Today’s news highlights include the rising influence of AI in sectors like finance, with fintech firms adopting event sourcing for compliance; computing firms outperforming Bitcoin by shifting to high-performance AI models; and Broadcom benefiting from surging AI chip demand amid layoffs. Cyber threats also intensified, with Chinese APT groups targeting Russian supply chains, while UK energy issues threaten AI data center expansion.
▶️ Internet Infrastructure
Fintech Uses Event Sourcing and Tiered Storage to Boost Compliance and Efficiency
The fintech project adopted event sourcing in the Transaction-Portfolio Service for auditability, using snapshots and tiered storage to optimize performance and manage storage costs, enabling compliance and scalable data reconstruction.
- Implemented event sourcing in the Transaction-Portfolio Service to ensure auditability and compliance, storing events with fields: Event_ID, Account_ID, Event_Type, Amount, Timestamp
- Used snapshots and delta replay to optimize event replay performance, reducing balance reconstruction time from 2–5 seconds to 50–200ms for active accounts
- Employed tiered storage: hot (Azure Premium SSD, last 3 months, ~2TB), warm (Azure Standard SSD, 3–12 months, ~5TB), cold (Azure Archive, over 1 year, ~50TB), with storage costs of $800/month versus $15,000/month on premium
Computing Firms Pivot to AI and High-Performance Models Outperform Bitcoin
Crypto miners shifting to AI and high-performance computing models are outperforming Bitcoin, reversing previous declines, as sector profitability and competition dynamics evolve since the 2023 AI boom.
- Shares of large-scale computing companies pivoting to AI and high-performance computing are outperforming Bitcoin, driven by a shift in mining models.
- The sector experienced a boom two years ago during the AI surge but saw share prices decline in the following year due to decreased profitability and increased competition.
- Companies are increasingly adopting hybrid models that integrate AI and high-performance computing, moving away from traditional Bitcoin mining approaches.
Broadcom Cuts Staff Amid Post-Acquisition Layoffs as AI Demand Surges
Broadcom cut staff in sales and account management roles amid ongoing layoffs post-VMware acquisition, benefiting from AI chip demand, with a $1.65 trillion market cap and a $10 billion AI accelerator deal with OpenAI.
- Broadcom cut staff primarily in sales, customer success, account management, and solutions roles this week, with exact numbers unspecified
- The layoffs follow Broadcom’s 2023 acquisition of VMware, which resulted in workforce reductions of approximately 50%
- Broadcom benefits from the AI boom by designing chips for AI technology, reaching a $1.65 trillion market cap late last year and signing a $10 billion AI accelerator agreement with OpenAI on October 16, 2025
Chinese APT Jewelbug Targets Russian IT Firms in Long-Range Supply Chain Infiltration
Symantec reports Chinese APT group “Jewelbug” infiltrated a Russian IT provider from early 2025 to May 2025, using stealth techniques and cloud exfiltration to enable supply chain attacks.
- Symantec Threat Hunter Team identified a Chinese APT group “Jewelbug” (REF7707, CL-STA-0049, Earth Alux) infiltrating a Russian IT services firm from early 2025 to May 2025
- The intrusion lasted several months, allowing access to build servers, code repositories, and infrastructure, enabling potential supply chain attacks on Russian clients
- Attackers used renamed Microsoft cdb.exe (“7zup.exe”) for stealth, employing credential dumps, scheduled-task persistence, log clearing, and exfiltration via Yandex Cloud
- The operation indicates Chinese actors are now probing Russian infrastructure, marking a rare east-on-east cyber espionage activity
- The attack posed risks of espionage or disruption across Russian companies, highlighting evolving cyber threat dynamics between China and Russia
Datacenter Shift to 21-Inch Racks Expected by 2030
Most datacenters will standardize on 21-inch racks by 2030, offering improved airflow, larger fan arrays, and better support for AI infrastructure, with over 70% of kit shipped adopting this format.
- Most datacenters are expected to adopt 21-inch racks based on OCP standards by 2030, surpassing the traditional 19-inch standard.
- The larger rack format, popularized by the Open Compute Project, will constitute over 70% of shipped equipment by 2030, driven by vendors like Dell and HPE.
- Major cloud providers including Microsoft, Amazon, Meta, Google, ByteDance, Huawei, and Oracle have already transitioned to 21-inch racks, citing advantages in airflow, thermal management, and future scalability.
Britain’s Energy Crisis Threatens AI Data Center Expansion
Britain faces electricity supply challenges for AI datacenter growth, relying on delayed nuclear projects, expensive gas, and limited renewables, risking blackouts and higher consumer bills.
- UK Energy Secretary Ed Miliband advocates for renewables and nuclear power to support AI datacenter expansion amid infrastructure underinvestment
- Britain has the world’s most expensive electricity, heavily reliant on gas-fired backup due to limited renewable capacity
- Construction delays and high costs affect nuclear projects like Hinkley Point C, with operational dates pushed to 2030+ and costs rising from £26B to £34B
Meta Secures $27B Debt for Hyperion AI Data Center in Louisiana
Meta secured $27 billion in debt and $1.5 billion equity financing from Blue Owl Capital for its Hyperion AI datacenter, structured to keep debt off Meta’s books, with operations starting in 2029.
- Meta persuaded Blue Owl Capital to finance its Hyperion AI datacenter project in Louisiana with approximately $27 billion in debt and $1.5 billion in equity, structured to keep debt off Meta’s balance sheet
- The deal, brokered by Morgan Stanley, involves a 20% stake for Meta, with debt maturing in 2049, fully amortizing, and the project expected to be operational by 2029
- The Hyperion datacenter, initially announced in December 2024 with a $10 billion estimate, has expanded to over five gigawatts of compute capacity, requiring significant power infrastructure including a new 2.2 GW natural gas generator plant by Entergy
Microsoft Hosts NASA Earth Observation Data Amid Funding Uncertainty
Microsoft hosts NASA’s HLS satellite data on Azure’s Planetary Computer, enabling frequent Earth observation analysis; funding depends on US government budget, with AI integration suggested.
- Microsoft hosts NASA’s Harmonized Landsat and Sentinel-2 (HLS) dataset on Azure via the Planetary Computer platform
- HLS combines data from Landsat 8/9 (30m resolution, 16-day revisit) and Sentinel-2A-D (10-20m resolution, 5-day revisit), enabling observations every 2-3 days at 30m
- Dataset supports climate change, land use, and agricultural research; NASA funding is uncertain due to US government shutdown
HPC Brings 200-Million-Year-Old Dinosaur Footprints to Life
HPC enables detailed simulations of 200-million-year-old dinosaur footprints, advancing paleontology and inspiring cross-disciplinary collaborations in robotics, medicine, and art.
- Researchers at Liverpool John Moores University combine paleontology, biomechanics, and HPC to simulate dinosaur footprints at the sand grain level.
- The simulations utilize HPC systems like LJMU Prospero, launched in August 2020, handling millions of particles and days-long computations.
- The work involves finite element analysis, with Falkingham and Griffin emphasizing the importance of mentorship and scalable HPC for detailed, realistic modeling.
▶️ Open Source
Forgejo v13.0 Launches Content Reporting and Security Enhancements
Forgejo v13.0, released on 16 October 2025, introduces content reporting, security improvements, migration tools from Pagure, configurable global 2FA, avatar EXIF removal, and enhanced Actions features.
- Forgejo v13.0 was released on 16 October 2025, with a complete list of changes in the release notes
- Introduces user and admin reporting of abusive content, with reports accessible via the admin interface, and supports migration from Pagure repositories
- Adds security enhancements including a more secure module for Forgejo Actions secrets, optional global 2FA enforcement, and removal of EXIF data from avatar uploads
ripgrep 15.0.0 Boosts Performance with Bug Fixes and Platform Updates
ripgrep 15.0.0 introduces bug fixes, performance enhancements, and new features, including improved gitignore handling, platform support updates, and binary optimizations, released on October 16, 2023.
- ripgrep 15.0.0 is a major release with bug fixes, performance improvements, and new features.
- Fixed gitignore matching bugs, memory regressions with large gitignore files, and issues with
rg -vf filewhenfileis empty. - Platform support updates include Windows
aarch64artifacts;powerpc64artifacts removed; binaries now compiled with full LTO for performance gains.
Free-Programming-Books: Extensive Multilingual Resources Since 2014
The free-programming-books repository offers extensive, multilingual free programming resources, including books, courses, and tutorials, maintained collaboratively since 2014.
- The repository free-programming-books hosts a curated list of freely available programming books and resources in multiple languages.
- Contains over 9,830 commits, 373,000 stars, and 64,900 forks, maintained by a community of 3,090 contributors.
- Includes resources grouped by programming language, subject, and resource type, with links to books, courses, cheat sheets, podcasts, interactive tutorials, problem sets, and more.
Arduino Library Connects Behringer WING and Midas DP48 via A²B Audio Technology
The StageConnect Arduino library facilitates A²B-based audio connections between devices like Behringer WING and Midas DP48, supporting 32 channels at 48kHz over 15-meter cables.
- Repository hosts an Arduino library for StageConnect devices using Analog Devices AD242x-ICs, based on automotive A²B technology.
- Enables connection of Behringer WING to Midas DP48 via uncompressed 48kHz 32-bit audio over XLR cables up to 15 meters.
- Uses virtual I2C communication routed through AD242x chips, with setup instructions including code snippets and hardware connection details.
WebMCP Enables Client-Side LLM Support and Resource Sharing on Websites
WebMCP is an open-source framework that enables websites to support client-side large language models through a websocket-based MCP protocol, facilitating tool sharing and resource management without API keys.
- WebMCP is a project enabling websites to support client-side LLMs via a WebSocket-based MCP protocol.
- It allows websites to share tools, resources, and prompts without API key sharing, functioning as an MCP server.
- The system includes a widget for websites, a websocket server for communication, and tools like token generators and schema definers, with Docker support for deployment.
Lux: A Lua Package Manager for Dependencies, Formatting, and More
Lux is a Lua package manager providing dependency management, lockfile support, code formatting, type checking, and compatibility with luarocks, aiming to streamline Lua project workflows.
- Lux is a Lua package manager designed to manage dependencies, build steps, and project configuration via
lux.toml. - Features include parallel builds, automatic rockspec generation, code formatting with
stylua, type checking withemmylua-analyzer-rust, and lockfile support. - Lux offers a comparison with luarocks, highlighting features like dependency management, lockfile support, code linting, formatting, and native Lua detection, with some features still in progress.
Flowistry: Rust IDE Plugin Visualizing Code Flow with Ownership-Based Analysis
Flowistry is a Rust IDE plugin that visualizes information flow, enabling focus on relevant code by highlighting influence regions, with analysis based on MIR and ownership models.
- Flowistry is an IDE plugin for Rust that analyzes information flow within Rust programs.
- It provides a “focus mode” to highlight code influencing or influenced by a selected code region.
- Based on the paper “Modular Information Flow through Ownership” at PLDI 2022.
AI-Generated Static Website with Security Leak and Broken Links
A static website built entirely by AI (Claude), featuring broken links, a security leak exposing “hunter2” password, and a messy project structure, exemplifying rapid, untested AI-driven web development.
- The website is entirely AI-generated using Claude responses, with minimal human oversight.
- It features a static HTML structure with intentionally broken or placeholder links, and a security leak exposing the password “hunter2.”
- The project directory includes multiple HTML files, scripts, CSS folders, and documentation, all stored in a GitHub repository here.
Skyvern Boosts Web Automation Efficiency with AI-Generated Playwright Scripts
Skyvern’s AI-driven automation platform uses reasoning models to generate maintainable Playwright scripts, achieving 2.7x cost reduction and 2.3x faster runs by separating exploration and replay phases.
- Skyvern enables AI to generate and maintain its own code for web automation, reducing costs by 2.7x and execution time by 2.3x
- Utilizes reasoning models to improve agent accuracy and produce maintainable Playwright scripts
- Implements a two-phase approach: explore mode for learning website flow and replay mode for deterministic execution, leveraging intent metadata for recovery
RubyGems and Bundler Repos Moving to Ruby Core Amid Community Controversy
Ruby Central transferred RubyGems and Bundler repositories to the Ruby core team for stability, following a contentious takeover involving revocation of maintainers’ access and community disputes.
- Ruby Central transferred repository ownership of RubyGems and Bundler to the Ruby core team to improve stability, without restoring control to original maintainers
- The move followed a power struggle initiated when Ruby Central’s GitHub organization was renamed and admin rights revoked by Ruby core members
- Ruby creator Yukihiro Matsumoto announced the core team will oversee these projects, maintaining licenses and contributor rights, amid community controversy and allegations of a hostile takeover
▶️ Software Development
Microsoft Patches Critical ASP.NET Core Request Smuggling Flaw
Microsoft fixed a CVSS 9.9 ASP.NET Core vulnerability in Kestrel enabling request smuggling, affecting all versions, with patch options including SDK updates or NuGet packages.
- Microsoft patched a high-severity (CVSS 9.9) vulnerability in ASP.NET Core affecting all supported versions, including 2.3, 8, 9, and 10 pre-release
- The flaw resides in the Kestrel web server and enables request smuggling, allowing hidden requests to bypass security checks
- Exploitation could lead to actions such as user impersonation, CSRF bypass, or injection attacks, depending on application code; patch via latest SDK or NuGet package (2.3.6)
▶️ Management and Leadership
Ukrainian Founder Turns Headway into $720M Edtech Giant Amid Wartime Growth
Ukrainian founder Anton Pavlovsky built Headway into a 160-million-user edtech platform valued at $720 million, expanding globally and maintaining profitability amid wartime challenges.
- Anton Pavlovsky, Ukrainian entrepreneur, founded Headway Inc in 2019, reaching 160 million users and generating an estimated $160 million revenue, valued at $720 million
- Headway’s apps include 15-minute book summaries and the brain game app Impulse; revenue streams include subscriptions, paywalled features, advertising, and affiliate sales
- During the 2022 war, Pavlovsky led the evacuation of employees from Kyiv, maintaining profitability and launching new products like Nibble, AddMile, and Skillsta; company growth driven by AI-enhanced marketing and language expansion
Metropolis 1998: Pixel-Art City Builder with Large-Scale Management and Future Modes
Metropolis 1998 is an early-stage city builder allowing detailed isometric pixel-art city design, supporting large populations, with planned features like crime and zombie modes, aiming for release between Q4 2024 and Q2 2025.
- Metropolis 1998 is an isometric, pixel-art city-building game inspired by SimCity, Dwarf Fortress, and Rollercoaster Tycoon, available in pre-alpha demo on Steam and Itch.io.
- Players can design detailed buildings, including room layouts, doors, windows, and interior objects, with options to save and reuse blueprints; early development version features a complex interface requiring extensive interaction.
- The game supports large populations with an algorithm capable of tracking hundreds of thousands of objects; planned features include crime scenarios, zombie mode, and city management mechanics, with a roadmap and ongoing updates from developer YesBox.
Zorin OS 18 Launches with Redesigned Interface and Enhanced Windows Support
Zorin OS 18, launched on October 14, 2025, offers a redesigned interface, improved Windows compatibility, and new productivity features, supporting over 170 Windows apps and long-term updates until 2029.
- Zorin OS 18 was released on October 14, 2025, featuring a redesigned desktop, new themes, and enhanced features.
- Supports seamless Windows 10 replacement, with improved Windows app support powered by WINE 10, and long-term support until June 2029.
- Introduces advanced window tiling, Web Apps integration, OneDrive support, new desktop layouts, and expanded app collections in Pro and Education editions.
Early 90s Text-Based IDEs Outshine Modern Console Editors in Features
Early 1990s IDEs like Turbo C++ provided full-featured, integrated development environments in under 9 MB, whereas modern terminal editors lack such comprehensive, user-friendly features, indicating regression in usability.
- The article reviews text-based IDEs from the late 1980s and early 1990s, highlighting features like TUIs, syntax highlighting, integrated debugging, and project management.
- Borland Turbo C++ (1990) offered full-screen TUIs with syntax highlighting, compiler diagnostics, project management, debugging, and integrated help, all under 9 MB.
- Modern console IDEs like Neovim, Doom Emacs, and Helix lack the integrated, user-friendly experience of early IDEs; current tools are multi-language and less specialized.
WhatsApp to Ban General-Purpose AI Chatbots in Business API from 2026
WhatsApp will ban general-purpose AI chatbots via its Business API from January 15, 2026, restricting AI model providers and emphasizing customer support use cases to manage system load.
- WhatsApp changed its business API policy to ban general-purpose chatbots, effective January 15, 2026
- The new terms prohibit AI providers, including large language models and generative AI platforms, from distributing AI assistants on WhatsApp
- Meta states the move aims to focus on customer support use cases and reduce system burden caused by unanticipated chatbot volume
Robinhood Shifts Focus to Gen Z and Retirement Investing for Long-Term Growth
Robinhood is diversifying beyond meme stocks by targeting Gen Z, 401(k)s, and political trends, aiming for sustained growth through long-term investing and demographic expansion.
- Robinhood aims to expand its user base by targeting Gen Z, 401(k) investors, and leveraging political trends, including Trump-era policies
- CEO Vlad Tenev emphasizes staying relevant to younger generations and listening to Gen Z to maintain growth
- The platform is shifting focus from meme stocks to long-term investing, including retirement accounts and new demographic segments
BCG Urges 5 Hours of AI Training to Unlock Value and Drive Innovation
BCG’s Steven Mills advocates for five hours of AI training per employee and urges companies to reimagine processes, as only 5% currently extract value, with public sector adoption accelerating.
- Employees desire approximately five hours of hands-on AI training; only about one-third currently receive it, according to BCG research.
- Mills emphasizes that the real value of AI begins once employees experience its capabilities, creating a virtuous cycle of increased usage and innovation.
- BCG’s recent report indicates only 5% of companies are deriving measurable value from AI; organizations are urged to reimagine business processes and services leveraging AI’s transformational potential.
Military Leaders Accelerate AI Chatbot Use for Faster Decision-Making Amid Security Risks
Top military leaders, including Maj. Gen. William ‘Hank’ Taylor, are adopting AI chatbots for decision support, reflecting a broader military push toward AI integration in combat and logistics, emphasizing faster, data-driven decisions amid security concerns.
- US Army Maj. Gen. William ‘Hank’ Taylor is experimenting with generative AI chatbots to enhance decision-making in command and daily operations.
- The military is integrating AI into weapons systems, aircraft, combat tech, and back-end logistics, emphasizing faster decision processes aligned with the OODA Loop.
- The Pentagon warns of risks such as data leaks and flawed outputs from generative AI, urging caution in high-stakes military applications.
Marc Benioff Apologizes for Suggesting National Guard Deployment in SF
Marc Benioff apologized for recommending Trump send National Guard troops to San Francisco, citing safety concerns during Dreamforce 2025, amid criticism and the resignation of Ron Conway.
- Marc Benioff apologized for suggesting President Trump should deploy National Guard troops to San Francisco.
- His remarks were made “from an abundance of caution” during the Dreamforce conference, which had over 50,000 attendees.
- Salesforce declined to comment further on safety incidents; Ron Conway resigned from the Salesforce Foundation board citing value misalignment.
Nothing Aims at Gen Z with Unique Design, Valued at 1.3 Billion
Nothing, led by Carl Pei, targets younger consumers to penetrate the smartphone market, focusing on unique designs and form factors, with a valuation of $1.3 billion and sales forecasted at $1 billion in 2025.
- Nothing CEO Carl Pei targets Gen Z consumers, with an average user age of 26, to differentiate from Apple and Samsung.
- The company is valued at $1.3 billion, with sales projected to reach $1 billion in 2025.
- Nothing’s smartphones, including Phone 2 and Phone 3, range from <$300 to $700, featuring a design with a transparent back exposing internal hardware.
Salesforce CEO Apologizes for Support of Federal Troops in San Francisco
Salesforce CEO Marc Benioff apologized for supporting Trump’s possible deployment of federal troops to San Francisco, clarifying that the National Guard is unnecessary following local feedback and Dreamforce’s success.
- Salesforce CEO Marc Benioff apologized for supporting President Trump’s potential deployment of federal troops to San Francisco.
- Benioff stated he no longer believes the National Guard is needed after listening to local officials and the success of Dreamforce.
- Benioff previously indicated he would welcome troops to San Francisco, citing insufficient police presence, but faced backlash from local politicians and figures like Ron Conway and Elon Musk.
How to Classify Postgres Workloads as Read-Heavy or Write-Heavy
Postgres workload classification into read-heavy or write-heavy informs optimization, with read costs lower due to memory caching and write costs involving WAL, index, and disk I/O; a custom query aids workload assessment.
- The article analyzes whether Postgres workloads are read-heavy or write-heavy, emphasizing workload characteristics influence tuning strategies.
- Reads in Postgres are generally less costly, involving 8KB blocks often served from memory, while writes require WAL logging, index updates, and disk I/O.
- A custom SQL query estimates read and write activity by analyzing pg_stat_user_tables, pg_statio_user_tables, and pg_class, classifying tables as read-heavy, write-heavy, or balanced based on a configurable ratio (default 5:1).
Jeep Wrangler 4xe OTA Update Causes Powertrain Failures and Safety Concerns
Jeep’s October 2025 OTA update for Wrangler 4xe caused powertrain failures; a subsequent patch reverts to previous software, but many owners still face vehicle inoperability.
- Over-the-air software update for Jeep Wrangler 4xe models caused major powertrain glitches, rendering many vehicles inoperable and causing safety concerns.
- The faulty update scrambled vehicle electronics, leading to issues such as inability to start or loss of drive power, with some owners experiencing multiple shutdowns mid-drive.
- Jeep issued a second update to revert the software to the previous version; the update does not alter the infotainment system, making version identification difficult, and many owners remain without their vehicles.
Senator Cassidy Questions Cisco Over Critical Firewall Vulnerabilities and Breaches
Senator Cassidy demands Cisco disclose its knowledge of and response to critical firewall flaws CVE-2025-20333 and CVE-2025-20362, exploited since May 2025, which compromised federal and possibly other systems, prompting urgent government patching and raising concerns over vendor transparency.
- US Senator Bill Cassidy sent a letter to Cisco questioning its awareness and response to critical firewall vulnerabilities CVE-2025-20333 and CVE-2025-20362.
- The vulnerabilities led to an emergency patching directive from CISA, with Cisco admitting exploitation as early as May 2025, linked to the ArcaneDoor campaign and Chinese-linked threat group “UAT4356.”
- The flaws allowed attackers to drop implants, run commands, and siphon data from Cisco ASA 5500-X firewalls, with breaches affecting at least one federal agency; Cisco has not publicly confirmed or denied specific breaches.
Majority of SAP Users Struggle with S/4HANA Migration Amid Support Changes
Over 95% of SAP legacy users find building a business case for S/4HANA migration difficult; SAP’s support deadlines, licensing changes, and cloud focus influence strategic decisions.
- 95% of legacy SAP users report that building a business case for S/4HANA migration is challenging or requires significant effort
- Mainstream support for SAP ECC ends in 2027; extended support ends in 2030, with optional extensions until 2033 for early adopters
- 94% of respondents see value in extending existing SAP systems; SAP’s support policy changes and product renaming increase user uncertainty
- 61% of SAP ECC users have not licensed S/4HANA, despite being available since 2015; SAP now favors cloud SaaS deployments with AI innovations exclusive to these platforms
- SAP’s recent licensing changes, including the end of RISE with SAP and new Full-Use-Equivalent (FUE) metrics, have raised transparency concerns among users
- Customers are increasingly adopting open “composable” architectures, integrating third-party solutions to meet ERP needs
- SAP emphasizes cloud ERP Private Edition as a strategic move, recommending users rebuild or adapt business processes during migration
Vatican Advocates for Global AI Regulation to Promote Justice and Sustainability
Vatican’s seminar on October 17, 2025, urged global AI regulation, emphasizing AI as a tool for social justice, environmental sustainability, and fair benefit distribution, with calls for enforceable international frameworks.
- Vatican hosted a seminar titled Digital Rerum Novarum on October 17, 2025, advocating for global AI regulation and equitable distribution of AI benefits.
- Pope Leo XIV emphasized AI as a tool with potential to foster a just, human-centric society, highlighting the need for a shared, enforceable international regulatory framework.
- Participants called for binding global AI regulations supported by appropriate institutions, addressing environmental sustainability, energy reduction in datacenters, and social justice in AI’s future of work.
PC Shipments Surge as Windows 11 Boosts Growth and AI PCs Rise to 31%
Windows 11’s release has stimulated PC upgrades, boosting Q3 shipments by 8.2% to 69 million units, with AI PCs rising to 31% of shipments in 2025.
- Global PC shipments reached 69 million units in Q3 2025, up 8.2% YoY, driven mainly by Windows 11 end-of-support refresh cycle
- North American growth was limited to 1.6% due to earlier demand pull-in from Q1 and Q2 caused by import tariff concerns
- Industry forecasts indicate AI PCs will constitute 31% of shipments in 2025, up from 15% in 2024, driven by industry adoption rather than consumer demand
▶️ Technology
GPU Algorithms for Fast Distance Computation to Cubic Bézier Curves
Develops GPU-based algorithms for fast distance calculation to cubic Bézier curves using polynomial root solvers, including optimized GLSL implementations for degrees 2 to 5, enabling efficient rendering and shape analysis.
- Presents GPU-optimized methods for computing distances to cubic Bézier curves, crucial for text and shape rendering.
- Derives polynomial equations up to degree 5 for root solving, including quadratic, cubic, and quintic solutions.
- Implements GLSL code for root finding using Aberth–Ehrlich, Newton bisection, and ITP methods, with detailed polynomial differentiation and root filtering.
AMD Unveils Strix Halo: The Largest High-Performance Mobile APU to Date
AMD’s Strix Halo is a large, high-performance mobile APU with dual Zen 5 CCDs, RDNA 3.5 iGPU, 256GB/s memory bandwidth, and CPU performance near desktop flagship levels.
- AMD’s Strix Halo is the largest client APU to date, announced at CES 2025, with TDP from 55W to 120W, integrating CPU and GPU without dedicated graphics.
- Equipped with dual 8-core Zen 5 CCDs (16 cores), 512b FPUs, and a 5.1GHz max boost clock; features a RDNA 3.5 iGPU with 40 compute units, 32MB Infinity Cache, and up to 2.9GHz boost clock.
- Uses a 256b LPDDR5X-8000 memory bus, providing up to 256GB/s shared bandwidth; memory latency is approximately 123ns, with single CCD bandwidth around 103GB/s, and CPU performance comparable to desktop flagship CPUs.
Open-Source Quartz Watch Merges Hardware Firmware and Design
The ./watch project embraces electronic components in its quartz watch design, emphasizing hardware, firmware, and case development, with comprehensive documentation and open-source access.
- The watch design exposes electronic components, inspired by command-line interfaces and computer hardware aesthetics
- Features include PCB design, firmware development, and hardware case prototypes, with detailed decisions and rationales documented
- Source code and project details are available on GitHub
Recent Findings Confirm LLMs Can’t Achieve AGI Anytime Soon
Recent developments disprove LLMs as near-AGI solutions, with key papers and expert opinions confirming they cannot overcome core reasoning and distribution challenges, delaying AGI by at least a decade.
- Recent months have discredited the idea that LLMs can achieve AGI, citing failures in reasoning, distribution shift, and overhyped claims.
- June 2025: Apple’s reasoning paper confirmed LLMs cannot handle distribution shift; August 2025: GPT-5 underperformed; September 2025: Rich Sutton acknowledged limitations of LLMs; October 2025: Andrej Karpathy and Demis Hassabis emphasized AGI is at least a decade away.
- Experts agree current paradigm of LLMs is insufficient for AGI, with research indicating formal reasoning and understanding remain significant hurdles.
Oracle Expands AI Integration Across Cloud and Legacy Systems
Oracle is heavily investing in AI across its platform, enabling legacy system integration, expanding AI agent support, and partnering with hyperscalers, but user adoption remains uncertain due to data quality and governance challenges.
- Oracle announced extensive AI integrations across cloud infrastructure, applications, and data analytics at its AI World conference in Las Vegas.
- AI additions include enhancements to Fusion, Oracle’s cloud ERP, finance, HR, and payroll, with support for third-party LLM providers via AI Agent Studio and an agent marketplace.
- Oracle supports legacy systems like E-Business Suite and PeopleSoft with AI capabilities, offering extended support until 2036, longer than competitors like SAP.
Carnegie Mellon’s Unobtrusive AI Lets Objects Predict and Perform Tasks
Carnegie Mellon University’s “unobtrusive physical AI” uses vision and LLMs to enable objects like staplers and shelves to predict and perform tasks autonomously, enhancing trust in physical AI.
- Carnegie Mellon University researchers developed a computer vision system enabling everyday objects to predict user actions and reposition proactively.
- The system combines ceiling-mounted cameras, computer vision, and large language models (LLMs) to generate scene descriptions and control robotic bases.
- Prototype demonstrations include a self-moving hospital tray and shelves that fold out when groceries are detected; presented at ACM UIST 2025.
Boris Johnson Praises ChatGPT Despite Its Inaccuracies
Boris Johnson publicly declared his affection for ChatGPT, citing its usefulness for writing and questioning, while acknowledging its tendency for inaccuracies and fabrications.
- Boris Johnson publicly expressed love for ChatGPT in October 2025, describing it as “fantastic” and “strange.”
- Johnson admitted to using ChatGPT for writing books and asking questions, and adopted a affected accent during a TV interview.
- He acknowledged that ChatGPT sometimes provides misleading or incomplete information, reflecting its propensity for hallucinations and inaccuracies.