Alvaro Lopez Ortega / 2025-09-06 Briefing

Created Sat, 06 Sep 2025 02:09:28 +0000 Modified Sat, 13 Sep 2025 02:06:52 +0000
6942 Words

Today’s tech news highlights include Europe’s Jupiter supercomputer reaching exascale performance with Nvidia GPUs, Amazon’s Kuiper satellites achieving speeds over 1.28 Gbps, and the EU fining Google €3.5 billion for antitrust violations. Additionally, OpenAI and Broadcom are developing custom AI chips, while legal actions target AI training data misuse and cybersecurity vulnerabilities escalate industry concerns.

▶️ Internet Infrastructure

Why Switching from Docker to Podman Boosts Security and Efficiency

Switching from Docker to Podman enhances security, reduces resource footprint, and simplifies container orchestration by eliminating the root-privileged daemon, with seamless CLI compatibility and Kubernetes support.

  • Transitioned from Docker to Podman for container management, citing security, resource efficiency, and architecture benefits.
  • Docker’s root-privileged daemon architecture poses security vulnerabilities, including CVE-2019-5736, CVE-2024-21626, and others.
  • Podman operates without a persistent daemon, runs containers as the invoking user, and supports systemd integration, Kubernetes alignment, and containerized workflows.

UC Santa Cruz Develops Cost-Effective WiFi Heart Rate Monitor Using Machine Learning

UC Santa Cruz researchers developed Pulse-Fi, a WiFi-based heart rate monitoring system using low-cost devices and machine learning, achieving clinical accuracy within five seconds at distances up to 10 feet.

  • Pulse-Fi uses WiFi signals and machine learning to measure heart rate with clinical accuracy, using low-cost ESP32 chips ($5-$10) and Raspberry Pi (~$30).
  • The system accurately detects heart rate within five seconds, with only 0.5 beats per minute error, regardless of the person’s position or distance up to 10 feet.
  • Experiments involved 118 participants across 17 body positions; performance remains robust at distances up to 3 meters, with potential for longer-range applications.

High-Capacity Servers Outperform Cloud Costs for Heavy Workloads

Using one large server with high core count and memory capacity can be cost-effective and efficient for most workloads, challenging the necessity of distributed architectures and cloud reliance.

  • A typical modern server, like Microsoft’s Azure AMD server, has 128 cores, 256 threads, 2-2.5 GHz clock speed, 256 MB L3 cache per CPU, and supports up to 8 TB of RAM.
  • Capabilities include serving 400 Gbps video streams, 1 million IOPS on NoSQL databases, 70k IOPS in PostgreSQL, 500k requests/sec to nginx, kernel compilation in 20 seconds, and 4K video rendering at 75 FPS.
  • Cost comparisons: OVHCloud offers similar servers for $1,318/month; Hetzner for €140/month; AWS m6a.metal costs ~$6,055/month; Dell servers around $40,000; buying servers breaks even in 8 months versus cloud, 30 months versus renting.

Tempo Blockchain Delivers High-Speed, Low-Cost Stablecoin Payments

Tempo is a layer 1 blockchain designed for high-volume, low-cost payments, supporting over 100,000 TPS with sub-second finality, stablecoin interoperability, and compliance features.

  • Tempo is a purpose-built, layer 1 blockchain optimized for payments, supporting all major stablecoins, developed with partners including Stripe, Paradigm, and Deutsche Bank.
  • Processes over 100,000 TPS with sub-second finality, offering predictable near-zero transaction fees payable in any stablecoin.
  • Features include stablecoin interoperability, batch transfers, compliance controls (blocklists/allowlists), memo fields, and a dedicated payments lane.

X Shadow Bans Turkish Opposition Leader’s Posts Ahead of Elections

X (Twitter) appears to secretly shadow ban Turkish opposition leader Ekrem İmamoğlu’s posts, limiting visibility despite active posting, amid political suppression following his disqualification.

  • Turkish opposition leader’s Twitter account with 9.7 million followers is restricted in Turkey; a new account created by him shows limited growth and engagement.
  • Evidence suggests X (formerly Twitter) is secretly shadow banning Ekrem İmamoğlu’s posts, with 56% of users never seeing his tweets, despite regular posting.
  • The shadow banning appears politically motivated, following his disqualification and legal actions perceived as efforts to suppress his influence ahead of elections.

US Judge Allows Google to Keep Chrome but Limits Monopoly Practices

A federal judge permitted Google to keep Chrome without exclusive contracts, ordered data sharing, and limited remedies in the antitrust case, avoiding severe penalties and impacting search market competition.

  • U.S. District Judge Amit Mehta ruled Google can retain its Chrome browser but cannot enforce exclusive contracts for distribution.
  • The decision follows nearly one year after Google was found to hold an illegal monopoly in internet search; Google pays Apple billions annually to be default search on iPhones.
  • Google is ordered to share certain search index and user interaction data on commercial terms, but not ads data; the court rejected proposals to divest Chrome or Android assets.

Amazon’s Kuiper Satellites Hit 1.28 Gbps Speeds as JetBlue Plans In-Flight Wi-Fi

Amazon’s Project Kuiper achieved peak download speeds of 1.28 Gbps in staged tests at 630 km altitude, with JetBlue planning to deploy Kuiper satellite Wi-Fi on flights starting 2027.

  • Amazon demonstrated peak download speeds of 1.28 Gbps via Project Kuiper satellites, with an average of 1,189 Mbps and peak over 1,289 Mbps during staged tests.
  • The Kuiper network operates at 630 km altitude, with initial satellite deployment totaling just over 100 satellites; compared to Starlink’s over 8,000 satellites.
  • JetBlue announced plans to install Kuiper satellite internet hardware on its fleet in 2027, offering free in-flight Wi-Fi, with current service via Viasat until then.

IBM Cloud Ends Free Human Support in 2026, Shifting to Self-Service and AI Tools

IBM Cloud will end free human support in January 2026, shifting to self-service, with AI tools like Watsonx support and paid plans from $200/month to improve support efficiency and reduce costs.

  • IBM Cloud will transition its Basic Support tier to a self-service model in January 2026, removing the ability to open or escalate support cases via portal or APIs.
  • The free Basic Support includes 24x7 case raising but lacks guaranteed response times and dedicated account managers; customers can still report hardware or backup issues via the Cloud Console and open billing cases.
  • IBM will continue offering the Watsonx-powered AI Support Assistant and introduce a “Report an Issue” tool with faster routing and expanded self-help documentation; paid plans start at $200/month for enhanced support.

SAP invests €20 billion to build European sovereign cloud solutions

SAP invests €20 billion over ten years to develop sovereign cloud solutions in Europe, focusing on data sovereignty, compliance with GDPR, and avoiding replication of hyperscaler infrastructure.

  • SAP commits €20 billion over ten years to expand sovereign cloud infrastructure in Europe, targeting public sector, regulated industries, and defense.
  • Offers three solutions: SAP Cloud Infrastructure (EU-stored data within SAP datacenters), Sovereign Cloud On-Site (customer-owned datacenter), and Delos Cloud (Germany-specific sovereign cloud).
  • Investment includes infrastructure, R&D, operations, compliance, and partnerships, building on existing European datacenters in Walldorf, Sankt Leon-Rot, and Frankfurt.

enterprises may spend 7.3 billion on Windows 10 support by 2025

Nexthink estimates that enterprises may spend over $7.3 billion on Windows 10 ESU support for about 121 million devices by October 14, 2025, amid rising migration challenges and stability issues with Windows 11.

  • Nexthink estimates the first-year cost of extended security updates (ESU) for Windows 10 could exceed $7.3 billion
  • Approximately 181 million enterprise Windows 10 devices are projected to remain in use by support deadline, assuming a 33% decrease from May to August 2025
  • Support for many Windows 10 editions ends October 14, 2025; devices missing upgrade deadlines will incur significant support costs at $61 per device

Mozilla Extends Firefox ESR 115 Support for Windows 7 Until 2026

Mozilla extends Firefox ESR 115 support for Windows 7 until March 2026, with support ending for 32-bit Linux in Firefox 145; security updates continue while OS upgrades are encouraged.

  • Mozilla extends Firefox ESR 115 support for Windows 7 until March 2026
  • Firefox 115 was the last supported version for Windows 7, 8, and 8.1
  • Support for 32-bit Linux will end with Firefox 145; 32-bit Linux support continues in Firefox ESR 140 until at least September 2026

Huawei to End UK Battery Storage Sales by 2025 Amid Market Challenges

Huawei plans to cease UK sales of its BESS products by end of 2025, citing commercial reasons; the move follows US and UK bans impacting Huawei’s UK revenue and infrastructure.

  • Huawei to discontinue its battery energy storage systems (BESS) in the UK by the end of 2025
  • Existing inventory will be cleared, warranties honored, but extended warranties are no longer sold
  • Suppliers informed on August 22; Huawei declined to comment on official plans

Europe’s Jupiter achieves exascale with Nvidia GPUs and plans Rhea-based cluster

Europe’s Jupiter supercomputer reached exascale performance at inauguration on September 5, 2025, with a GPU cluster of 6,000 nodes using Nvidia GH200 chips, while its general-purpose cluster based on SiPearl Rhea1 processors is expected in 2026.

  • Europe’s Jupiter supercomputer was inaugurated on September 5, 2025, at Jülich Supercomputing Center, surpassing the exascale threshold of 10^18 operations/sec
  • The Booster module, a GPU cluster with approximately 6,000 nodes, features Nvidia GH200 Grace Hopper superchips and Quantum-2 InfiniBand networking
  • The system targets large-scale simulations and AI training, with a planned general-purpose Cluster Module based on SiPearl Rhea1 processors (80-core Arm Neoverse V1), expected in late 2026, comprising 1,300 nodes

▶️ Open Source

Nuclear: Free Open-Source Music Player with Tauri Rewrite Plans

Nuclear is a free, community-driven streaming music player sourcing from platforms like YouTube and SoundCloud, with plans for a Tauri-based rewrite featuring Rust performance enhancements.

  • Nuclear is a free, open-source streaming music player focused on sourcing music from free internet platforms, licensed under AGPL-3.0.
  • Supports searching and playing music from YouTube, Jamendo, Audius, SoundCloud, with features like playlists, album info, lyrics, and scrobbling to Last.fm.
  • The project is currently unmaintained, with a planned rewrite to use Tauri instead of Electron, incorporate Rust for performance, and add features like auto-updates, theming, plugin system, and expanded metadata support.

Mastering Linear Algebra: Key Concepts and Applications in Data Science and More

The article explains core linear algebra topics—vectors, matrices, eigenvalues, spectral theorem, and applications in data science, graphics, and differential equations—highlighting their theoretical foundations and practical uses.

  • The article provides a comprehensive overview of linear algebra concepts, including vectors, matrices, transformations, eigenvalues, and spectral theorems.
  • It details applications in data science, machine learning, computer graphics, differential equations, and network analysis.
  • Emphasizes the importance of orthogonality, diagonalization, and positive definiteness for stability, optimization, and geometric interpretation.

FreeDroidWarn Alerts Users of 2026 Developer Verification Deadline

FreeDroidWarn is an Android library that shows a deprecation warning about Google’s 2026/2027 developer verification requirements, which will disable app functionality on certified devices.

  • The FreeDroidWarn library alerts users about upcoming Android developer verification requirements starting in 2026/2027, which the app developer does not plan to comply with.
  • It displays a deprecation warning dialog informing that the app will cease functioning on certified Android devices after the deadline.
  • The library is licensed under Apache V2.0, available via JitPack, and can be integrated with implementation 'com.github.woheller69:FreeDroidWarn:V1.4'.

Herman Revises Bear License to Block Hosted Services and Protect Platform

Herman updated Bear’s license to a source-available Elastic License variant to prevent hosted service use, aiming to protect the platform from forks and commercial copying.

  • Herman changed Bear’s license from MIT to a copyleft license based on the Elastic License on September 1, 2025
  • The new license prohibits providing the software as a hosted or managed service, while allowing other uses
  • The license aims to prevent “free-ride competition” and protect the platform from forks and commercial exploitation

Nokia Sans Font Reignites as a Top Choice for High DPI UI Design

Nokia Sans, historically used by Nokia from 2002–2013, proves to be a highly legible and characterful UI font, especially the Wide variant, suitable for high DPI displays and Linux environments.

  • Nokia Sans font, used by Nokia devices from 2002 to 2013, was tested as a user interface font.
  • The Nokia Sans Wide variant is recommended for UI use due to its legibility and character.
  • The author found Nokia Sans effective on high DPI displays, especially in KDE on Wayland, and prefers it over Inter.

Linux Mint 22.2 “Zara” Launches with Cinnamon 5.8 and Kernel 6.14

Linux Mint 22.2 (“Zara”) launched on September 5, 2025, features Linux kernel 6.14, Ubuntu 24.04.3 base, Cinnamon 5.8, and hardware enablement updates, improving hardware compatibility and user interface.

  • Linux Mint 22.2 (“Zara”) released on September 5, 2025, based on Ubuntu 24.04.3 with Linux kernel 6.14 and hardware enablement updates
  • Introduces Cinnamon 5.8, minor UI enhancements, new fingerprint configuration app, improved Hypnotix IPTV, and companion apps for Android and iOS
  • Maintains Xfce 4.18, with themes slightly bluer, updated Software Manager, and better distinction between native and Flatpak apps; supports Ubuntu’s Hardware Enablement stack

OpenAI Teams Up with Broadcom to Develop Cost-Reducing Custom AI Chips

OpenAI is developing a proprietary AI accelerator with Broadcom’s support, leveraging 3.5D XDSiP packaging and custom compute units, to reduce costs and dependence on Nvidia, with chips expected in 2026.

  • OpenAI developing custom AI accelerator with Broadcom, aiming to reduce reliance on Nvidia and lower GPT model costs
  • Broadcom’s $10 billion customer, likely OpenAI, orders over $10 billion of AI racks based on XPUs, with three existing customers and a fourth upcoming
  • OpenAI’s chip expected to debut in 2026, primarily for internal use; design involves Broadcom’s 3.5D XDSiP tech and high-performance matrix multiply-accumulate units

▶️ Software Development

GitHub Community Demands Blocking of Copilot Amid Unresolved Concerns

Community complaints about Microsoft’s forced Copilot features on GitHub include unresolved requests to disable AI code reviews, concerns over training data, and migration efforts, amid Copilot’s 20 million users and 75% enterprise growth.

  • GitHub users request blocking Copilot from generating issues and pull requests; discussions remain unresolved since May and June 2025
  • Microsoft’s GitHub Copilot reached 20 million users, with enterprise adoption up 75% quarter-over-quarter as of July 2025
  • Critics cite AI training on proprietary code, lack of opt-out options, and forced integration, prompting community migration to alternatives like Codeberg

▶️ Management and Leadership

Critic Challenges Next.js Middleware for Poor Context and Data Handling

Dominik Meca criticizes Next.js’s middleware and logging architecture, citing limitations in context propagation, reliance on headers, and poor developer experience, advocating for more flexible solutions like SvelteKit.

  • The author criticizes Next.js middleware for limited parameter passing, lack of middleware chaining, and inconsistent async context handling.
  • Demonstrates issues with logging setup using AsyncLocalStorage, which fails to maintain context across middleware, pages, and server code.
  • Highlights that Next.js’s reliance on headers and route rewrites limits effective data passing, contrasting with SvelteKit’s locals approach for persistent request data.

The Power of Emotions in Design: Feeling Your Work Beyond Technical Metrics

Mitchell Hashimoto emphasizes that feeling the emotional impact of work is essential, as it influences user satisfaction and product quality beyond technical metrics.

  • Article emphasizes the importance of experiencing and understanding the emotional impact of work, beyond meeting technical requirements
  • Highlights that feelings like frustration, joy, relief, and confidence are integral to user experience
  • Advocates for designers and developers to “feel” their work by engaging with it directly, not just checking boxes

Palantir CEO Predicts AI Experts Will Become Highly Valuable and Well Paid

Palantir CEO Alex Karp predicts that skilled tech workers, especially AI experts, will become highly valuable and highly paid, while the company may adopt leaner teams despite strong revenue growth.

  • Palantir CEO Alex Karp states skilled tech workers will become increasingly valuable amid rising AI competition
  • Top technical talent will demand higher pay, with “artist-shaped” experts becoming “crazy valuable”
  • Despite revenue exceeding $1 billion in Q2 and nearly doubling US commercial revenue, Palantir plans to reduce team sizes

Amazon Limits AI Talent War Due to Pay Structure and Policies

Amazon’s internal document reveals its limited participation in the AI talent war, citing pay structure, reputation, and RTO policies as key obstacles to competing with Meta, OpenAI, and Microsoft for top AI engineers.

  • Amazon has largely abstained from the AI talent war due to its pay model, reputation lag, and strict return-to-office policies.
  • An internal document from late 2024 cites fixed salary bands, egalitarian pay philosophy, and limited salary increases as barriers to attracting top AI talent.
  • Competitors like Meta, OpenAI, Google, and Microsoft actively recruit high-profile AI engineers with more aggressive compensation packages, leading to Amazon’s lower retention and hiring rates.

Anthropic Valuation Soars to $183B, Boosting Employee Stock Options

Anthropic’s rapid valuation increase from $12.5 billion to $183 billion has significantly boosted startup equity values, with stock options now estimated between $5 million and $23 million.

  • A senior software engineer at Anthropic received a $571,000 compensation package, including approximately $250,000 in stock options, 14 months ago
  • The equity’s estimated value has increased from about $1 million to between $5 million and $23 million due to Anthropic’s valuation growth
  • Anthropic’s valuation surged from approximately $12.5 billion in early 2024 to $183 billion by September 2025

Legal and legislative changes in Delaware, Texas, and Nevada have facilitated easier approval of massive executive compensation, exemplified by Musk’s $1 trillion Tesla pay plan, following Delaware’s legal challenges and corporate relocations.

  • Elon Musk’s previous $56 billion Tesla pay package was blocked by a Delaware judge due to conflicts of interest and lack of board independence.
  • Tesla and other corporations relocated from Delaware to Texas and Nevada, prompting Delaware to amend its corporate laws to ease executive compensation approvals.
  • Delaware’s legal reforms made it significantly easier for public companies to authorize large executive pay packages, exemplified by Musk’s $1 trillion Tesla compensation plan.

Tesla’s $1 Trillion Pay Plan Pushes Elon Musk to Develop Succession Framework

Tesla’s $1 trillion pay package for Elon Musk mandates succession planning, indicating long-term leadership stability; Musk must develop a CEO succession framework to earn later milestones.

  • Tesla’s $1 trillion pay package for Elon Musk requires him to develop a CEO succession plan.
  • The package includes 12 milestones over 10 years, such as reaching an $8.5 trillion valuation, selling 12 million cars, and deploying 1 million robotaxis.
  • Tesla’s SEC filing states Musk must create a “CEO Succession Framework” for the 11th and 12th tranches to earn shares.

Andy Jassy’s Amazon Leadership Shift Focuses on Cost and Profitability

Since becoming Amazon CEO in 2021, Andy Jassy has restructured the company’s culture to prioritize cost efficiency, performance, and focused innovation, moving away from Bezos’s broad project bets.

  • Andy Jassy became Amazon CEO in 2021, succeeding Jeff Bezos, overseeing over 1.5 million employees during the COVID-19 pandemic.
  • Jassy has implemented a culture shift emphasizing cost stabilization through layoffs, project cuts, and tighter financial controls.
  • He promotes a performance-first “arena” leadership tone and shifts Amazon’s innovation focus from spreading bets to concentrating on fewer, profitable “big bets.”

Atlassian to Acquire The Browser Co. for $610M to Boost AI-Enhanced Browsers

Atlassian plans to acquire The Browser Co. for $610 million to integrate AI browser technology, enhancing enterprise tools like Jira, amid interest from AI startups and ongoing browser market competition.

  • Atlassian announced agreement to acquire The Browser Co. for $610 million in cash, aiming to close in fiscal Q2 ending December.
  • The Browser Co., founded in 2019, develops AI-enhanced browsers Arc and Dia, competing with Chrome and Safari, with Arc launched in 2022 and Dia in beta in June 2025.
  • The startup’s valuation was $550 million last year; investors include Atlassian Ventures, Salesforce Ventures, Dylan Field, and Reid Hoffman.
  • The Browser Co. has attracted interest from AI search startup Perplexity, which offered $34.5 billion for Chrome, and OpenAI, both reportedly considered acquiring the startup.
  • Atlassian intends to integrate The Browser Co.’s AI features and SaaS experience into its products, including Jira, to enhance enterprise capabilities and user productivity.

Trump Threatens Trade Investigation Over EU Fines on U.S. Tech Giants

Trump threatened a Section 301 trade probe against EU fines on Google and Apple after Google’s $3.5 billion antitrust penalty, alleging discrimination and unfair economic impact on U.S. tech firms.

  • Donald Trump threatened to initiate a Section 301 trade investigation to counteract EU fines against U.S. tech companies, citing discrimination.
  • The threat followed Google’s nearly $3.5 billion EU antitrust penalty related to advertising technology.
  • Trump’s social media post accused Europe of unfairly taking money from American firms like Google and Apple, claiming they paid $13 billion in false claims and $17 billion in back taxes respectively.

Stanford Study Finds 13% Drop in Young Workers’ Jobs Due to AI

Stanford research finds a 13% drop in employment for young workers in AI-exposed jobs since ChatGPT’s emergence, indicating AI is already impacting youth employment and job composition.

  • Stanford study analyzing ADP payroll data from mid-2025 shows a 13% employment decline among 22–25-year-olds in highly AI-exposed jobs like software development and customer service
  • Employment for older workers and less-exposed jobs, such as home health aides, remained steady or increased
  • The decline is concentrated in occupations with high AI automation potential, with evidence of within-firm employment shifts between exposed and less-exposed roles

Magic Lantern 2025 Midsummer Edition Boosts Camera Support and Features

Magic Lantern 2025: Midsummer Edition introduces a unified build system, supports new cameras (200D, 6D2, 750D, 7D2), and enhances features like raw video, extended recording time, with ongoing bug fixes.

  • Magic Lantern 2025: Midsummer Edition announced on June 21, 2025, with regular releases for all supported cameras, supported models include 200D, 6D2, 750D, 7D2
  • Major updates include a new build system, cleaner code, support for Digic 6 and 7, and new supported models; support for Digic 8 and above planned
  • Support for new features such as raw video with DPAF and Dual ISO on 200D, extended maximum recording time for MOV/MP4, and bug fixes across all cameras; no new code for EOS M or other unsupported models at this time

Why Startups Should Focus on Great Engineers Over the Best

Hiring “the best engineers” is impractical for startups; focusing on “great” engineers with realistic trade-offs accelerates hiring and aligns with startup speed and resource constraints.

  • The article argues against the common startup goal of hiring “the best engineers,” emphasizing that top-tier engineers are often unavailable or unaffordable for startups.
  • It highlights that “best” engineers typically have high salaries, strong opinions on tech debt, remote work preferences, and may avoid standard interview processes.
  • The author advocates for prioritizing “great” engineers over “best” ones by accepting trade-offs, such as experience level, work hours, or compensation, to reduce hiring time and increase candidate pool.

Apple sued for using copyrighted books without permission in AI training

Apple is sued for allegedly using copyrighted books without permission to train AI models, reflecting a wave of legal actions over intellectual property in AI development.

  • Apple sued in federal court in Northern California for allegedly using copyrighted books without consent, credit, or compensation to train AI systems.
  • The proposed class action claims Apple used pirated works, including known pirated books, to develop its “OpenELM” large language models.
  • Authors Grady Hendrix and Jennifer Roberson, who filed the lawsuit, state their works were part of the pirated dataset used by Apple.
  • The lawsuit follows similar cases against Microsoft, Meta, and Anthropic, with Anthropic settling for $1.5 billion in a related author class action.
  • The lawsuit highlights ongoing legal disputes over copyright protections in AI training data, amid broader industry concerns over intellectual property rights.

Vincent Quigley’s 6-Week Experience with Claude Code for AI-Assisted Development

Vincent Quigley details his 6-week journey using Claude Code for AI-driven development, emphasizing iterative coding attempts, context management via Claude.md files, tool integrations, and adapted review processes to improve productivity and code quality.

  • Vincent Quigley, Staff Software Engineer at Sanity, documents a 6-week experience with Claude Code for AI-assisted coding
  • AI development process involves three iterative attempts: 95% garbage on first, 50% on second, finally workable on third
  • Key solutions include Claude.md files for context, tool integrations (Linear, Notion, GitHub), managing multiple AI instances, and a three-step review process

Atos Recruits Leaders to Drive Growth Toward €10B Revenue by 2028

Atos recruited four experienced leaders, including Florin Rotar and Laurent Soulier, to support its transformation plan targeting €9-10 billion revenue and 10% margin by 2028 amid ongoing financial and strategic challenges.

  • Atos appointed four senior executives, including Florin Rotar as Group CTO and Laurent Soulier as VP for strategy and operational excellence
  • Atos aims to generate €9-10 billion ($10.48-11.64 billion) in revenue with 10% operating margin by 2028, up from €8.5 billion in 2025
  • The company has a history of leadership turnover, failed divestitures, and recent restructuring efforts focused on AI and digital transformation

UK public support for moving away from Microsoft as government plans £9B partnership

A poll shows 93% of Register readers favor shifting from Microsoft in UK public sector; government plans to spend £9B with Microsoft over five years amid calls for open source alternatives.

  • 93% of over 1,500 Register readers support moving away from Microsoft in the UK public sector
  • UK government plans to spend £9 billion with Microsoft over five years, with £1.9 billion annually under existing agreement
  • The 2024–2029 Strategic Partnership Arrangement (SPA24) aims to provide “enhanced value” across Microsoft products for public bodies

Oracle Cuts Over 12,000 Jobs Amid Cloud and AI Expansion

Oracle laid off over 12,000 employees worldwide, including 101 in Washington and 250+ in California, amid plans for additional cuts in October, following FY2025 revenue of $57.4 billion.

  • Oracle laid off over 100 employees in Washington and more than 250 in California, with estimates of nearly 12,000 affected globally.
  • Notices filed include 101 jobs in Seattle (effective Nov 3), 36 in Pleasanton, 187 in Redwood Shores, and 31 in Santa Clara, affecting roles such as Application Developer, IT Manager, Data Scientist, and Support positions.
  • Oracle reported $57.4 billion revenue for FY2025, up 8%, with CEO Safra Catz optimistic about FY2026 growth; layoffs follow a period of aggressive cloud talent recruitment and AI expenditure.

Critical SAP S/4HANA Flaw Allows Full System Takeover via Code Injection

A critical, actively exploited SAP S/4HANA bug (CVE-2025-42957) enables arbitrary ABAP code injection, bypassing authorization, creating backdoors, and risking full system compromise; patch immediately.

  • Critical code-injection vulnerability in SAP S/4HANA (CVE-2025-42957) allows low-privileged attackers to take over systems
  • Exploited actively; SAP issued August patches addressing the flaw
  • Demonstrated in labs how attackers can create SAP superuser accounts with SAP_ALL privileges, enabling full system control and data manipulation

FCC plans to revoke funding for Wi-Fi on school buses and library hotspots, citing unlawful expansions and safety concerns, potentially impacting internet access for students and disadvantaged communities.

  • FCC Chair Brendan Carr declared Biden-era E-Rate expansions funding Wi-Fi on school buses and library hotspots unlawful, proposing their elimination.
  • The expansions, funded by the E-Rate program, were deemed unauthorized by Congress for off-campus broadband after pandemic subsidies ended.
  • Carr and Senator Ted Cruz linked the program’s termination to children’s safety concerns, citing risks of unsupervised screen time and online harms.

US Army awards $98.9M contract to TurbineOne for AI threat detection system

The US Army contracted TurbineOne for its FPS platform, a no-code AI system that detects battlefield threats using multi-sensor data, operating offline for rapid deployment.

  • US Army awarded San Francisco startup TurbineOne a five-year contract up to $98.9 million for its Frontline Perception System (FPS)
  • FPS is a model-agnostic, no-code machine learning platform that ingests multi-sensor data to detect threats like drones and enemy positions
  • Operates offline on local devices, enabling soldiers to build, retrain, and deploy custom models, with actionable intelligence delivered to mobile devices

Wikipedia’s Resilience Amid Challenges in Rapid News Editing and Moderation

Wikipedia remains resilient as the web’s factual foundation but faces challenges in rapid content moderation during breaking news, exemplified by the Elon Musk controversy on January 20, 2025.

  • The article discusses Wikipedia’s role as the web’s primary factual resource and its current vulnerabilities.
  • An incident on January 20th, 2025, highlighted challenges in editing speed and content moderation during breaking news, exemplified by the Elon Musk Nazi salute controversy.
  • A single edit added a report of Musk’s gesture, which was quickly removed for violating standards, illustrating the tension between rapid updates and content accuracy.

Trump Hosts Tech CEOs Pledging Billions in US Investments

Trump hosted a dinner with top tech CEOs at the White House, where they pledged billions in US investments and praised his AI and infrastructure policies, emphasizing power dynamics.

  • A White House dinner on September 4, 2025, featured US President Donald Trump hosting major tech CEOs including Mark Zuckerberg, Bill Gates, and Sam Altman.
  • CEOs publicly pledged significant US investments, with Zuckerberg estimating approximately $600 billion through 2028; Apple committed $600 billion to US manufacturing.
  • The event highlighted bipartisan praise for Trump’s AI policies, infrastructure investments, and efforts to attract tech industry capital, with notable seating arrangements reflecting status.

MIT’s “How to Speak” Course Teaches Public Speaking Heuristics

MIT’s “How to Speak” course by Patrick Winston, available on YouTube, offers 1-hour guidance on public speaking heuristics, tools, and techniques for effective communication.

  • The YouTube video titled “How to Speak” was uploaded by MIT OpenCourseWare on December 20, 2019, with 20 million views.
  • The lecture, part of MIT’s “How to Speak” course taught by Patrick Winston, lasts 1 hour, 3 minutes, and 42 seconds, covering public speaking heuristics.
  • Key segments include rules of engagement, starting techniques, heuristics, tools like time, place, boards, props, slides, and methods for informing and persuading audiences.

▶️ Technology

Inside Nano-GPT: How 85K Parameters Power Token Sorting

The article explains the architecture and functioning of a nano-GPT model with 85,000 parameters, illustrating token embedding, transformer layers, and next-token prediction for sorting tokens.

  • Demonstrates the architecture of a nano-GPT model with 85,000 parameters for token sorting
  • Uses a vocabulary of three tokens: A (0), B (1), C (2), with input sequence “2 1 0 1 1 2”
  • Explains embedding, transformer layers, and next-token prediction process, including probability outputs

Sustainable Off-Grid Blogging with a Recycled Pixel 5 and Solar Power

Dom Corriveau’s blog operates on a recycled Pixel 5 with solar power, utilizing Termux for hosting and automation, demonstrating a sustainable, off-grid web hosting solution with technical setup details.

  • The blog runs on a recycled Google Pixel 5, powered by a 100W solar panel connected to a Jackery 160W power station
  • Uses Termux to install utilities like git, hugo, ssh, rsync, and cronie for site management and automation
  • Site performance remains fast and reliable, with setup involving SSH access, cron jobs, and screen sessions for hosting the Hugo static site

HunyuanWorld-Voyager: Real-Time 3D Scene Generation from Single Images

HunyuanWorld-Voyager, released on September 2, 2025, is a video diffusion framework that produces world-coherent 3D scene videos and real-time RGB-D reconstruction from a single image.

  • HunyuanWorld-Voyager is an interactive RGB-D video generation model conditioned on camera input, supporting real-time 3D reconstruction.
  • Released on September 2, 2025, with code and model weights available here.
  • Generates world-consistent 3D point-cloud sequences from a single image along user-defined camera paths, enabling scene exploration and aligned depth and RGB videos.

Le Chat Launches 20+ MCP Connectors and Memories for Enterprise Workflow

Le Chat integrates 20+ MCP-powered connectors and Memories (beta) for personalized context, offering flexible deployment, with full control over data, enhancing enterprise workflow automation and data access.

  • Le Chat now integrates with 20+ enterprise platforms via MCP, including Databricks, Snowflake, GitHub, Atlassian, and Stripe.
  • Introduces Memories (beta), enabling personalized, reliable context retention with full control over stored data and fast import from ChatGPT.
  • Features a flexible, secure MCP connector directory (beta), supporting data, productivity, development, automation, commerce, and custom integrations; deployable on-premises, in cloud, or Mistral Cloud.

Cloudflare Radar tracks AI bot and crawler activity, including HTTP trends, crawl purposes, and AI service popularity, highlighting AI model usage and robots.txt directives across domains.

  • Cloudflare Radar reports on AI bot and crawler traffic over the past 7 days, including HTTP request trends, crawl purposes, and domain category data
  • Analyzes crawl-to-refer ratios, AI bot best practices, and popularity of generative AI services based on DNS resolver traffic
  • Provides insights into AI user agents in robots.txt files across top 10,000 domains, and distribution of AI models and tasks on Cloudflare Workers

Python 3.14 Boosts Concurrency Amid Ongoing Async Limitations

Python 3.14 enhances concurrency with free-threading and multiple interpreters, but async remains niche due to GIL limitations, error complexity, and partial framework support, limiting its overall popularity.

  • Python 3.14 release is weeks away, introducing PEP779 (free-threading support) and PEP 734 (multiple interpreters in stdlib)
  • Async support, introduced in Python 3.5, remains limited in popularity due to error handling, GIL constraints, and partial framework support
  • Main use-case for async is I/O-bound tasks like HTTP requests; CPU-bound tasks and disk I/O are less suited, with asyncio lacking native filesystem async support

Anthropic Raises $13B in Series F at $183B Valuation Amid Rapid Growth

Anthropic raised $13 billion in Series F funding at a $183 billion valuation, boosting revenue from $1 billion to over $5 billion by August 2025, supported by rapid customer growth and advanced AI models.

  • Anthropic completed a $13 billion Series F funding round led by ICONIQ, valuing the company at $183 billion post-money.
  • The round included investors such as Fidelity, Lightspeed Venture Partners, Altimeter, Baillie Gifford, BlackRock, Blackstone, Coatue, D1 Capital, and others.
  • Since March 2023, Anthropic’s revenue grew from approximately $1 billion to over $5 billion by August 2025, with over 300,000 business customers and a nearly 7x increase in large accounts (> $100,000 revenue).

CPU Utilization Overestimates at Low Loads and Underestimates at High Loads Due to Hyperthreading and Turbo

CPU utilization reports are inaccurate due to hyperthreading and Turbo effects, causing actual work capacity to be significantly higher than indicated, necessitating benchmarking and direct work measurement.

  • CPU utilization is overestimated at low loads and underestimated at higher loads due to hyperthreading and Turbo behavior, with actual work capacity exceeding reported metrics.
  • Stress tests on a Ryzen 9 5900X (12 cores / 24 threads) showed that at “50%” CPU utilization, actual work done ranges from 60-85%, with matrix math reaching 80-100%.
  • Benchmarking with stress-ng and real-world Nginx tests revealed discrepancies between reported CPU utilization and actual request throughput, influenced by shared resources and clock speed adjustments.

Best 48- and 50-Inch TVs of 2025: OLED, QLED, Mini LED Picks

The article compares top 48- and 50-inch TVs of 2025, emphasizing OLED, QLED, and Mini LED models like LG C4, Samsung QN90D, and Amazon Omni QLED, with detailed specs and performance insights.

  • The article reviews the best 48- and 50-inch TVs of 2025, including models from LG, Samsung, Amazon, TCL, Vizio, and more
  • Highlights include OLED, QLED, Mini LED, and budget options, with detailed specs, features, and pricing
  • Testing criteria encompass contrast, HDR brightness, color gamut, viewing angles, refresh rate, smart OS, and value

OpenAI researchers attribute hallucinations in large language models to training methods favoring guessing over uncertainty, proposing evaluation metric redesigns to reduce false factual claims.

  • OpenAI researchers identify that large language models hallucinate because they are trained to prioritize guessing over uncertainty acknowledgment.
  • Hallucinations occur when models confidently generate inaccurate information as facts, affecting models like GPT-5 and Claude.
  • Redesigning evaluation metrics to penalize guessing and encourage uncertainty expression can mitigate hallucinations, as suggested in a recent paper and blog post.

Minimal Phone MP01: E-Paper Android with Long Battery and Keyboard

The MP01 is an Android epaper phone with a physical keyboard, offering 2-day battery life, adjustable refresh rates, and improved reading experience, but with bugs and slow updates.

  • The Minimal Phone (MP01) is an Android device with a black-and-white epaper display, featuring a physical keyboard and a size comparable to Pixel 8
  • It offers adjustable refresh rates (slow, hybrid, fast) to balance ghosting, flickering, and image quality, with a battery life of approximately 2 days per charge
  • The device has bugs including occasional fingerprint reset, screen refresh rate issues affecting keyboard input, and slow software updates; a pending update aims to address these

Nvidia Warns U.S. AI Export Restrictions May Hamper Innovation and Competition

Nvidia warns that the U.S. CHIPS Act’s AI export restrictions and the proposed 2025 AI diffusion rule could restrict competition and innovation in AI development.

  • Nvidia states the U.S. CHIPS Act’s AI export restrictions could limit competition and innovation
  • The company compares the AI diffusion rule to the 2025 AI Act, implying similar restrictive impacts
  • Nvidia warns that such regulations may hinder global AI development and market dynamics

Goldman Sachs Warns AI Boom to Drive Datacenter Growth and Energy Use

Goldman Sachs warns the AI-driven datacenter capacity may reach 92 GW by 2027, increasing energy use and semiconductor revenues, but market weakness risks remain amid overhyped adoption.

  • Goldman Sachs forecasts datacenter capacity will increase by 50% to 92 GW by 2027, driven by AI demand.
  • Sector’s energy consumption is expected to double by 2030, with power use rising 165% globally by 2030.
  • AI workloads will constitute 28% of capacity by 2027, up from 13% in early 2023, with cloud share decreasing from 58% to 50%.

AI System A2 Discovers Over 100 Zero-Day Android Vulnerabilities

Researchers created A2, an AI-driven Android vulnerability system that found over 100 zero-day flaws, with 57 validated via automated exploits, achieving 78.3% coverage on benchmarks.

  • Researchers from Nanjing University and University of Sydney developed A2, an AI system for vulnerability discovery and validation in Android apps.
  • A2 achieved 78.3% coverage on the Ghera benchmark, surpassing static analyzers like APKHunt (30%), and identified 104 zero-day vulnerabilities in 169 production APKs, including a flaw with over 10 million installs.
  • The system uses multiple AI models (OpenAI o3, Gemini 2.5 Pro, Gemini 2.5 Flash, GPT-OSS) in roles: planner, executor, validator, to emulate human bug hunting, with validation based on concrete tasks such as extracting hardcoded keys, forging tokens, and proving authentication bypasses.

Chinese Cybercrime Group Infects 65 Servers to Manipulate Google Rankings with Custom Malware

GhostRedirector, a Chinese cybercrime crew, infected 65 Windows servers since August 2024, deploying custom malware to manipulate Google search rankings via SEO fraud techniques.

  • GhostRedirector, a China-aligned cybercrime group, compromised at least 65 Windows servers globally, active since August 2024, with infections detected in June 2025
  • Utilizes custom malware Rungan (passive C++ backdoor), Gamshen (IIS trojan manipulating Google search results), and other tools like privilege escalation exploits based on EfsPotato and BadPotato
  • Gains initial access via SQL injection, downloads payloads from 868id[.]com, and employs PowerShell for privilege escalation, creating backdoors and webshells
  • Malware signed with a TrustAsia RSA certificate issued to Shenzhen Diyuan Technology; tools include Comdai and Zunput for web data collection and dropper functions
  • Payloads Rungan and Gamshen enable backdoor commands and SEO fraud, respectively, manipulating Googlebot responses to artificially boost gambling sites’ rankings, primarily targeting South America and South Asia

Attackers Exploit Sitecore Vulnerability to Deploy Malware and Escalate Privileges

Attackers exploited CVE-2025-53690 in Sitecore products via publicly exposed sample machine keys, enabling remote code execution, malware deployment, and privilege escalation, with ongoing threat implications.

  • Attackers exploited a configuration vulnerability in Sitecore products (XM, XP, XC, Managed Cloud) related to CVE-2025-53690, a ViewState deserialization flaw.
  • The vulnerability affects versions deployed in multi-instance mode with customer-managed static machine keys, particularly those using sample keys from Sitecore XP 9.0 or earlier and Active Directory 1.4 or earlier.
  • Exploited publicly documented machine keys enabled remote code execution, malware deployment (WEEPSTEEL), and unauthorized data access, with attackers escalating privileges and conducting reconnaissance.

Researchers Uncover TAG-150’s Stealthy CastleRAT Malware Spread via ClickFix Attacks

Recorded Future reports that TAG-150 developed Python and C variants of CastleRAT malware, spreading via ClickFix social engineering, with the Python version optimized for stealth and the C version capable of advanced data theft.

  • Researchers at Recorded Future identified two CastleRAT malware variants in Python and C, spreading via ClickFix social engineering attacks.
  • The Python version is designed for stealth with minimal antivirus detection; the C version can harvest keystrokes, capture screens, and establish persistence.
  • Both variants can establish remote shell access and download additional malware; the C build can self-delete if needed.
  • The malware is operated by the TAG-150 cybercrime crew, which also developed CastleLoader and CastleBot, using encrypted Tox Chat for command and control.
  • The ClickFix technique involves fake login prompts and instructions to paste malicious code into PowerShell or Windows Run, achieving a 28.7% success rate.
  • The gang uses virtual private servers in the Netherlands and hosts malware from cloud providers like Google Cloud.
  • Recorded Future recommends monitoring ports 443, 7777, and 80 for suspicious activity to detect malware presence.

Microsoft Develops Light-Based Analog Optical Computer for Faster, Energy-Efficient Optimization

Microsoft’s Analog Optical Computer prototype employs light-based computation with 256 weights, targeting optimization in finance, clinical, and AI applications, with future scalability and datacenter integration.

  • Microsoft researchers in Cambridge developed an Analog Optical Computer (AOC) prototype utilizing light for solving complex optimization problems in finance, healthcare, and logistics
  • The prototype uses commercially available parts, including micro-LEDs, optical lenses, and sensors, with the technology based on decades-old photon fiber optics
  • The current machine has 256 weights, up from 64 in previous versions, with potential to scale to millions or over a billion weights; aims for room temperature operation, 100x faster and 100x more energy-efficient for specific tasks

EU Fines Google €3.5 Billion for Antitrust Violations in Ad Tech

The EU fined Google €3.5 billion for antitrust violations in its ad tech business, citing abuse of market dominance through restrictive clauses on third-party platforms.

  • European Union fined Google €3.5 billion ($3.7 billion) on March 20 for antitrust violations related to its ad tech business
  • The EU accused Google of abusing its market dominance by imposing restrictive clauses on third-party websites and app developers
  • The fine is part of ongoing EU investigations into Google’s advertising practices, which began in 2018

Zed Launches Public Beta of Claude Code Integration Using Open ACP Standard

Zed launched a Claude Code integration in public beta, enabling native, real-time code editing with syntax highlighting and granular change review via ACP, an open standard for agent integration.

  • Claude Code integration is now available in public beta within Zed, utilizing the open Agent Client Protocol (ACP) standard
  • Enables Claude Code to run natively in Zed, providing real-time editing, syntax highlighting, language server support, and granular change review
  • The integration is built using an open-source ACP adapter for Claude Code, available under Apache license, compatible with other ACP-supported editors like Neovim