Alvaro Lopez Ortega / 2025-06-15 Briefing

Created Sun, 15 Jun 2025 18:36:08 +0000 Modified Sat, 28 Jun 2025 02:44:34 +0000
7137 Words

Today’s tech highlights include the launch of Android 16 with live notifications and enhanced security, Apple’s debut of Liquid Glass for immersive interfaces, and Google Cloud’s introduction of Nvidia RTX PRO 6000 VMs propelling AI workloads. Meanwhile, Meta’s massive $14.3B investment in Scale AI signals fierce industry AI competition.

▶️ Internet Infrastructure

Containerization Swift Package Powers Linux Containers on macOS

Containerization is a Swift package for running Linux containers on macOS, offering APIs for OCI image management, remote registry interaction, and lightweight VM orchestration, relying on an optimized Linux kernel.

  • Containerization is a Swift package enabling Linux container runtime on macOS, utilizing Virtualization.framework on Apple Silicon.
  • Provides APIs for managing OCI images, interacting with remote registries, creating ext4 filesystems, and managing lightweight VMs via vminitd.
  • Requires macOS 15+, Xcode 26 beta, and macOS on Apple Silicon; supports fast container startup with an optimized Linux kernel configuration.

Android EthernetTracker ignores CDC Ethernet interfaces due to regex limitations

Android’s EthernetTracker ignores usbX interfaces due to regex eth\d, preventing CDC Ethernet gadgets from being recognized, despite kernel support for CDC standards like ECM, EEM, and NCM.

  • Android’s EthernetTracker only recognizes interfaces named ethX; Linux CDC Ethernet drivers create interfaces named usbX.
  • The default config_ethernet_iface_regex in Android is eth\d, causing usb0 to be ignored.
  • Support for CDC Ethernet standards (ECM, EEM, NCM) exists in the kernel (CONFIG_USB_NET_CDCETHER=y), but interface naming prevents detection on Android.

Google Cloud Report Shows No Major Incidents as of June 15, 2025

Google Cloud status page reports no major incidents as of June 15, 2025, 09:22 PDT, with comprehensive regional and product-specific health monitoring and incident tracking.

  • No major incidents reported on Google Cloud status page as of June 15, 2025, 09:22 PDT
  • Provides real-time status updates for Google Cloud services across multiple regions and products
  • Offers personalized service health, incident history, and external links for detailed monitoring and support

Waymo Rides Cost More Than Uber and Lyft in San Francisco

Obi’s analysis shows Waymo’s self-driving rides are consistently more expensive than Uber and Lyft, with higher variability and longer wait times, despite 250,000 weekly trips and customer preference for driverless cars.

  • Obi analyzed nearly 90,000 ride records in San Francisco from March 25 to April 25, 2025
  • Average ride prices: Lyft $14.44, Uber $15.58, Waymo $20.43
  • Waymo’s rides are approximately $11 more expensive than Lyft and $9.50 more than Uber during peak hours

Cloudflare Outage Disrupts Services Due to Third-Party Dependency Failure

Cloudflare’s June 12, 2025, outage, caused by a third-party dependency failure affecting Workers KV, disrupted multiple services including WARP, Access, and Dashboard, but was resolved within two hours.

  • Cloudflare experienced broad service outages affecting multiple products, including Workers KV, WARP, Access, Dashboard, and Turnstile, due to a third-party dependency failure.
  • The outage was first reported at 19:57 UTC on June 12, 2025, caused by an outage of a third-party service impacting Cloudflare’s critical Workers KV service.
  • All services were restored and fully operational by June 12, 2025, at 21:31 UTC, with ongoing monitoring to confirm stability.

Building Tech Independence: Self-Hosting with Open-Source Tools and Own Domains

The article details the author’s journey of building self-hosted services, advocating for owning a domain, leveraging open-source Linux tools, and achieving tech independence through DIY infrastructure.

  • The author shares personal experience self-hosting a blog, second brain, book, subscriber list, and home server with SSL, SSH, and reverse proxy setup
  • Emphasizes the importance of owning a domain for long-term content preservation and avoiding migration issues
  • Highlights open-source tools and Linux’s role in achieving tech independence, including Gitea, Listmonk, GoatCounter, and others

AI Accelerates JFK Files Declassification and Boosts Secret Cloud Infrastructure

AI has accelerated declassification of JFK files and supports classified government work; AWS will open a second Secret cloud region in 2025, enabling secure top-secret workloads.

  • AI enabled the US intelligence community to declassify tens of thousands of JFK assassination documents more rapidly than manual review.
  • Declassified documents revealed CIA activities in the 1960s, including attempts to set up foreign dignitaries with female companionship.
  • The US government has deployed an AI chatbot across all 18 intelligence agencies, supporting classified workloads up to the “Secret” level.
  • AWS announced the launch of its second Secret-level cloud region in 2025, supporting workloads up to the US Secret classification, complementing its first region established in 2017.
  • The “top-secret clouds” are designed for processing and storing sensitive information at the Secret classification level, with AWS, Microsoft Azure, and Google Cloud providing compliant offerings.
  • Gabbard emphasized partnerships with tech vendors like AWS to improve government technology solutions and reduce internal development efforts.

Ransomware Exploits Unpatched SimpleHelp Vulnerability Causing Disruptions

Ransomware groups exploited unpatched SimpleHelp CVE-2024-57727 in June 2025, leading to service disruptions and double extortion; the flaw was patched in January but not all users updated.

  • CISA issued a warning on June 12, 2025, about ransomware exploiting unpatched SimpleHelp CVE-2024-57727
  • The vulnerability is a high-severity path traversal flaw affecting SimpleHelp 5.5.7 and earlier, fixed in January 2025
  • Attackers exploited the flaw to infect downstream customers, causing service disruptions and double extortion schemes, including data theft and encryption

Google Cloud Launches Nvidia RTX PRO 6000 VMs for AI and Data Workloads

Google Cloud is the first to offer Nvidia RTX PRO 6000 Blackwell GPU-based VMs, featuring 8 GPUs with 3,753 TFLOPS sparse FP4, 96GB GDDR7 memory, supporting AI inference, fine-tuning, and visualization workloads, with general availability by late 2025.

  • Google Cloud announced the availability of Nvidia RTX PRO 6000 Blackwell Server Edition VMs, first to market with this GPU.
  • G4 VMs incorporate eight Nvidia RTX PRO 6000 GPUs, two AMD Turin CPUs, and Google Titanium offload processors.
  • Each GPU delivers 3,753 teraFLOPS of sparse FP4 compute, with 96GB GDDR7 memory and 1.6TB/s bandwidth; G4 VMs have 768GB memory, 384 vCPUs, 12 TiB SSD, expandable to 512 TiB.
  • The RTX Pro 6000 GPU is based on PCIe, with 96GB GDDR7, 12 TFLOPS FP32, and 3,753 TFLOPS sparse FP4, aimed at AI inference, model fine-tuning, and data visualization.
  • G4 VMs are currently in preview, with general availability expected by the end of 2025; they support workloads like physical AI, robotics, generative AI, and game rendering.
  • Google claims the G4 VM can handle larger models and higher throughput due to increased memory and bandwidth, compared to earlier G2 VMs.
  • The move signals Google’s focus on high-performance, GPU-accelerated cloud workloads, integrating Nvidia’s latest GPU technology into its AI Hypercomputer system.
  • Industry experts view this as a strategic showcase of Google’s cloud capabilities and a response to increasing demand for scalable AI infrastructure.

Google Cloud Outage Disrupts Services Including Cloudflare on June 12, 2025

Google Cloud’s outage on June 12, 2025, caused widespread service disruptions, including Cloudflare impacts, due to an IAM issue; recovery was underway within hours.

  • Google Cloud experienced a major outage on June 12, 2025, affecting over 40 locations and 26 services, including Cloud Console, Cloud Storage, and Cloud SQL.
  • The incident was caused by an Identity and Access Management Service issue, with recovery efforts ongoing as of 12:41 PDT.
  • Cloudflare’s services, which rely on Google Cloud, also experienced disruptions, with Cloudflare reporting impact on limited services before recovery began.

Oracle Reports Strong Q4 Revenue Fueled by Cloud Growth and Major TEMU Deal

Oracle’s Q4 2025 revenue exceeded forecasts at $15.9 billion, driven by 50% IaaS growth to $3 billion, with a significant deal involving TEMU, a large Chinese e-commerce firm migrating to Oracle Cloud.

  • Oracle reported $15.9 billion revenue for Q4 2025, an 11% YoY increase, with net income rising 9% to $3.4 billion
  • Cloud infrastructure (IaaS) revenue reached $3 billion, a 50% growth; full-year revenue grew 8% to $57.4 billion, net income up 19% to $12.4 billion
  • Oracle secured a major deal with TEMU, a rapidly growing Chinese e-commerce company, moving infrastructure to Oracle Cloud; Ellison described TEMU as “a very large company” transitioning its infrastructure

Concerns Rise Over Chinese-Owned VPN Apps on Apple and Google Stores

Apple and Google’s app stores host Chinese-owned VPN apps, raising concerns over user data access and compliance with US export controls, with Chinese law enabling government surveillance.

  • Apple and Google offer free VPN apps owned by Chinese companies in their app stores without clear disclosure of their Chinese origin
  • Researchers at Tech Transparency Project identified 20 top US VPNs with Chinese ownership; five were initially pulled, but two (Turbo VPN and VPN Proxy Master) remained available as of early May
  • The apps are linked to Chinese firms, including Qihoo 360, which has alleged PLA links and is on the US Commerce Department’s Entity List; Chinese law can compel data sharing with Beijing authorities

AST SpaceMobile gains 45 MHz spectrum from Ligado after a $550 million settlement, aiming for nationwide coverage with five satellites, competing against Starlink’s extensive constellation and Musk’s expanding D2C plans.

  • AST SpaceMobile secures 45 MHz spectrum from bankrupt Ligado after a legal settlement, involving a $550 million deal funded by a non-recourse senior-secured loan.
  • Ligado filed for bankruptcy in January 2025 due to US DoD objections over potential GPS interference; Ligado’s SkyTerra 1 satellite declared a total loss in 2024.
  • AST’s initial five satellites do not provide full US coverage; future launches aim for approximately 100% nationwide coverage, competing with Starlink’s 7,500 satellites.
  • Major US telcos—AT&T, Verizon, and T-Mobile—are developing D2C services, with T-Mobile’s Starlink beta launched in February 2025; Verizon’s service limited to select devices since March.
  • Elon Musk’s Starlink has over 7,500 operational satellites, with exclusivity for T-Mobile’s D2C service ending after one year, opening opportunities for other carriers.
  • AST’s satellite constellation is limited, with no specified launch date; its short window is threatened by Starlink’s expanding D2C dominance and Musk’s plans to serve multiple carriers.

Barclays Corporate Banking Services Suffer Unexpected Outage

Barclays’ corporate banking platforms, including iPortal and Barclays.Net, suffered an unexpected outage on June 13, 2025, disrupting digital services during peak hours, with ongoing efforts to restore functionality.

  • Barclays experienced unplanned outages affecting its corporate banking services, including iPortal and Barclays.Net, starting before 8am UTC on June 13, 2025.
  • The outage disrupted access to digital services such as balance viewing, transactions, and reporting, with reports flooding in via Down Detector.
  • Barclays confirmed ongoing issues and stated they are working to resolve the problems; the outage followed unlisted scheduled maintenance on June 1 and 7.

Enterprise AI Adoption Stalls Amid Rising Cloud Costs and Unpredictable Inferencing Expenses

Enterprise AI deployment stalls due to unpredictable inferencing costs amid rising cloud expenses, with public cloud costs deemed unsustainable for large-scale AI inference, prompting shifts to colocation and specialized providers.

  • Enterprise AI adoption is hindered by unpredictable inferencing costs amid rising cloud infrastructure expenses.
  • Global cloud spending reached $90.9 billion in Q1 2025, up 21% YoY, driven by workload migration and generative AI exploration.
  • Inferencing costs are recurring and volatile, leading enterprises to restrict usage, reduce model complexity, or seek alternative hosting solutions.

HDD sales decline sharply outside hyperscale datacenters, with 2025 shipments dropping to 150 million; nearline HDDs remain vital for large-scale, infrequently accessed data, amid AI-driven storage demand growth.

  • In 2025, approximately 150 million HDD units are expected to ship, down from 600 million in 2010, with most sales to hyperscale datacenter operators.
  • HDD sales to non-hyperscale enterprises are increasingly rare; buyers prefer all-flash storage and QLC SSDs for primary workloads.
  • Despite declining volumes, total HDD capacity shipped annually is rising due to larger disk sizes, with nearline HDDs remaining the dominant data storage solution.

PCIe 7.0 Standard Doubles Speed to 128 GTps, paving Way for 1 TBps PCIe 8.0

PCIe 7.0 standard finalized at 128 GTps (512 GBps bi-directional in x16), with PCIe 8.0 pathfinding underway to meet rising datacenter bandwidth demands, potentially reaching 1 TBps.

  • PCIe 7.0 specification finalized with a raw bit rate of 128 GTps, doubling previous PCIe 6.0 (64 GTps)
  • Supports up to 512 GBps bi-directional bandwidth in an x16 lane configuration
  • PCIe 8.0 development has begun, with potential to reach 1 TBps bandwidth, but no guarantee of continued doubling; optical and electrical implementations expected

Rocky Linux 10 and AlmaLinux 10 Diverge on Hardware Support and AI Features

Rocky Linux 10 and AlmaLinux 10 diverge in hardware support and features, with Rocky dropping x86-64-v2 support and AlmaLinux maintaining it; RHEL 10 introduces an AI shell assistant, impacting distribution differentiation.

  • Rocky Linux 10 “Red Quartz” reached general availability, supporting RISC-V and Raspberry Pi 4/5; drops support for Pi 3 and Zero.
  • AlmaLinux 10 “Purple Lion” released end of May, supporting x86-64-v2 hardware; Rocky Linux 10 now requires x86-64-v3, dropping v2 support.
  • RHEL 10 introduced an AI assistant called Red Hat Enterprise Linux Lightspeed, accessible via the c command, with a 8.5 GB ISO size; Lightspeed requires online connection and is absent in AlmaLinux and Rocky Linux.

▶️ Open Source

Spyware from Israeli Firm Paragon Targets European Journalists in 2025

Citizen Lab revealed that spyware from U.S.-backed Israeli firm Paragon Solutions targeted European journalists’ devices in 2025, raising concerns over abuse of commercial spyware in democracies.

  • Citizen Lab forensic evidence shows spyware from Israeli company Paragon Solutions targeted at least three European journalists, including an Italian editor, in 2025.
  • Paragon’s spyware, Graphite, infected iPhones via iMessage and Android devices, allowing covert access to encrypted apps like WhatsApp and Signal without user action.
  • Paragon, backed by former Israeli Prime Minister Ehud Barak and acquired by AE Industrial Partners for over $500 million, has U.S. government contracts including a $2 million deal with ICE and reportedly with the DEA.

Effective AI-Assisted Development Tips for Better Code Quality

The post outlines effective AI-assisted development practices, emphasizing structured documentation, guardrails, human-written tests, and context management to optimize productivity and code quality.

  • The post shares practical insights on AI-assisted software development, emphasizing deliberate practices to enhance productivity and code quality.
  • It details infrastructure tools like CLAUDE.md, commit strategies, guardrails, and anchor comments to guide AI interactions effectively.
  • Highlights the importance of humans writing tests, establishing boundaries, and managing context to prevent AI errors and ensure maintainability.

Emulating Apple Silicon iPhone 11 on QEMU with Open Source Tools

Emulates Apple Silicon iPhone 11 on QEMU; repository contains source code, documentation, and licensing info, with over 1,500 stars, supporting GPL-2.0 and LGPL-2.1 licenses.

  • Emulates Apple Silicon devices on QEMU, currently only iPhone 11.
  • Repository includes code, documentation, and licensing info, with 1.5k stars and 81 forks.
  • License details include Unknown, GPL-2.0, and LGPL-2.1 licenses.

Resemble AI’s Chatterbox: Open-Source TTS Model with Advanced Features

Resemble AI’s Chatterbox is a leading open-source TTS model with 0.5B Llama backbone, emotion control, trained on 0.5M hours, watermarked outputs, and superior performance.

  • Resemble AI’s Chatterbox is a state-of-the-art open-source TTS model licensed under MIT, benchmarked against closed-source systems like ElevenLabs
  • Features include 0.5B Llama backbone, emotion exaggeration control, alignment-informed inference, trained on 0.5 million hours of cleaned data, and watermarked outputs
  • Supports easy voice conversion and outperforms ElevenLabs in quality; available via Hugging Face Gradio app

jemalloc Ends 20-Year Development After Major Technical Milestones

jemalloc’s open-source development concluded in 2025 after 20 years, with key phases including FreeBSD integration, Firefox porting, Facebook/Meta adoption, and technical innovations like size segregation and huge page support.

  • jemalloc was conceived in 2004, with active development ending by 2025 after nearly 20 years
  • Development phases include Lyken (2004-2006), FreeBSD integration (2005-2006), Firefox porting (2007-2008), Facebook involvement (2009-2017), and Meta reorganization (2016-2025)
  • Major technical milestones: transition to size-segregated regions (2006), extensive profiling and telemetry support (2010s), huge page allocation (2016+), and removal of Valgrind support (2019)

Y Combinator’s Spring 2025 Cohort Highlights 70 Agentic AI Startups

Y Combinator’s first spring 2025 cohort featured 70 agentic AI startups, with notable companies like Aegis, Approval AI, and Galen AI developing autonomous systems across healthcare, fintech, and B2B sectors.

  • YC’s spring 2025 batch included 70 startups focused on agentic AI, with a $500,000 investment per company
  • The cohort ran from April to June in San Francisco, marking YC’s first spring program
  • BI identified 10 notable AI agent startups across industries including healthcare, fintech, B2B, consumer, and industrials

Denmark to Replace Microsoft with Linux and LibreOffice by 2025

Denmark’s Ministry of Digital Affairs plans to eliminate Microsoft dependency by adopting Linux and LibreOffice, supporting national digital sovereignty and open-source collaboration by fall 2025.

  • Danish Ministry of Digital Affairs to fully replace Microsoft software with Linux and LibreOffice by fall 2025, starting with 50% of employees this summer
  • Move aligns with Denmark’s digital sovereignty strategy and aims to reduce dependence on US tech providers amid geopolitical concerns
  • The initiative follows municipal efforts in Copenhagen and Aarhus to phase out Microsoft services and promote open-source solutions

▶️ Software Development

Enhancing Coding Efficiency with Feedback-Driven LLM Agents and Tools

Agents are feedback-driven LLMs with tools like bash, patch, and web eval, enabling automated code editing, testing, and navigation, significantly improving programming efficiency despite higher time and cost.

  • Defines an agent as a 9-line code loop containing an LLM call enabling command execution without human input (source)
  • Agents utilize environmental feedback tools: bash(cmd), patch(hunks), todo(tasks), web_nav(url), web_eval(script), web_logs(), web_screenshot(), keyword_search(keywords), codereview()
  • Benefits include improved API use, reduced syntax errors, better dependency management, enhanced testing, handling larger codebases, and end-to-end code validation; drawbacks are increased time and cost (e.g., $1.15 per API commit)

▶️ Management and Leadership

Microsoft Successfully Migrates Office Source Control to Git with VFS and Parallel Sync

Microsoft’s Office source control migrated from Source Depot to Git over years, utilizing VFS for Git, parallel syncing, rigorous testing, and extensive communication to ensure seamless transition for 4,000+ engineers.

  • Microsoft migrated Office source control from Source Depot to Git over several years, involving complex planning and execution.
  • The migration required developing Virtual File System for Git (VFS for Git) to handle a 200 GB repository, enabling efficient cloning and operations.
  • The process included creating a parallel universe for syncing Source Depot with Git, proving equivalence through extensive testing, and implementing a rollback strategy.

Phil McKinney on HP’s $1.2B Palm Deal and Rapid WebOS Collapse

Phil McKinney recounts how HP’s leadership, under flawed decision frameworks and leadership mismatches, bought Palm for $1.2B based on solid WebOS technology, then rapidly destroyed it in 49 days, illustrating systemic errors in innovation decision-making.

  • Phil McKinney, former HP CTO, led the technical due diligence and presented the case for HP’s $1.2 billion acquisition of Palm in April 2010, believing WebOS was a breakthrough platform.
  • HP acquired Palm in July 2010; within 49 days, CEO Leo Apotheker announced the immediate discontinuation of all WebOS devices, including the TouchPad, after poor sales of 25,000 units out of 270,000 shipped.
  • McKinney was confined to bed during the decision to kill WebOS, witnessing from afar the rapid dismantling of the platform, which he attributes to systematic thinking errors and leadership misjudgments.

BT CEO Foresees AI-Led Job Cuts and Openreach Spin-Off Potential

BT CEO Allison Kirkby suggests AI could enable deeper job cuts and possibly spin off Openreach if its network value isn’t reflected in BT’s share price, amid ongoing cost reductions and network upgrades.

  • Allison Kirkby, CEO of BT, indicates AI advancements could enable further job cuts beyond the planned 40,000 by 2030 and reduce costs by £3bn.
  • BT has doubled down on its 2023 plan to cut up to 55,000 jobs, refocused on the UK market, and sold its Italian and Irish divisions.
  • The company is considering spinning off Openreach if its market value does not reflect its network value; Openreach aims to reach 30 million homes by 2030, with a current valuation of about £30bn versus BT’s £18.5bn market cap.

Tailscale: Frequent Reauthentication Hinders Security and User Experience

Tailscale argues that frequent reauth does not enhance security but causes user frustration; instead, continuous verification and real-time security checks provide stronger, less intrusive protection.

  • Frequent reauthentication disrupts user workflow and can weaken security posture by increasing MFA fatigue and phishing risks.
  • Modern OS screen locks and device possession checks effectively verify user presence, reducing the need for repeated logins.
  • Web session expiry times (7-30 days) are ineffective for security and cause unnecessary user frustration; continuous verification methods are preferred.
  • Security should rely on real-time device posture checks, instant revocation upon security events, and adaptive policies rather than frequent login prompts.

Amazon Grocery Restructuring Under New Leader Jason Buechel

Jason Buechel, promoted in January 2025 to oversee Amazon’s grocery sector, formed a leadership team to streamline operations and integrate Whole Foods more tightly, aiming to accelerate growth and improve efficiency.

  • Jason Buechel assembled a new leadership team to reorganize Amazon’s grocery business, including Whole Foods.
  • Buechel was promoted to oversee Amazon’s entire grocery operations earlier in 2025.
  • Whole Foods corporate staff is being integrated under Amazon’s main employee programs as part of the restructuring.

Tech Leaders and Founders Dismiss College Over Cost and Ideological Concerns

Big Tech leaders and startup founders are increasingly dismissing college degrees, citing high costs and ideological issues, with some opting for alternative paths like gap years or direct industry entry.

  • Tech leaders increasingly oppose higher education, citing high costs, inefficiency, ideological bias, and perceived limited value of college degrees.
  • Palantir CEO Alex Karp and venture investors Peter Thiel and Joe Lonsdale criticize colleges as disconnected from real-world applications and stifling free thought.
  • Some young founders and high school dropouts are skipping college entirely or taking gap years, influenced by industry skepticism and entrepreneurial opportunities.

Cannes Lions 2025 Highlights Industry Shakeup and Retail Media Rise

Cannes Lions 2025 emphasizes industry upheaval, including agency mergers, leadership changes, retail media growth, and AI-driven marketing shifts, reflecting evolving client needs and media consumption.

  • Cannes Lions 2025 will focus on industry shifts like agency consolidation, high-profile executive departures, and technological changes.
  • The festival highlights include retail media’s rise, with Albertsons showcasing a grocery store activation, and the introduction of retail media categories in awards.
  • Major agency mergers are pending (Omnicom and IPG), with notable leadership exits (WPP CEO Mark Read, David Droga), and industry share declining from 40% to 30% since 2015.

BT CEO Warns AI May Lead to More Job Cuts Beyond 2030 Targets

BT CEO Allison Kirkby indicated AI advancements may cause further workforce reductions beyond the 55,000 jobs targeted by 2030, leveraging AI to optimize customer service and operational efficiency.

  • BT CEO Allison Kirkby warned AI could lead to additional job cuts beyond the 55,000 planned by 2030
  • BT’s 2023 plan aimed to reduce costs, but Kirkby stated AI’s full potential might enable even smaller workforce sizes
  • BT has implemented generative AI for customer service and sales, with EE’s virtual assistant “Aimee” handling up to 60,000 weekly conversations

Microsoft Develops Copilot AI for DoD Launching Summer 2025

Microsoft is preparing Microsoft 365 Copilot for the Department of Defense, targeting summer 2025 release after security and compliance checks, with potential adoption by the Pentagon’s 2.87 million personnel.

  • Microsoft is developing a version of its Copilot AI tool for the Department of Defense, with availability for DoD environments expected no earlier than summer 2025.
  • Microsoft 365 Copilot for DoD will undergo security and compliance validation before release.
  • A customer with over 1 million Microsoft 365 licenses, likely the Pentagon with 2.1 million military and 770,000 civilian personnel, is expected to adopt Copilot, representing a major enterprise deployment.

Ex-Apple Director Urges New Employees to Observe Culture Before Acting

Bob Baxley recommends slowing down and deeply observing company culture when starting a new role, to avoid carrying over old behaviors from powerful tech environments like Apple and Pinterest.

  • Ex-Apple director Bob Baxley advises taking time to recalibrate and observe company culture when starting a new job
  • Baxley emphasizes the importance of slowing down, deeply observing, and avoiding immediate behavioral carryover from previous workplaces
  • Major tech companies, including Apple and Pinterest, have “powerful cultures” that influence employee behavior and are difficult to transition between without adjustment

YouTube Surpasses Netflix in Viewership and Boosts Scripted Content Competition

YouTube is expanding scripted content and episodic series, attracting creators and advertisers, with creator earnings projected at $185 billion in 2025, challenging Hollywood’s traditional model.

  • YouTube surpassed Netflix in TV viewership and is developing scripted shows like “Shanked” to compete with traditional TV.
  • Top creators such as Dhar Mann and Alan Chikin Chow produce TV-like episodic series; YouTube is promoting tools and matching advertisers to support quality content.
  • Advertisers are shifting spend, with WPP estimating creators will earn $185 billion in 2025, surpassing traditional TV ad revenue; Unilever plans to allocate 50% of ad spend to social media and influencers.

Over 700 Marines Deployed to Support Federal Forces During LA Protests

Over 700 Marines have been mobilized to support federal forces during LA protests, augmenting National Guard presence amid unprecedented military deployment within the US, with roles and rules of engagement still being defined.

  • Over 700 Marines from 2nd Battalion, 7th Marines, 1st Marine Division, were mobilized at the Marine Corps Air Ground Combat Center in California to respond to Los Angeles protests.
  • The Marines will join approximately 1,700 National Guard members already deployed, with an additional 2,000 National Guard troops ordered by Trump, though timing is unclear.
  • The deployment marks a significant escalation in military involvement in domestic protests, with tasks potentially including crowd control and perimeter security; use-of-force guidelines are under finalization.

Bill Atkinson’s Impact on Apple and Multimedia Innovation

Bill Atkinson joined Apple in April 1978, contributed to Macintosh graphics and interface development, and created HyperCard, influencing personal computing and multimedia before leaving in 1990.

  • Bill Atkinson joined Apple Computer on April 27, 1978, leaving his neuroscience PhD at UC San Diego.
  • He developed key software components including QuickDraw, Lisa Window Manager, and MacPaint, contributing to the Macintosh’s graphical interface.
  • Atkinson designed HyperCard in 1985, enabling non-programmers to create interactive media, published in 1987, predating Mosaic web browser.

Boosting Productivity with Receipt Printer and Tangible Feedback System

Using a receipt printer to automate daily task printing, combined with micro-task breakdowns and tangible feedback, significantly enhanced productivity and consistency for managing routines.

  • The author improved productivity by applying video game feedback loop principles to task management.
  • Implemented a system using sticky notes, crumpling them into a jar for tangible feedback.
  • Switched to a thermal receipt printer (e.g., Epson TM-T20III, ~$150) to automate task printing, increasing consistency and reducing procrastination.

UK Government to Revamp Digital Spending in 2025 Review

UK government’s 2025 spending review aims to overhaul digital investment strategies, emphasizing zero-based budgeting and digital-first policies amid concerns over legacy system costs and delayed digital projects.

  • UK government’s upcoming three-year spending review will be announced by the Chancellor of the Exchequer on June 11, 2025, including capital budgets for four years.
  • The review emphasizes reforming departmental technology investments, with a focus on a “digital-first approach” and a Zero-Based Review (ZBR) to justify budgets from the ground up.
  • Past digital investments have been underfunded, leading to increased remediation costs, legacy system issues, and potential revenue losses, exemplified by HMRC’s delays and cost overruns.

NHS Professionals Breach Exposes Active Directory Data in May 2024

NHS Professionals experienced a cyberattack in May 2024, resulting in the theft of its Active Directory database; remediation efforts included password resets and disabling drive mapping, but full security hardening was ongoing.

  • Attackers stole NHS Professionals’ Active Directory database (ntds.dit) in May 2024 via a compromised Citrix account, with the breach detected on May 15, 2024
  • The breach involved privilege escalation to domain admin, lateral movement via RDP and SMB, and exfiltration of the AD database through a ZIP archive copied to a physical drive
  • Deloitte’s incident response report recommended extensive security improvements, including multi-factor authentication, endpoint detection and response deployment, enhanced logging, and network hardening

Denmark Replaces Microsoft Office with LibreOffice to Boost Digital Sovereignty

Denmark’s Ministry for Digitalization intends to phase out Microsoft software, starting with LibreOffice, as part of a broader European push for digital sovereignty and self-hosted open-source alternatives.

  • Denmark’s Ministry for Digitalization plans to replace Microsoft Office 365 with LibreOffice for half of its staff, including the minister, aiming for full replacement by year-end.
  • The move follows similar actions by Copenhagen and Aarhus city governments, driven by digital sovereignty concerns amid geopolitical tensions and US tech dependence.
  • The transition emphasizes open-source solutions like Collabora’s CODE, enabling self-hosted LibreOffice deployments, and aligns with broader European efforts to reduce reliance on US-based technology providers.

UK Universities Sign £9.86M Oracle Java License Deal Amid Cost Rise

UK higher education institutions agreed to pay up to £9.86 million for Oracle Java SE subscriptions, including historic fee waivers, amid licensing cost hikes and audit pressures since Oracle’s 2023 licensing model change.

  • UK universities and colleges signed a framework worth up to £9.86 million ($13.33 million) with Oracle for Java SE Universal Subscription licenses, including a waiver of historic fees since 2023
  • The Java SE Universal Subscription, introduced in January 2023, shifted licensing from per-user to per-employee basis, causing significant cost increases—up to five times more for large organizations
  • Licensing experts and Gartner analysis indicate the new model can be two to five times more expensive than legacy licensing; Oracle’s aggressive sales and audit strategies are increasing licensing concerns among users

Larry Ellison’s Net Worth Hits $242 Billion After Oracle’s Q4 Success

Larry Ellison’s net worth rose to $242 billion after Oracle’s Q4 2025 earnings, boosting him to No. 2 on Forbes’s billionaire list, driven by a 14% share increase amid high cloud demand.

  • Larry Ellison’s net worth surged approximately $25 billion to $242 billion on June 12, 2025, after Oracle’s Q4 results
  • Shares increased 14% due to better-than-expected fiscal Q4 results ending May 31, and Ellison highlighted “astronomical” cloud demand
  • Ellison’s fortune places him second on Forbes’s real-time billionaire list, behind Elon Musk, with a net worth of $406 billion

Mattel and OpenAI to Embed AI in Toys Amid Privacy Concerns

Mattel and OpenAI’s 2025 partnership will embed AI into toys and products, focusing on youth over 13, with unspecified data use, raising privacy and security considerations.

  • Mattel signed a deal with OpenAI to incorporate AI into toys and products later in 2025, targeting children over 13.
  • The partnership aims to develop smarter toys, potentially including internet-connected physical toys with ChatGPT integration.
  • Past issues with Mattel’s Wi-Fi-enabled Hello Barbie, which faced security and privacy concerns, highlight the need for careful data handling in new AI toys.

Meta Offers Up to $10 Million Annually to Top AI Researchers in Bid for Superintelligent AI

Meta is offering elite AI researchers up to $10 million per year, with Zuckerberg personally recruiting, to build a superintelligent AI team amid fierce competition and talent scarcity.

  • Meta offered at least $10 million annually to a top AI researcher, with CEO Mark Zuckerberg personally reaching out.
  • The offer did not specify job details; the researcher declined to be identified.
  • Meta’s aggressive talent acquisition aims to develop superintelligent AI amid intense industry competition and talent shortages.

User’s Wireless Laptop Outrage Highlights Need for Clear Device Communication

A user demanded a wireless laptop in June 2025, became angry when its battery died, illustrating misconceptions about wireless devices and the need for proper user education.

  • A user requested a wireless laptop during a device upgrade at a PR firm in June 2025
  • The user was outraged when the laptop’s battery died, believing it was “wireless” and did not require power
  • The incident highlights user misconceptions about wireless technology and the importance of clear communication about device capabilities

US Army Launches Detachment 201 with Tech Leaders to Drive AI Military Modernization

The US Army established Detachment 201, recruiting senior executives from Palantir, Meta, and OpenAI as lieutenant colonels to accelerate AI-driven military modernization efforts.

  • US Army creates Detachment 201: Executive Innovation Corps, recruiting senior tech executives from Palantir, Meta, OpenAI as lieutenant colonels in the Army Reserve.
  • Recruits include Palantir CTO Shyam Sankar, Meta CTO Andrew Bosworth, OpenAI Chief Product Officer Kevin Weil, and ex-OpenAI CRO Bob McGrew.
  • Members will serve at least 120 hours annually, work remotely, and are exempt from basic training, advising on AI integration and military modernization projects.

Meta emplea “localhost tracking” para evadir GDPR y puede enfrentar multas de €32 mil millones

Meta utilizó “localhost tracking” para vincular actividad web con identidades reales, evadiendo protecciones legales, lo que podría generar multas combinadas de aproximadamente €32 mil millones.

  • Meta implementó “localhost tracking” para identificar usuarios en móviles, incluso con VPN, modo incógnito y cookies eliminadas.
  • La técnica combina datos del app en segundo plano y scripts de seguimiento en el navegador, usando WebRTC y SDP Munging.
  • La recopilación ilícita puede costar a Meta hasta €32 mil millones en multas por violaciones de GDPR, DSA y DMA, por uso no consentido de datos personales.

▶️ Technology

Android 16 Launches with Live Notifications, Enhanced Security and Tablet Windowing

Android 16, released on June 10, 2025, introduces real-time notifications, hearing aid enhancements, Advanced Protection security, and desktop windowing for tablets, expanding device capabilities.

  • Android 16 launched on June 10, 2025, initially available on supported Pixel devices with broader device support later this year.
  • Introduces live notifications with real-time updates from compatible apps like food delivery and ride-share, along with grouped notifications to reduce clutter.
  • Enhances hearing aid support by allowing users to switch to their phone’s microphone for clearer calls and provides native control for hearing devices.
  • Adds stronger security features through Google’s Advanced Protection, protecting against online threats, harmful apps, and scam calls.
  • Implements desktop windowing on tablets, enabling multiple resizable app windows for improved multitasking, with features like custom keyboard shortcuts and taskbar overflow, rolling out later in 2025.

macOS 26 Tahoe Introduces High-Speed ASIF Disk Image Format

macOS 26 Tahoe’s ASIF format offers significantly faster disk image performance, suitable for VMs and potentially Time Machine backups, with creation via diskutil and compatibility considerations.

  • macOS 26 Tahoe introduces ASIF, a new disk image format capable of near-native speeds, documented here.
  • ASIF disk images are flagged as com.apple.disk-image-sparse, are sparse files in APFS, and can be created via Disk Utility or diskutil with --format ASIF.
  • Performance tests on a MacBook Pro M3 Pro show read/write speeds of 5.8/6.6 GB/s unencrypted and 4.8/4.6 GB/s encrypted; similar speeds observed on a Mac mini M4 Pro.

miniDiffusion: PyTorch Reimplementation of Stable Diffusion 3.5 for Education

miniDiffusion reimplements Stable Diffusion 3.5 in PyTorch with minimal dependencies, including core models, text encoders, tokenizers, and diffusion components, for educational and experimental purposes.

  • miniDiffusion is a reimplementation of Stable Diffusion 3.5 in pure PyTorch, designed for educational and experimental use, with approximately 2800 lines of code.
  • Core files include dit.py (main model), dit_components.py (embedding, normalization, patch embedding), attention.py (Joint Attention), noise.py (Euler Scheduler), t5_encoder.py, clip.py (text encoders), tokenizer.py (tokenizers), and metrics.py (FID).
  • The repository contains modules for VAE, CLIP, T5 encoders, Byte-Pair & Unigram tokenizers, Multi-Modal Diffusion Transformer, Euler Scheduler, Logit-Normal Sampling, and training/inference scripts.

Mistral AI Launches Magistral: A Multilingual Reasoning Model for Enterprise

Mistral AI launched Magistral, a dual-variant reasoning model (Small and Medium) with 24B parameters, achieving up to 90% accuracy on AIME2024, emphasizing transparent, multilingual, and fast multi-step reasoning for diverse professional applications.

  • Mistral AI announced Magistral on June 10, 2025, as its first reasoning model focused on domain-specific, transparent, and multilingual reasoning.
  • Released in two variants: Magistral Small (24B parameters, open-source) and Magistral Medium (enterprise version), scoring 73.6% and 90% on AIME2024 respectively.
  • Features native multi-language chain-of-thought reasoning, 10x faster responses with Le Chat’s Flash Answers, and supports applications in enterprise, regulated industries, software engineering, and creative content.

Apple Unveils Liquid Glass for Spatial Ambient Interfaces at WWDC 2025

Apple’s WWDC 2025 debut of Liquid Glass signifies a paradigm shift toward spatial, ambient interfaces, inspired by visionOS, emphasizing translucency, environmental awareness, and integrated hardware-software design.

  • Apple introduced Liquid Glass at WWDC 2025 as a strategic shift toward ambient, spatial interfaces inspired by visionOS, emphasizing translucency, layering, and environmental responsiveness.
  • The design pattern follows Apple’s historical paradigm: major visual changes precede fundamental interaction shifts, moving from skeuomorphic to flat design, now toward less screen relevance.
  • Liquid Glass leverages tight hardware-software integration, showcasing GPU-intensive real-time effects like dynamic transparency and lighting, reinforcing Apple’s ecosystem advantage and influencing industry design standards.

GPU-Based Real-Time Glyph Rendering with Curves and Efficient Atlas Packing

Rubén Osorio López introduces a GPU-driven glyph rendering pipeline that rasterizes font curves at runtime, enabling high-quality, resolution-independent text with subpixel anti-aliasing and efficient atlas management.

  • Presents a GPU-based method for real-time glyph rendering using curves, avoiding texture baking and atlas size limitations
  • Converts font curves (lines, quadratic, cubic beziers) into quadratic beziers, then rasterizes at runtime with up to 512 samples per glyph
  • Utilizes Z-Order (Morton codes) for efficient atlas packing, supports transposed packing for long/thin glyphs, and employs temporal accumulation for high-quality anti-aliasing

Rapid LLM Growth: 30+ Models, New Benchmarks, and Innovative Evaluation

The last six months saw rapid LLM development with over 30 models, emphasizing local inference, innovative evaluation, and tool integration, notably Meta’s Llama 3.3 70B, DeepSeek v3, and GPT-4.1 Mini, alongside new benchmarks and tool-based model assessments.

  • Over 30 significant LLM models released in the last six months, including Meta’s Llama 3.3 70B, Amazon Nova, DeepSeek v3 685B, and Google Gemini 2.5 Pro.
  • Notable advancements include Llama 3.3 70B capable of running on a 64GB RAM laptop, DeepSeek v3 685B with an estimated training cost of $5.5 million, and GPT-4.1 Mini offering high performance at low cost.
  • Evaluation methods shifted from benchmarks and leaderboards to custom, LLM-driven comparisons, including a side-by-side SVG pelican on bicycle test and Elo ranking system for model performance.

Spark: A High-Performance 3D Gaussian Splatting Renderer for THREE.js

Spark is a high-performance 3D Gaussian Splatting renderer for THREE.js, enabling fast, integrated rendering with support for multiple formats and dynamic effects.

  • Spark is an advanced 3D Gaussian Splatting renderer for THREE.js
  • Supports integration with other meshes and splats, fast rendering across all devices, programmable dynamic splat effects, and wide format support (ply, spz, splat, ksplat)
  • Documentation available at Getting Started and other sections

New York RAISE Act Sets Safety Rules for Large AI Labs to Prevent Disasters

New York’s RAISE Act, pending governor approval, establishes transparency standards for large AI labs to mitigate risks of catastrophic AI-related disasters involving models trained with over $100 million in resources.

  • New York state lawmakers passed the RAISE Act on June 12, 2025, targeting frontier AI models from OpenAI, Google, and Anthropic.
  • The bill aims to prevent AI-related disasters involving over 100 deaths or injuries, or damages exceeding $1 billion.
  • If signed into law, it mandates AI labs to publish safety/security reports, report incidents, and allows civil penalties up to $30 million for non-compliance.

Sam Altman Cuts O3 Price by 80% to Boost Accessibility

Sam Altman announced an 80% price drop for O3 on June 10, 2025, aiming to enhance accessibility and usage, while O3-pro pricing maintains performance advantages.

  • Sam Altman announced an 80% price reduction for O3 on June 10, 2025
  • O3-pro pricing remains unchanged, emphasizing performance benefits
  • The post received 3.4 million views, 1.7K replies, and 2K likes

Anthropic’s Multi-Agent System Boosts Research Performance by 90%

Anthropic developed a multi-agent research system with a lead agent coordinating parallel subagents, enabling dynamic, scalable exploration of complex topics, improving performance by 90.2%, and managing over 200,000 tokens.

  • Anthropic’s Research feature employs multi-agent architecture with a lead agent and specialized subagents to perform complex web and tool searches.
  • System outperforms single-agent setups by 90.2% on internal evaluations, especially for breadth-first queries involving multiple independent directions.
  • The architecture uses dynamic, multi-step search with iterative planning, external memory, parallel tool calls, and citation processing, handling over 200,000 tokens in context.

Apple Unveils Liquid Glass Design for All Platforms Enhancing Visuals and Content Focus

Apple introduced a universal Liquid Glass-based design across platforms (iOS 26, iPadOS 26, macOS Tahoe 26, watchOS 26, tvOS 26) that enhances content focus and visual vitality through dynamic, translucent elements.

  • Apple previewed a new software design featuring Liquid Glass material across iOS 26, iPadOS 26, macOS Tahoe 26, watchOS 26, and tvOS 26.
  • Liquid Glass is translucent, reacts dynamically to content and environment, with real-time rendering, specular highlights, and adaptive color.
  • The design updates include redesigned controls, toolbars, tab bars, sidebars, and system elements like Lock Screen, Home Screen, notifications, and system apps, enhancing harmony and personalization.

Apple Unveils iPadOS 26, visionOS 26 and New CarPlay at WWDC 2025

Apple unveiled iPadOS 26, visionOS 26, and a redesigned CarPlay at WWDC 2025, with Siri AI upgrades delayed to spring 2026 and increased OpenAI collaboration.

  • Apple announced iPadOS 26 with enhanced multitasking features, visionOS 26, and a new CarPlay interface at WWDC 2025
  • iPadOS 26 introduces improved multitasking capabilities; visionOS 26 focuses on spatial computing; CarPlay receives a redesigned interface
  • Delayed Siri AI upgrades are now scheduled for spring 2026; Apple’s AI leadership remains low-profile at WWDC; increased collaboration with OpenAI

Meta invests $14.3B in Scale AI to lead its Superintelligence lab and compete in AI

Meta invested $14.3 billion in Scale AI to bolster its AI division, appointing CEO Alexandr Wang to lead its Superintelligence lab, aiming to compete with industry AI rivals.

  • Meta invested $14.3 billion in Scale AI, its largest outside investment, representing about 10% of its 2024 revenue.
  • The deal includes appointing Scale AI CEO Alexandr Wang, 28, to a top leadership role in Meta’s new Superintelligence lab.
  • This is Meta’s second-largest deal after the $19 billion WhatsApp acquisition, as it aims to compete with AI leaders like Google, Microsoft, OpenAI, and Anthropic.

LLM Inference Costs Drop Below Search API Prices

LLMs are substantially cheaper to operate than many believe, with inference costs often below search API prices, due to rapid cost reductions and efficient models, challenging assumptions of high expense.

  • Inference costs for LLMs have decreased faster than model improvements, making them significantly cheaper than commonly perceived.
  • Public API prices for search engines range from $5 to $35 per 1,000 queries; comparable LLM inference costs are often an order of magnitude lower.
  • Estimated token output per query ranges from 361 to 1,145 tokens, with models priced between $0.20 and $8.00 per million tokens, indicating low per-query inference costs.

Cybercriminals Use Fake AI Sites to Spread Malware and Steal Data Worldwide

Cybercriminals used fake AI model sites and malicious ads to infect devices with “BrowserVenom” malware, enabling traffic interception and data theft via a fake DeepSeek installer.

  • Cybercriminals created a fake installer for Chinese AI model DeepSeek-R1, loaded with malware called “BrowserVenom”
  • BrowserVenom redirects browser traffic through attacker-controlled servers, enabling data theft, monitoring, and plaintext traffic exposure
  • The malware has infected multiple computers across Brazil, Cuba, Mexico, India, Nepal, South Africa, and Egypt; campaign detected via phishing and malicious ads promoting fake AI sites
  • The campaign used the URL https://deepseek-platform.com, promoted via Google ads, which were later suspended by Google
  • Infection process involves fake CAPTCHA prompts, leading to download of malicious installer AI_Launcher_1.21.exe from https://r1deepseek-ai.com/gg/cc/
  • The malware checks for admin privileges, installs a hardcoded certificate for persistent access, and adds a proxy server to monitor traffic
  • The campaign code contains Russian comments; no attribution to specific cybercrime group
  • The malware mimics legitimate CAPTCHA and download pages, tricking users into executing the infection