Today’s headlines include a major setback for Nextcloud’s Android app due to Google’s permission revocation, impacting millions of users, and the announcement of EuroHPC’s upgrade of the Leonardo supercomputer with NVIDIA H100 GPUs for advanced AI workloads. Additionally, Intel aims for a 2027 break-even in its foundry business and Salesforce envisions open source LLMs transforming AI markets.
▶️ Internet Infrastructure
Nextcloud Android App Limited by Google Permission Revocation
Nextcloud’s Android app cannot upload all files due to Google’s permission revocation in September 2024, citing security concerns, impacting millions and exemplifying Big Tech gatekeeping practices.
- Nextcloud Android app’s file upload feature has been limited since Google revoked a critical permission in September 2024, restricting uploads to photos and videos only.
- Google cites security concerns for permission revocation; Nextcloud has used this feature since 2016 without reported issues, and Google’s own apps still have it.
- Despite multiple appeals, Google refuses to reinstate full file upload permissions, forcing Nextcloud to limit functionality for approximately one million users on Google Play.
EuroHPC to Upgrade Leonardo Supercomputer with 166 NVIDIA H100 GPU Nodes
EuroHPC’s €28.2 million contract with Eviden will upgrade Leonardo with 166 NVIDIA H100 GPU nodes for AI workloads, supporting LLMs and multi-modal AI, deployment scheduled for late 2025.
- EuroHPC signed a €28.2 million ($31.4 million) contract with Eviden to upgrade the Leonardo supercomputer with an AI-focused architecture, called LISA.
- The LISA system will include 166 server nodes, each with 8 GPU accelerators, interconnected via a high-performance non-blocking network and equipped with high-bandwidth memory (HBM).
- The specifications remain undisclosed; the contract price suggests a configuration with NVIDIA H100 GPUs, contrasting earlier estimates of older Nvidia A100 systems.
Intel Targets 2027 Break-Even for Foundry Business Through External Customers
Intel seeks external foundry customers for its 14A process node to achieve 2027 break-even, relying on high NA EUV tools and partnerships, amid workforce reductions and strategic organizational changes.
- Intel aims for its foundry division to break even by 2027, targeting external customers for its 14A process node.
- The 14A node, utilizing high NA EUV tools from ASML, is expected to require external volume to be financially viable.
- Intel’s 18A process will primarily serve internal products, with first SKU (Panther Lake) expected by end of 2025; 18A’s internal volume reduces reliance on external foundry business.
- CFO David Zinsner states that 14A will be more expensive, but expects external volume to increase, driven by partnerships with UMC and Tower Semiconductor.
- Intel procured at least two high NA EUV tools from ASML, with the process expected to be used mainly for high-performance and AI chips.
- The company anticipates low to mid-single digit billions of dollars in external revenue for the foundry, with a mix of older and newer process generations.
- The global supply chain issues have increased customer interest in second sourcing, positioning Intel Foundry as a potential alternative.
- Intel plans to shed up to 20% of its workforce (~20,000 employees) to improve execution and organizational efficiency under CEO Lip-Bu Tan.
Intel Xeon 6 Unlocks Up to 128 Cores for AI and Data Center Efficiency
Intel’s Xeon 6 processors, introduced in April 2024, feature dual microarchitectures optimized for AI, HPC, and edge workloads, offering up to 128 cores, AI acceleration, and significant data center efficiency improvements.
- Intel unveiled revamped Xeon 6 processors at April 2024 Intel Vision event, featuring two microarchitectures: P-core for compute-intensive workloads and E-core for high-density, low-power environments
- The Xeon 6 family includes CPUs with up to 128 cores, integrated AI acceleration via Intel AMX, support for MRDIMM, CXL enhancements, and a range of four series for diverse workloads
- Deployment of P-core and E-core CPUs enables workload-specific optimization, with server consolidation reducing server count by up to 5:1, cutting power, TCO, and carbon emissions significantly
Ivanti Patches Zero-Day Flaws Exploited in the Wild Amid Critical ASD Warning
Ivanti patched two actively exploited zero-day vulnerabilities (CVE-2025-4427 and CVE-2025-4428) in EPMM, linked to unnamed open source libraries, following a critical ASD warning.
- Ivanti released patches for two zero-day vulnerabilities (CVE-2025-4427 and CVE-2025-4428) linked to open source libraries, actively exploited in the wild
- The vulnerabilities affect Ivanti Endpoint Manager Mobile (EPMM) versions 11.12.0.4 and earlier, 12.3.0.1 and earlier, 12.4.0.1 and earlier, 12.5.0.0 and earlier
- Australian Signals Directorate issued a critical warning, noting exploitation of these CVEs for remote code execution, with ongoing investigation into the open source library CVEs
Ransomware Targets Critical In-Between Systems Threatening Infrastructure Safety
Ransomware groups target in-between IT and OT systems in critical infrastructure, risking safety and operational integrity; attacks are evolving to manipulate systems causing long-term damage.
- Ransomware attackers increasingly target systems between IT and operational technology (OT), known as no man’s land, affecting critical infrastructure.
- In-between systems, such as those in fuel storage or industrial control, are vulnerable, with attacks risking product integrity and safety.
- Criminal focus has shifted toward these middle systems, with malware variants targeting ICS increasing from 7 to 9 in recent years, and attacks on critical infrastructure rising since 2022.
Red Hat Leaks RHEL 10 with May 2025 Release and New Hardware Requirements
Red Hat quietly leaked RHEL 10, with GA date May 13, 2025, featuring x86-64-v3 requirements, no LibreOffice or X.org, and GNOME on Wayland only, ahead of the official announcement at Red Hat Summit.
- RHEL 10 leaked ahead of Red Hat Summit, appearing on the Japanese site with a GA release date of May 13, 2025
- The release code name is Coughlan; details briefly visible on the English site before vanishing
- RHEL 10 moves to x86-64-v3 baseline, requiring Intel Haswell or newer, AMD Excavator or later, with AVX2 and FMA support; 32-bit x86 support removed
▶️ Open Source
Continuous Thought Machine Enables Time-Based Reasoning and Outperforms Traditional Models
The CTM uses neural synchronization over a decoupled internal dimension to enable time-based reasoning, achieving strong performance and interpretability across tasks like maze navigation, ImageNet classification, and parity prediction.
- The Continuous Thought Machine (CTM) is a neural network architecture that models neural activity synchronization over time as its core representation.
- It incorporates internal recurrence, neuron-level models with private weights, and neural synchronization to enable dynamic, biologically-inspired reasoning.
- The CTM demonstrates capabilities in maze solving, image classification, parity prediction, and memory tasks, often outperforming traditional models like LSTMs, with good calibration and interpretability.
Community-Led CoMaps Fork of Organic Maps Gains Momentum
A community-led fork of Organic Maps, named CoMaps, is under active development with principles of transparency, open source, and privacy; community voting on the final name ends May 20, with ongoing contributions on Codeberg.
- Community-led fork of Organic Maps named CoMaps is progressing rapidly, emphasizing transparency, community decision-making, not-for-profit, open source, and privacy focus
- Project’s provisional name is CoMaps; community voting on the final name is open until May 20th via Gitberg
- Development efforts are ongoing on Codeberg, including feature development, documentation, governance, website building, and fundraising through OpenCollective
- No significant progress in negotiations with Organic Maps shareholders; Viktor seeks guarantees against sale and full control, while shareholder disagreements remain unresolved
Salesforce CEO Predicts Open Source LLMs Will Transform AI Market and Cut Costs
Salesforce’s Benioff forecasts open source LLMs will dominate, enabling cost reductions and innovation, challenging proprietary models, with DeepSeek’s MIT-licensed R1 model exemplifying this shift.
- Salesforce CEO Marc Benioff predicts open source LLMs will commoditize the AI market, reducing reliance on proprietary models like OpenAI, Google Gemini, and Meta’s Llama 2
- Benioff highlights China’s DeepSeek R1 model, licensed under MIT, as a cost-effective alternative enabling embedding in products at significantly lower costs, potentially reducing Salesforce’s model costs by 90%
- He emphasizes open source AI’s role in innovation, citing Salesforce’s contributions and the transformation from transformer to MOE (mixture of experts) models to achieve cost savings
- Benioff criticizes Microsoft’s strategy, claiming they are merely reselling ChatGPT and are developing their own models under the Prometheus program, moving away from open source approaches
- The article notes that open source models like DeepSeek are smaller, suitable for edge devices, and that Gartner predicts a decline of large LLM providers, with open models driving market evolution
▶️ Software Development
Transformative Resources That Reshaped My Understanding of Compiler Design
The author highlights key programming language and compiler resources that profoundly changed their understanding, covering garbage collection, optimizer design, formal verification, and compiler construction techniques.
- The article shares influential programming language and compiler resources that significantly altered the author’s understanding, including topics like garbage collection, instruction rewriting, abstract domains, register allocation, regex engines, neural networks, SSA form, and compiler construction.
- Notable references include Andy Wingo’s semi-space collector, CF Bolz-Tereick’s optimizer techniques and Z3 as a proof engine, Russ Cox’s regex virtual machine, Andrej Karpathy’s micrograd, and Abdulaziz Ghuloum’s single-pass compiler approach.
- The author emphasizes how these resources changed their perspectives on topics such as union-find, code verification, optimizer pass ordering, and IR storage, with links to each resource for further exploration.
Master Rust: Embrace Its Concepts and Use Clippy for Better Coding
The article advocates embracing Rust’s mental model, practicing with small code snippets, leveraging compiler guidance, and building intuition through consistent, accurate coding to flatten the learning curve.
- Published on May 5, 2025, by Matthias Endler on corrode.dev
- Emphasizes adopting a mindset shift to accept Rust’s unique concepts like lifetimes, ownership, and traits
- Recommends using all Clippy lints from day one, writing small code snippets in the playground, and focusing on type-driven development
▶️ Management and Leadership
The Rise and Fall of Bell Labs: Autonomy, Innovation, and Cultural Shifts
Bell Labs thrived due to researcher autonomy, selective talent, and long-term focus; its decline correlates with corporate and cultural shifts toward short-term metrics and reduced scientific freedom.
- Bell Labs was founded by Alexander Graham Bell with a management style that emphasized researcher autonomy and shared credit, starting in the early 20th century.
- During WWII, Bell Labs rapidly reverse-engineered the British Magnetron, developed anti-aircraft technology, and invented key innovations like pulse code modulation and the first anti-aircraft missile.
- Bell Labs declined after AT&T’s breakup, as the culture of radical freedom was replaced by short-term shareholder pressures, with modern tech giants investing more in research but lacking similar institutional independence.
pkill -H flag improves process safety by targeting only signals with handlers
pkill --require-handler (-H) prevents signals from being sent to processes without handlers by inspecting /proc/[pid]/status, reducing outages caused by unhandled signals like SIGHUP.
- The
pkill --require-handler(-H) flag ensures signals are only sent to processes with registered handlers, preventing unintended termination. - Default signal dispositions, such as for
SIGHUP,SIGUSR1, andSIGUSR2, often terminate processes if no handler exists, causing production outages. pkill -Hreads/proc/[pid]/statusand checks theSigCgtbitmap to verify if a process has a handler for a specific signal before sending it.
Sony Forecasts Lower Profit Amid US Tariffs and Plans Spinoff and Buyback
Sony projects ¥1.28 trillion operating profit for FY2026 amid US tariffs, plans ¥250 billion buyback, and a financial unit spinoff, while PS5 sales target 15 million units.
- Sony Group Corp. forecasts FY operating profit of ¥1.28 trillion ($9.4 billion), below the analyst estimate of ¥1.5 trillion, citing a ¥100 billion US tariff impact.
- The company plans to sell 15 million PlayStation 5 consoles in FY2026, down from 18.5 million in the previous year, with potential price increases to offset tariffs.
- Sony announced a share buyback of up to ¥250 billion and a partial spinoff of its financial unit, listing it on September 29 and treating it as a discontinued business from the current quarter.
Uber Launches “Route Share” with Up to 50% Savings for Commuters
Uber launches “Route Share,” a cheaper pooled ride service up to 50% less expensive than UberX, targeting commuters with 20-minute interval pickups on busy corridors.
- Uber introduces “Route Share,” a budget pooled ride option up to 50% cheaper than UberX, available only during weekday commuting hours.
- Route Share operates on 20-minute intervals along busy corridors, with passengers potentially walking up to 15 minutes to pickup points and sharing with two others.
- Existing pooled rides reduce costs by up to 20%, but Route Share aims to attract more daily commuters amid rising transportation costs.
Alphabet’s 2024 Executive Pay Surpasses $215 Million Amid Revenue Growth
Alphabet’s 2024 executive compensation exceeded $215 million, led by stock awards, with CEO Pichai earning $10.725 million amid revenue of $350 billion and a 32:1 CEO-to-median employee pay ratio.
- Alphabet’s 2024 total executive compensation exceeded $215 million, with CEO Sundar Pichai earning $10.725 million, including $8.304 million in “all other compensation”
- Pichai’s 2024 pay increased 21.8% from 2023 but was significantly lower than $225.9 million in 2022; his compensation mainly comprised stock awards and other incentives
- CFO Anat Ashkenazi received $49.98 million, including $38.5 million in stock awards and a $9.9 million bonus; predecessor Ruth Porat earned $30.16 million
- Senior VP Prabhakar Raghavan and Chief Business Officer Philip Schindler each earned approximately $47 million, with stock awards of about $43.97 million; legal officer Kent Walker earned $30.16 million
- The median employee total compensation was $331,894, resulting in a CEO-to-median employee ratio of 32:1; revenue grew 14% in 2024 to $350 billion, with net income of $100 billion
NCSC Warns Rapid AI Deployment Raises Cybersecurity Risks by 2027
Organizations deploy AI rapidly without sufficient security measures, increasing vulnerabilities; NCSC warns this could enable advanced cyberattacks and systemic risks by 2027.
- At CYBERUK 2025, only 3 of 200 security professionals had banned generative AI; none understood AI security risks.
- The UK National Cyber Security Centre (NCSC) warns organizations deploy AI insecurely, increasing attack surface and vulnerability to cyber threats.
- NCSC report predicts that by 2027, AI-enabled attackers will significantly reduce vulnerability exploitation times, heightening cyber risks across supply chains.
Gartner Expert Warns Against Rushing Patch Updates to Outpace Threats
Gartner’s Craig Lawson recommends delaying Patch Tuesday updates, arguing organizations can’t outpace threat actors and should develop tailored patching strategies with compensating controls.
- Gartner Research VP Craig Lawson advises against rushing patching efforts, stating “Nobody has ever out-patched threat actors at scale.”
- Most organizations cannot keep up with patching, and accelerating efforts may be counterproductive due to patch complexity and potential for breaking systems.
- Criminals exploit only 8-9% of vulnerabilities, often ignoring even zero-day flaws, and most targeted vulnerabilities are not rated critical.
Post Office Ends NBIT Project, Launches £410M Procurement for Horizon Support
Post Office ceases in-house NBIT development, initiates £410M procurement for Horizon support and cloud migration, and seeks COTS EPOS software, amid ongoing Horizon scandal remediation efforts.
- UK Post Office ends in-house NBIT project and launches £410M procurement for alternative suppliers
- NBIT replacement plans halted; existing development repurposed for incremental branch improvements
- Procurement includes support for Horizon datacenter migration to cloud and a new off-the-shelf EPOS system, with contracts lasting 12 years
FTC Delays ‘Click-to-Cancel’ Rule Enforcement to 2025
The FTC postpones enforcing the ‘click-to-cancel’ rule until July 14, 2025, requiring companies to simplify subscription cancellations, after assessing compliance burdens amid commissioner vacancies.
- The FTC delays enforcement of its ‘click-to-cancel’ rule from May 14 to July 14, 2025
- The rule mandates that subscription cancellations be as easy online as signing up
- The delay follows a “fresh assessment” of compliance burdens, with a 3-0 vote amid two absent commissioners due to illegal firing
Microsoft Extends Windows 10 Security Updates to 2028 to Support Windows 11 Upgrade
Microsoft announced it will continue providing security updates for Office 365 apps on Windows 10 until October 10, 2028, delaying support end from 2025 to aid transition to Windows 11.
- Microsoft extends security updates for Microsoft 365 apps on Windows 10 through October 10, 2028
- Originally planned to end support after October 14, 2025, but now support is extended by three years
- Updates will be delivered via standard channels; users are advised to upgrade to Windows 11 to avoid performance issues
▶️ Technology
OpenAI Launches GPT-4.1 and Mini Models to Boost Coding and Speed
OpenAI released GPT-4.1 and GPT-4.1 mini models in ChatGPT on May 14, 2025, improving coding, instruction following, and speed, with broader transparency and safety evaluation efforts.
- OpenAI announced the release of GPT-4.1 and GPT-4.1 mini models in ChatGPT on May 14, 2025
- GPT-4.1 enhances coding and instruction-following capabilities, is faster than GPT-4o, and aims to assist software engineers
- GPT-4.1 is available to ChatGPT Plus, Pro, and Team subscribers; GPT-4.1 mini is free for paying users; GPT-4.0 mini is being removed
Google Prepares DeX-Like Desktop Mode for Android Release
Google is working on an Android desktop mode resembling Samsung DeX, featuring a taskbar, multitasking in resizable windows, and external display management, possibly arriving in Android 17.
- Google is developing a DeX-like desktop mode for Android, previewed on a Pixel device, not yet part of Android 16 stable.
- The desktop mode includes a taskbar with pinned/recent apps, supports multiple floating, resizable windows, and drag-and-drop multitasking.
- It adapts Android’s tablet windowing environment for external displays, with potential release in a quarterly update or Android 17, likely as an opt-in developer feature.
Nintendo Switch 2 Unveils Powerhouse Specs with Nvidia T239 SoC and Advanced Features
Nintendo’s Switch 2 features a custom Nvidia T239 SoC with ARM Cortex A78C CPU cores, Ampere GPU with 1536 CUDA cores, 12GB LPDDR5X RAM, and 256GB UFS storage, supporting DLSS, ray tracing, HDR10, and VRR up to 120Hz.
- Nintendo officially announced the Switch 2’s technical specifications on May 14, 2025, featuring a custom Nvidia T239 SoC.
- The CPU uses ARM Cortex A78C cores at 998MHz (docked) and 1101MHz (mobile), with a maximum of 1.7GHz; the GPU is Ampere architecture with 1536 CUDA cores at 1007MHz (docked) and 561MHz (mobile).
- Memory includes 12GB LPDDR5X with 102GB/s bandwidth (docked) and 68GB/s (mobile), reserving 3GB for system, leaving 9GB for developers; storage is 256GB UFS with MicroSD support up to 2TB.
Understanding HDR: Camera Mode, Displays, and New Tone-Mapping Features
HDR means either “HDR mode” in cameras or high contrast displays; Halide offers single-shot tone mapping inspired by analog techniques, with HDR support in iOS 18 and customizable levels.
- HDR (High Dynamic Range) has two meanings: “HDR mode” introduced in iPhone cameras in 2010 and displays with higher vibrancy and contrast.
- Dynamic range refers to the difference between the darkest and brightest parts of a scene; scenes with high contrast challenge cameras and screens.
- Halide’s new optional tone-mapping feature derives from analog dodging and burning, working from a single capture to enhance local contrast without multi-exposure algorithms.
Noyb Challenges Meta’s GDPR Compliance in AI Data Training
Noyb accuses Meta of GDPR violations by training AI on EU user data without explicit opt-in, challenging Meta’s legitimacy interest claim; legal action may follow.
- Noyb sent a cease and desist letter to Meta on May 14, 2025, alleging GDPR violations in AI training data collection.
- Meta claims its data collection complies with EDPB guidance and relies on legitimate interest under GDPR Article 6(1).
- Noyb argues Meta’s reliance on legitimate interest is incorrect, asserting no overriding legitimate interest exists and compares it to previous GDPR violations regarding ad targeting.
Chinese Researchers Develop Breakthrough Silicon-Free 2D Transistor Outperforming Silicon Chips
Chinese scientists created the fastest, most efficient silicon-free 2D transistor using bismuth-based materials and GAAFET design, outperforming existing silicon chips by 40% in speed and 10% in energy efficiency.
- Researchers at Peking University developed a 2D, silicon-free transistor using bismuth oxyselenide (Bi₂O₂Se) and bismuth selenite oxide (Bi₂SeO₅)
- The transistor employs a gate-all-around field-effect transistor (GAAFET) architecture, encircling the channel for improved control and energy efficiency
- It achieves 40% higher speed and 10% lower energy consumption compared to current 3-nanometer silicon chips, with potential for scalable manufacturing