V. to support development and funding, hiring paid personnel like @Profpatsch and Schabi in 2024 to ensure project stability
  • Current focus is on completing the app rewrite using Kotlin and Compose, aiming for a modern interface and features, while addressing challenges from Google’s efforts to restrict alternative app stores like F-Droid
  • The project faces increasing technical complexity, aging codebase, and legal threats from Google, but plans to continue development and seek creative solutions, emphasizing community contribution and resilience

ShadCN Themer Unlocks Custom Themes and Dark Mode for shadcn/ui

ShadCN Themer (website) allows users to build and share custom themes for shadcn/ui, featuring an intuitive editor and a collection of predefined themes with diverse color schemes and dark mode support.

  • ShadCN Themer enables creation, customization, and sharing of themes for shadcn/ui
  • Offers a theme editor for building color palettes and applying styles
  • Provides a gallery of pre-made themes with various color schemes and modes, including dark mode options

Apple Launches Swift SDK for Android Development in 2025

Apple released the Swift SDK for Android in October 2025, supporting cross-platform app development with official tools, Java bindings, and extensive setup instructions.

  • The first preview release of the Swift SDK for Android was published in October 2025, enabling development of Android apps in Swift with official tooling.
  • The SDK supports interoperability via swift-java, which generates bindings between Java and Swift, allowing use of Swift alongside Android APIs and Java code.
  • The SDK is available in nightly preview bundled with Swift for Windows or as a separate download for Linux and macOS; a comprehensive getting started guide is provided.

UC Berkeley Boosts Load Balancing Speed with AI-Driven Optimization

UC Berkeley researchers employed OpenEvolve to enhance EPLB load balancing, achieving a 5x speedup and demonstrating AI-driven optimization’s potential to outperform human-designed algorithms.

  • UC Berkeley researchers used OpenEvolve, an open-source implementation of Google DeepMind’s AlphaEvolve, to optimize a load balancing algorithm.
  • Achieved a 5x speedup for the Expert Parallelism Load Balancer (EPLB), reducing rebalancing time from 540 ms to 3.7 ms.
  • The optimized approach replaced loops with vectorized tensor operations and implemented zig-zag partitioning, using less than $10 and five hours.

Ghanaian Hand-Painted Movie Posters from the 80s-90s Turned Global Art Collectibles

Ghanaian hand-painted movie posters from the 1980s-90s, created on flour sacks with exaggerated, surreal imagery, became globally recognized collectible art reflecting resourceful cultural reinterpretation.

  • Ghanaian movie posters from the 1980s-90s were hand-painted on used flour sacks, featuring exaggerated, often grotesque imagery of Hollywood films.
  • Artists, working with minimal resources and referencing VHS covers or oral descriptions, created vibrant, chaotic posters with blood, explosions, and unrealistic character depictions.
  • These posters gained international recognition in the 2000s, becoming collectible art, exhibited in galleries worldwide, and regarded as a form of outsider art symbolizing ingenuity and cultural adaptation.

▶️ Software Development

Mastering Essential CS Tools: Command-Line, Git, Debugging, and Security Course

The course teaches practical CS tools mastery—command-line, editors, version control, debugging, security—through scheduled lectures from Jan 13-30, 2020, with resources on YouTube.

  • The course covers essential CS tools proficiency, including command-line, text editors, version control, debugging, and security, not typically taught in traditional classes.
  • Scheduled from January 13 to January 30, 2020, with lectures on topics like shell scripting, Vim, Git, debugging, and cryptography.
  • Video recordings are available on YouTube; instructors include Anish, Jon, and Jose.

Effective Code Reviews: Focus on Context, Limit Comments, and Approvals

Effective code review requires understanding the full system beyond diffs, limiting comments, and favoring approvals over blocking, with specific emphasis on context, quality, and collaborative standards.

  • The article highlights common mistakes in code reviews, emphasizing the importance of understanding the entire codebase rather than just the diff.
  • Recommends limiting comments to five or six per review to avoid overwhelming developers and losing critical feedback.
  • Advises reviewers to leave blocking reviews when they do not want changes merged and to favor approvals for most cases, especially in active SaaS environments.

▶️ Management and Leadership

Essential Skills and Strategies for Effective Principal Tech ICs

Effective principal tech ICs should develop diverse skills, influence across teams, mentor others, and focus on high-impact, strategic work while managing communication, time, and organizational relationships.

  • The article offers guidance for new principal tech ICs, emphasizing diverse roles, influence, communication, and strategic impact.
  • Key advice includes focusing on technical vision, connecting teams, mentoring, and managing time effectively to avoid being overwhelmed.
  • It highlights the importance of defining your charter, building organizational credibility, and balancing big-picture thinking with local solutions.

MyraOS: A Self-Developed Unix-Like OS with Virtual Memory and Real App Support

MyraOS is a self-developed x86 Unix-like OS with advanced features like protected mode, virtual memory, ext2 filesystem, and ELF loader, enabling real app execution and gaming.

  • MyraOS is an x86 Unix-like operating system built entirely from scratch, with features including protected mode, paging, virtual memory, and ext2 filesystem
  • Implements user-mode (ring 3) and kernel mode (ring 0), processes, scheduling, device drivers (PIT, RTC, Keyboard, Mouse, Framebuffer, PATA), and a UI compositor with window widgets and custom font
  • Supports ELF loader for running real applications, including Doom, and is capable of running on real hardware

How to Set Up a Git Server with SSH, Hooks, and Static Site Automation

Provides instructions for hosting a git server with SSH, configuring hooks for automatic updates, and enabling server-side static site generation, ensuring easy, backed-up code management.

  • Describes setting up a git server with SSH access for cloning, working locally, and pushing changes
  • Recommends configuring receive.denyCurrentBranch updateInstead to allow pushing to checked-out branches
  • Explains automating git update-server-info with post-update hooks for HTTP access and static site generation

Building a Minimal Linux RISC-V Distro with Kernel and QEMU Boot

Builds a basic Linux micro-distro on RISC-V, including kernel compilation, initramfs creation, and user-space tools, booted via QEMU, illustrating kernel, init, and network setup.

  • The article details building a minimal Linux distribution from scratch on RISC-V architecture, including kernel compilation, initramfs creation, and user-space setup.
  • Uses Linux kernel version 6.5.2, cross-compiled with riscv64-linux-gnu- toolchain, and QEMU for emulation.
  • Demonstrates booting Linux in QEMU, kernel configuration, loading initramfs, and running simple user-space programs like a custom init and shell, with network setup for internet access.

NVIDIA DGX Spark Benchmarks Show High Performance but Highlight FP16 Inference Issues

NVIDIA’s DGX Spark benchmarks are accurate but omit critical limitations like FP16 inference instability and memory fragmentation; effective production use requires workarounds, BF16 inference testing, and careful system management.

  • NVIDIA’s DGX Spark benchmarks report 82,739 tokens/sec for fine-tuning and high inference throughput, claiming 1 petaflop FP4 compute and <1% accuracy degradation with FP4
  • After 6+ days of intensive ML workloads on ARM64-based DGX Spark with Blackwell GB10 GPU and CUDA 13.0, real-world issues include FP16 inference numerical instability, memory fragmentation, and system hangs
  • Testing revealed training performance aligns with NVIDIA’s claims, inference speed scales proportionally with model size (~80 tokens/sec for 3.8B models), and 4-bit quantization maintains quality, but FP16 GPU inference is broken without BF16 testing

Ken Thompson Reflects on the Playful Origins of Unix at Bell Labs

Ken Thompson recounts the playful, community-driven origins of Unix at Bell Labs, emphasizing experimentation, open collaboration, and the cultural environment that fostered its development.

  • Ken Thompson, 82, recalls the early, rowdy days of Unix development at Bell Labs, highlighting community and experimentation.
  • Unix originated around 1966 from Thompson’s work on memory management after Multics project’s failure.
  • The Unix team, including Dennis Ritchie and others, built a collaborative environment with open source-like practices, including shared source and lock-picking antics.

AI Stock Surges Boost Employee Wealth and Retention Through RSUs

AI-driven stock surges have increased employee wealth via RSUs, creating “golden handcuffs” that incentivize retention but risk substantial forfeiture if employees leave early.

  • Employees at Nvidia and Broadcom hold RSUs worth nearly $2 million and $267,000 respectively, up from $420,000 and $66,000 in 2023 due to stock surges during the AI boom.
  • Stock payouts vest over time; leaving early results in forfeiting unvested RSUs, serving as a retention tactic to reduce employee turnover.
  • Nvidia’s turnover rate has halved, and Broadcom’s voluntary attrition is 6.2%, both citing equity awards as key retention incentives.

Australian Founder Moves to SF to Build AI Startup in Hacker House

Annie Liao relocated from Australia to San Francisco to build her AI education startup, living in a hacker house to foster community, while balancing the city’s inspiring tech scene with personal and safety challenges.

  • Annie Liao moved from Australia to San Francisco after raising $1.75 million in pre-seed funding for her startup, Build Club.
  • She lives in a hacker house called “The Stables,” which she describes as a supportive environment for founders, fostering collaboration and emotional support.
  • San Francisco’s tech culture and ambitious community inspire her, though she notes challenges like safety concerns, wealth disparity, and healthcare differences compared to Australia.

Teen Entrepreneurs Phoebe Gates and Sophia Kianni Raise 8M for AI Fashion App

Phoebe Gates and Sophia Kianni, both 23, raised $8 million for their AI fashion app Phia, which has over 600,000 users, using AI to optimize workflows and emphasizing team trust and prioritization.

  • Phoebe Gates and Sophia Kianni launched the Phia AI fashion app in April, amassing over 600,000 users and raising $8 million in seed funding led by Kleiner Perkins, with investors like Kris Jenner and Hailey Bieber.
  • The app compares prices across approximately 40,000 linked sites to help users find deals, leveraging AI to enhance workflow, including email writing, coding, and content creation.
  • Gates and Kianni, both 23, met at Stanford, developed the app independently without Gates’ family backing, and have learned key leadership lessons such as delegating tasks and focusing on strengths.

OpenAI Bypassed External Advisers in $1.5 Trillion Infrastructure Deals

OpenAI bypassed external advisers on $1.5tn deals, focusing on internal dealmakers and trust-based, multi-year infrastructure agreements with Nvidia, Oracle, and AMD, emphasizing technical development over detailed financial terms.

  • OpenAI largely avoided external advisers, relying on internal dealmakers for negotiations totaling approximately $1.5 trillion.
  • Sam Altman and team focused on technical infrastructure deals with Nvidia, Oracle, AMD, and Broadcom, emphasizing engineering over detailed financial terms.
  • Agreements were structured over multiple years with milestone-linked payments, allowing flexibility to scale back chip orders if needed.
  • Notable deals include an $11.9bn contract with CoreWeave, expanded to over $22bn, and a $300bn five-year partnership with Oracle.
  • OpenAI’s chip supply negotiations often originated from inbound approaches, with trust-based, open-ended agreements.
  • Neither Nvidia nor OpenAI sought external advice for the Nvidia investment deal, which involves up to $100bn investment for up to 10GW of chips.
  • AMD granted warrants to purchase up to 10% of OpenAI for $0.01 per share in exchange for 6GW of chips.
  • OpenAI’s internal team, including CFO Sarah Friar and President Greg Brockman, led deal structuring, with external advisers like Michael Klein involved in fundraising but not chip deals.
  • The company aims to rapidly increase computing capacity, targeting 1 gigawatt weekly, with infrastructure financing led by Peter Hoeschele.

Windows 10 End-of-Support Sparks Global PC Upgrade Boom and Mac Growth

The Windows 10 support deadline spurred a global PC upgrade cycle, boosting Mac shipments 14.9% YoY in Q3 2025, amid rising enterprise AI adoption and market consolidation.

  • The October 2025 end-of-support deadline for Windows 10 is driving a major PC replacement cycle, benefiting Apple with a 14.9% YoY increase in Mac shipments in Q3 2025.
  • Nearly 40% of the global installed PC base still runs Windows 10, prompting early fleet renewals across commercial and consumer sectors.
  • Major PC vendors: Lenovo grew 17.4%, Asus 14.1%, HP 10.3%, while Dell shipments declined 0.9%; top five vendors hold nearly 75% of the market.
  • PC manufacturers are integrating AI features and neural processing units, with enterprise buyers beginning to prioritize AI capabilities in procurement.

Microsoft Recommends Registry Hack to Fix Smart Card Issues After October 2025 Windows Update

Microsoft advises applying a registry hack (DisableCapiOverrideForRSA=0) to temporarily fix smart card authentication issues caused by the October 2025 Windows Update, which intentionally broke CSP-based RSA certificate support.

  • Microsoft recommends a registry hack to mitigate smart card authentication failures caused by the October 2025 Windows Update.
  • The update intentionally broke smart card certificate operations by requiring RSA certificates to use KSP instead of CSP; issues include sign-in failures and recognition problems.
  • The workaround involves setting the DisableCapiOverrideForRSA registry key to 0 on affected devices; this fix applies to nearly all supported Windows versions, including Windows 10 22H2, and is temporary until April 2026 when Microsoft plans to remove it.

US DOE Moves to Fast-Track Large Datacenter Grid Connections

US DOE proposes FERC rules to fast-track large datacenter grid hookups within 60 days, addressing multi-year delays caused by AI-driven power demand growth and regulatory hurdles.

  • US Energy Secretary directs FERC to implement new rules to expedite datacenter grid connections, limiting review time to 60 days.
  • Proposed regulations target large loads over 20 MW, requiring standardized study deposits, readiness, and withdrawal penalties.
  • The move aims to address delays, currently up to seven years, driven by increased datacenter construction and AI hardware power demands, with US power draw projected to be over 30 times higher by 2035.

Microsoft patches prompt injection flaw in Microsoft 365 Copilot enabling data theft

Microsoft patched an indirect prompt injection vulnerability in Microsoft 365 Copilot that enabled data exfiltration through Mermaid diagrams, involving malicious CSS/JavaScript to steal tenant emails.

  • Microsoft fixed a vulnerability in Microsoft 365 Copilot allowing indirect prompt injection that could exfiltrate tenant data, including emails.
  • The exploit used Mermaid diagrams’ support for CSS and JavaScript to generate fake login buttons that sent hex-encoded email data to attacker-controlled servers.
  • Researcher Adam Logue discovered the flaw, which involved embedding malicious instructions via Mermaid diagrams; Microsoft verified and patched the issue, but Logue did not receive a bug bounty as Copilot is out of scope.

OpenAI Launches Company Knowledge Feature for ChatGPT Enterprise Competitively Challenging Microsoft

OpenAI’s company knowledge adds organizational data access to ChatGPT, competing with Microsoft 365 Copilot, with connectors for Slack, SharePoint, Google Drive, Teams, and Outlook, but lacks OneDrive integration.

  • OpenAI introduces “company knowledge” feature for ChatGPT, available for ChatGPT Business, Enterprise, and Edu, integrating with Slack, SharePoint, Google Drive, Teams, and Outlook; OneDrive is absent.
  • Users authenticate each connector individually; data is encrypted, not used for training; administrators can review conversation logs via Enterprise Compliance API.
  • Data residency support varies by connector; OpenAI did not specify data processing locations; company knowledge requires manual selection per conversation and cannot search the web or generate images.
  • ChatGPT Business costs $25 per user/month (annual plan), compared to Microsoft 365 Copilot’s $30/month; OpenAI aims to position ChatGPT as an enterprise AI alternative challenging Microsoft’s deeper integrations.

Microsoft Celebrates 11 Years of Windows Insider Program with Special Wallpapers

Microsoft marks 11 years of the Windows Insider Program with special wallpapers; no signs of Windows 12, and a ZIP file hints at possible hidden messages.

  • Microsoft celebrates 11 years of the Windows Insider Program with custom wallpapers featuring the number “11” in dark and light modes
  • The program, launched in 2014, tests unreleased Windows versions across multiple channels: Release Preview, Beta, Dev, and Canary
  • No indication of Windows 12 development; a ZIP file containing Insider wallpapers includes a _MACOSX folder, possibly a leftover or secret message

UK MPs Urge Tech Firms to Fight Phone Theft with New Security Measures

UK MPs recommend the government enforce tech industry adoption of IMEI-based and cloud-linked device locks to reduce phone thefts, with 117,211 phones stolen in 2024 and minimal convictions.

  • UK House of Commons committee urges Home Secretary to compel Apple, Google, and Samsung to implement technical measures to render stolen phones worthless
  • 117,211 phones were stolen in 2024, a 25% increase from 2019; only 1% of London thefts result in charges or convictions
  • Apple has security features like Find My, Activation Lock, and Stolen Device Protection; considering IMEI blocking; Google and Samsung have implemented various anti-theft measures

ACCC Sues Microsoft Over Alleged Misleading Microsoft 365 Copilot Plans

Australia’s ACCC sued Microsoft for allegedly misleading Microsoft 365 users about Copilot plans, claiming concealment of lower-cost “Classic” plans that exclude Copilot at the previous price.

  • Australia’s ACCC filed legal proceedings against Microsoft for allegedly misleading users of Microsoft 365 Personal and Family plans regarding Copilot subscription options.
  • The ACCC claims Microsoft falsely advised users that maintaining their subscription required accepting Copilot and paying higher prices or canceling, omitting the existence of the “Classic” plans that retain features without Copilot at lower cost.
  • The ACCC alleges Microsoft concealed the “Classic” plans until after users initiated cancellation, violating consumer protection laws; Microsoft is reviewing the claim.

▶️ Technology

Building a 64-bit RISC-V Emulator in Rust for xv6 Compatibility

The ongoing project details building a RISC-V 64-bit emulator in Rust, enabling running xv6, by implementing hardware components and instruction sets including RV64I, M, and A.

  • The book guides creating a 64-bit RISC-V emulator in Rust, covering hardware components, instruction set, and system architecture.
  • Hardware components include CPU with two instructions, memory, system bus, CSRs, privileged architecture, exceptions, PLIC, CLINT, UART, interrupts, Virtio, and virtual memory.
  • Instruction sets implemented are RV64I base, “M” extension for multiplication/division, and “A” extension for atomic instructions; source code available at d0iasm/rvemu-for-book.

TinyKVM boosts performance with unmodified runtime support and innovative isolation

TinyKVM, open-sourced in February 2025, now supports running unmodified runtime executables with minimalist syscall emulation, hybrid per-request isolation, custom RPC, and VM snapshots, approaching native performance.

  • TinyKVM was open-sourced in February 2025; recent updates include limited support for running unmodified executables like Deno, Python WSGI, and Lune
  • Implements minimalist system call emulation with about 50 real syscalls, focusing on avoiding Linux kernel access and providing sanitized capabilities
  • Features per-request hybrid isolation modes using VM page rewriting and full VM resets, achieving near-native performance in benchmarks despite resetting entire VMs each request
  • Introduces a custom remote procedure call (RPC) mechanism allowing direct function calls between VMs via trapping and register switching, enabling zero-copy deep object passing
  • Supports VM snapshots by backing memory with a single file, enabling fast startup (0.7ms) and efficient state transport; plans include selective page preloading for faster cold starts

Microsoft Launches Agent Lightning for Zero-Code AI Agent Training

Microsoft’s Agent Lightning enables training and optimizing AI agents with minimal code, supporting multiple frameworks and algorithms, and streamlining the development pipeline.

  • GitHub repository microsoft/agent-lightning hosts an open-source framework for training AI agents.
  • Latest release is v0.2.0, published on October 22, 2025, with 2.3k stars and 185 forks.
  • Core features include zero-code agent optimization, compatibility with various frameworks (LangChain, OpenAI SDK, AutoGen, CrewAI), and support for reinforcement learning, prompt optimization, and supervised fine-tuning.

Linux Boot Sequence: From CPU Reset to Kernel Initialization and Security Features

The Linux boot process involves CPU real mode startup, firmware (BIOS/UEFI) initialization, bootloader loading, setup execution, mode switching to 64-bit long mode with paging, kernel decompression, relocation, and optional address randomization for security.

  • The Linux boot process begins with CPU reset to real mode at address 0xFFFFFFF0, executing a jump to firmware.
  • BIOS performs POST, checks boot devices for 512-byte boot sector ending with 0x55AA, and loads it at 0x7C00; UEFI replaces BIOS with direct filesystem access.
  • The bootloader (e.g., GRUB) loads the Linux kernel, which contains a setup program and compressed kernel; setup creates a predictable workspace, clears BSS, configures segment registers, and queries RAM via e820.
  • The setup code calls main in real mode, then switches to protected mode by disabling interrupts, opening A20 line, loading GDT/IDT, and setting CR0.PE, transitioning to 32-bit mode.
  • To run in 64-bit long mode, the kernel enables paging with identity mappings, sets LME in EFER, and performs a far return into 64-bit code, establishing virtual address translation.
  • The decompression stage unpacks the kernel ELF, applies relocations if loaded at a non-default address, and jumps to start_kernel.
  • Kernel Address Space Layout Randomization (kASLR) randomly chooses physical and virtual base addresses from free memory regions, unless disabled with nokaslr, enhancing security.

Apple Targets Vapor Chamber Cooling and Foldable iPad Innovations by 2027

Apple plans to introduce a vapor chamber cooling system in the iPad Pro by 2027 to improve thermal performance, while advancing plans for Apple Maps ads and a foldable 18-inch iPad, which faces development challenges.

  • Apple is developing a vapor chamber cooling system for the iPad Pro, expected as early as 2027
  • The vapor chamber aims to enhance thermal management and performance
  • Apple is also progressing on integrating ads into Apple Maps and developing a foldable iPad with an 18-inch screen, which has encountered development issues

AI invasiveness and ethical concerns overshadow consumer tech quality

AI features are increasingly invasive and degrade product quality, driven by marketing motives; models are trained on uncredited data, raising ethical concerns, while internet content becomes less authentic.

  • The article critiques the proliferation of unwanted AI features across consumer tech, emphasizing their tendency to degrade product quality and user experience.
  • Examples include Google’s Gemini replacing Assistant, which suffers from increased latency and reduced functionality; Apple’s Siri worsening under Apple Intelligence; and Microsoft’s Copilot cluttering Windows with intrusive AI elements.
  • The piece highlights that AI-driven updates often prioritize marketing over utility, replacing well-designed products with trend-chasing gimmicks, and notes that AI models are trained on data scraped without consent, often infringing on creators’ rights.

BBC Study Finds Nearly Half of AI Chatbot News Responses Are Misinformed

A BBC-led study revealed that OpenAI’s ChatGPT, Microsoft Copilot, Google Gemini, and Perplexity misrepresented news in 45% of responses, with Gemini worst at 76% error rate, highlighting AI hallucination and sourcing issues.

  • BBC investigation found that AI chatbots misrepresented news content in nearly 50% of responses
  • Google Gemini had a 76% error rate, the highest among four tested AI models
  • 45% of responses contained significant issues; 31% had sourcing problems; 20% had major accuracy errors, including hallucinations and outdated info

Co-packaged Optics Boost AI Data Center Scalability and Manufacturing

CPO technology aims to enable scalable, high-volume manufacturing of optical interconnects for AI datacenters, addressing the complexity of scaling from tens of thousands to millions of XPUs.

  • Co-packaged optics (CPO) are emerging as a solution for high-bandwidth, low-latency interconnects in datacenter AI systems, promising manufacturing at high volumes and low costs.
  • CPO will initially be adopted in scale-out networks, enabling increased link density without raising bandwidth, crucial for interconnecting thousands to millions of XPUs.
  • Industry experts will discuss scaling challenges and solutions in a webinar on November 5, featuring leaders from Alchip, Astera Labs, and Ayar Labs, focusing on manufacturing and integration hurdles.

Sora 2 AI Deepfakes Elicit Racial Slurs, Raising Misinformation and Privacy Concerns

Sora 2 AI can generate deepfakes of celebrities uttering racial slurs by bypassing filters through homophones, raising concerns over misinformation, reputation damage, and misuse of likenesses.

  • Researchers uncovered that Sora 2 AI video generator can produce deepfakes of celebrities, including Sam Altman and Mark Cuban, making them appear to utter racial slurs despite filters.
  • Users bypass built-in hate speech filters by prompting with homophones or coded terms like “knitters,” resulting in offensive content that mimics racist language.
  • Copyleaks reported that such videos, including recreations of a 2020 incident with celebrities, have been reposted on platforms like TikTok, with some videos garnering over 168,000 likes.
  • Sora’s Cameo feature allows uploading clips of users to insert into generated videos; many celebrities’ likenesses are used without permission, including deceased figures.
  • The platform’s evasion techniques highlight risks of deepfake misuse, including distortion of reality, damage to reputations, and intellectual property violations.