Today’s top stories include Microsoft’s Q4 earnings surpassing expectations with over $76 billion in revenue, driven by cloud and AI growth, and Meta’s $72 billion investment in AI infrastructure aiming to develop personal superintelligence by 2030. Meanwhile, Minnesota declared a state of emergency following a coordinated cyberattack on Saint Paul.
▶️ Internet Infrastructure
Copyparty: The Portable Cross-Platform File Server with Resumable Uploads
Copyparty is a self-contained, cross-platform file server supporting accelerated resumable uploads, deduplication via symlinks/hardlinks/reflinks, multiple protocols, zeroconf, media indexing, and customizable web UI, with no external dependencies.
- Copyparty is a portable file server supporting resumable uploads, deduplication, WebDAV, FTP, TFTP, zeroconf, media indexing, and thumbnails, with no dependencies.
- It is distributed as MIT-licensed, with 14.2k stars and 443 forks on GitHub.
- Supports cross-platform operation (Windows, Linux, macOS, Android, FreeBSD, ARM, PPC, RISC-V), with features like protocols, reverse proxy, zeroconf, and media transcoding.
Trae IDE Bypasses User Settings to Send ByteDance Telemetry Data
Trae IDE continuously transmits extensive user and system telemetry to ByteDance servers, bypassing user settings, raising privacy, security, and community trust concerns.
- Trae IDE, ByteDance’s Visual Studio Code fork, exhibits persistent telemetry transmission despite user disabling attempts
- Network analysis shows regular outbound connections to ByteDance servers (
mon-va.byteoversea.com,maliva-mcs.byteoversea.com) - Telemetry data includes detailed hardware info, user activity, session metrics, and project details, with data transmitted even when disabled
Oxide Raises $100M Series B to Scale On-Prem Cloud Hardware and Software
Oxide secured a $100M Series B funding, enabling large-scale manufacturing and operations, after developing integrated hardware and software solutions for on-premises cloud computing, with initial system shipped in 2023.
- Oxide raised a $100M Series B led by USIT, more than doubling total funding to $189M since 2019
- The company developed hardware and software components including custom board designs, microcontroller OS, platform software, host hypervisor, switch, storage service, and control plane
- First system shipped two years prior, with subsequent software updates, customer deployments, and general availability achieved
UK Extends G-Cloud 14 to 2026 Amid G-Cloud 15 Delays
UK extends G-Cloud 14 by six months at a cost of up to £1.65 billion due to delays in launching G-Cloud 15, which is planned to run from March 2026 to September 2027.
- UK government extended G-Cloud 14 framework by six months until October 28, 2026, at a cost of up to £650 million for lots 1-3 and £1 billion for lot 4
- Original G-Cloud 14 deal was set to end on April 28, 2026; replacement G-Cloud 15 procurement, worth up to £4.8 billion, was initiated in April 2025 with a scheduled period from March 18, 2026, to September 17, 2027
- The extension aims to facilitate procurement of G-Cloud 15 due to delays in finalizing the new framework, with more details to be announced later
Ofcom Fines Gigaclear Over Emergency Call Location Failures
Ofcom fined Gigaclear for failing to ensure accurate emergency caller location data from January 2022 to March 2024, impacting 948 calls, due to configuration and testing failures.
- Ofcom fined UK broadband provider Gigaclear £122,500 (approximately $164,000) for failing to deliver accurate caller location data during emergency calls from January 2022 to March 2024
- The failure affected 948 VoIP calls, providing either inaccurate or no location information to emergency services
- The regulator identified issues with third-party system configuration, lack of testing, and inadequate investigation of customer complaints, despite Gigaclear’s self-reporting and corrective actions
SafePay Ransomware Threatens Data Leak Despite Ingram Micro’s Claims
SafePay ransomware threatens to leak 3.5 TB of Ingram Micro data on August 1, indicating ongoing extortion despite the company’s claims of incident containment and partial website restoration.
- SafePay ransomware group posted Ingram Micro’s data on July 29, threatening to leak 3.5 TB on August 1
- The attack caused multi-day outage; Ingram Micro claims incident was contained but remains on leak site
- Ingram Micro’s public info page has not been updated since July 9, despite claiming global operations restored
Microsoft Q4 2025 Earnings Surge on Cloud and AI Growth
Microsoft’s Q4 2025 earnings surpassed expectations, driven by cloud and AI revenue, with Azure exceeding $75 billion in FY2025, reflecting 34% growth and market leadership in AI infrastructure.
- Microsoft reported Q4 2025 revenue of $76.4 billion, up 18% year-over-year, with operating income of $34.3 billion and net income of $27.2 billion.
- First public disclosure of Azure revenue, exceeding $75 billion for FY2025, up 34% from the previous year, averaging nearly $19 billion per quarter.
- Azure’s growth driven by all workloads, including AI; Microsoft claims to lead the AI infrastructure market and gained market share each quarter.
- Total FY2025 revenue reached $281.7 billion, up 15%, with operating income of $128.5 billion and EPS of $13.64.
- Microsoft’s cloud segment, Azure, contributed $29.9 billion in revenue, up 26%; other segments include Productivity and Business Processes ($33.1 billion) and More Personal Computing ($13.5 billion).
- Microsoft plans to spend $80 billion in FY25 on AI data center infrastructure; AI monetization expected to follow a per-user and consumption-based model.
- Stock price increased about 8% post-earnings, pushing market cap beyond $4 trillion, second only to Nvidia.
Minnesota Declares Emergency After Coordinated Cyberattack on Saint Paul
Minnesota’s governor called in the National Guard and declared a state of emergency after a deliberate, coordinated cyberattack on Saint Paul, disrupting city services and ongoing investigations.
- Minnesota Governor Tim Walz activated the National Guard and declared a state of emergency due to a deliberate, coordinated cyberattack on Saint Paul.
- The attack was identified as a sophisticated external digital assault, causing significant disruptions, including network outages and impaired city services.
- City officials first detected signs of intrusion on Friday; the attack persisted through the weekend, with ongoing investigations involving FBI, federal, state, and local agencies, and cybersecurity firms.
Major Earthquake Off Russia’s Coast Triggers Tsunami Without Disrupting Communications
A magnitude 8.8–8.9 earthquake off Russia’s eastern coast on July 30, 2025, caused a 30cm tsunami but did not impact submarine cables or cloud services, with no outages reported.
- Magnitude 8.8 (USGS) and 8.9 (Russian Geophysical Survey) earthquake occurred off eastern Russian coast on July 30, 2025, at 09:24:50 local time
- Tsunami waves of approximately 30cm reported in Japan; warnings issued across the Pacific, including US West Coast
- No disruptions reported at submarine cables (including Petropavlovsk-Kamchatsky–Anadyr link operated by Rostelecom) or major cloud providers (AWS, Azure, Google Cloud); no outages at communication or cloud facilities
▶️ Open Source
Spanish police link Pixel phones with crime due to privacy-enhancing GrapheneOS features
Police in Catalonia link Pixel phones with crime due to GrapheneOS’s privacy and security benefits, which limit data access, enhance sandboxing, and include features like a data wipe PIN.
- Spanish police associate Google Pixel devices with crime due to increased drug traffickers using them, but not because of Google’s Titan M2 chip.
- The association is linked to criminals favoring Pixel phones with GrapheneOS, a privacy-focused OS that enhances security and sandboxing.
- GrapheneOS offers granular app permission controls, sandboxing of Google apps, a duress PIN for data deletion, and improved security features, without sacrificing modern software compatibility.
EU Age Verification App Uses Google Play Integrity to Enforce Licensing Restrictions
The EU’s open-source age verification app incorporates Google Play Integrity for remote attestation, enforcing licensing and security checks that restrict use to Google-licensed Android systems.
- The EU is developing an open-source, privacy-preserving age verification app available at github.
- The app plans to include remote attestation using Google Play Integrity, requiring systems to be licensed by Google, download from Play Store, and pass device security checks.
- This implementation ties the app to Google services, excluding aftermarket Android OS like GrapheneOS, and prevents use on unlicensed systems, potentially rejecting custom or privacy-focused OS.
▶️ Software Development
Zed Code Editor Adds AI-Free Mode Amid Privacy Concerns
Zed, a fast Rust-based code editor, introduces an AI-free mode in version 0.197, enabling users to disable LLM integrations amid ongoing AI skepticism and privacy concerns.
- Zed, a Rust-based text editor for programmers, now allows users to disable all AI and LLM bot features in Preview build 0.197, announced in late July 2025
- The feature addresses user concerns about corporate restrictions on AI tool usage and privacy issues, responding to calls from the Zed community over the past year
- Zed emphasizes native performance, smaller size, and conflict-free replicated data types for collaborative editing, contrasting with Electron-based editors like Atom and VS Code
▶️ Management and Leadership
Pew Study Finds Google AI Overviews Cut Website Clicks by Half
Pew Research finds Google AI Overviews decrease website clicks by nearly 50%, with only 1% of AI summaries leading to source visits, raising concerns over web traffic and information accuracy.
- Pew Research Center analysis shows AI Overviews in Google search results significantly reduce clicks to websites, with a drop from 15% to 8%
- Searches with AI summaries result in only 1% of AI Overviews leading to source clicks; sources mainly include Wikipedia, YouTube, and Reddit
- About 20% of searches now feature AI Overviews, especially in longer or question-based queries, increasing the prevalence of AI-generated answers
Websites Overtake Apps as Safer Privacy Choice in 2025
In 2025, websites offer full functionality without privacy risks, as apps seek deeper device access for data collection, often using dark patterns to push users to download them.
- Article published on July 2, 2025, by Ibrahim Diallo, emphasizing the importance of using websites over apps for privacy reasons.
- Highlights how apps access deeper device functions (contacts, location, microphone, installed apps), enabling extensive data collection.
- Argues that web browsers are increasingly capable, making app installation unnecessary for functionality, while apps pose privacy risks and data sharing concerns.
Meta Reports 22% Revenue Growth and 36% Net Income Increase in Q2 2025
Meta’s Q2 2025 revenue grew 22% to $47.52 billion, with net income up 36% to $18.34 billion, driven by increased user engagement and ad pricing, amid $27.07 billion costs.
- Meta reported Q2 2025 revenue of $47.52 billion, up 22% year-over-year, with net income of $18.34 billion, increasing 36%
- Operating income rose 38% to $20.44 billion; operating margin increased to 43%
- Family of Apps DAU averaged 3.48 billion in June 2025, up 6% YoY; ad impressions increased 11%, and average ad price rose 9% YoY
Prem Qu Nair Joins Cognition to Advance AI-Driven Software Engineering
Prem Qu Nair joined Cognition to develop advanced IDEs and coding agents, citing a passion for software engineering and AI, after leaving Windsurf and a deal with Google DeepMind.
- Prem Qu Nair announced joining Cognition to focus on the future of software engineering, emphasizing AI+code development.
- Previously employee #2 at Windsurf, with over 3.5 years, and had a deal with Google DeepMind.
- Nair forfeited all vested shares at Windsurf, receiving only 1% of their estimated worth at the deal time.
ChatGPT Study Mode Reveals Car Ownership Costs and Encourages Alternatives
ChatGPT’s Study Mode guided the author through a detailed cost-benefit analysis, revealing that car ownership costs exceed $6,000 annually and recommending alternatives, leading to the decision to remain car-free.
- The author used ChatGPT’s Study Mode to evaluate whether to buy a car, focusing on decision-making rather than direct answers
- Study Mode prompted reflection on costs, benefits, and personal preferences, highlighting that annual expenses for car ownership could reach $6,000–$8,000
- It provided insights on EVs, depreciation, insurance costs, and alternatives like carpooling and renting, ultimately advising against purchasing a car
Meta Q2 Earnings Surpass Estimates Driven by AI Investments and Talent Costs
Meta’s Q2 earnings exceeded estimates with $47.52 billion revenue and $7.14 EPS, driven by AI talent costs, new AI hardware sales, and strategic investments, boosting stock over 12% to record levels.
- Meta’s Q2 revenue was $47.52 billion, surpassing estimates of $44.83 billion; EPS was $7.14, exceeding expected $5.89.
- Stock surged over 12% after hours to $778, a record high, driven by strong earnings and AI investment optimism.
- Employee compensation for “technical talent in priority areas” was Meta’s second-largest cost driver, amid a $15 billion investment in Scale AI and hiring of AI experts.
Microsoft surpasses earnings estimates, hits $4 trillion market cap
Microsoft’s Q4 earnings of $76.4 billion and $3.65 EPS exceeded estimates, propelling its market cap above $4 trillion; Azure revenue grew 34% to over $75 billion amid increased capital spending.
- Microsoft reported Q4 revenue of $76.4 billion and earnings of $3.65 per share, surpassing analyst estimates of $73.89 billion and $3.37.
- Shares increased over 9% after hours, elevating Microsoft’s market cap above $4 trillion, joining Nvidia.
- Azure revenue exceeded $75 billion, up 34%, driven by growth across all workloads; Microsoft plans to spend $30 billion on capital expenditures in the upcoming quarter.
Microsoft CFO Calls for “Intensity” in FY26 Amid Strong Profit and Azure Growth
Microsoft CFO Amy Hood announced FY26 will require “intensity,” emphasizing security, quality, and AI, following a $27 billion quarterly profit and Azure revenue growth of 34%.
- Microsoft CFO Amy Hood sent an internal email on July 31, 2025, emphasizing the need for “intensity” in FY26.
- Hood highlighted Azure revenue surpassing $75 billion with 34% growth and reiterated priorities: security, quality, and AI transformation.
- The email did not mention Microsoft’s workforce reductions exceeding 10,000 employees this year, despite profit growth.
Microsoft Q4 FY25 Revenue Soars 18% Driven by Cloud and AI Growth
Microsoft Q4 FY25 revenue grew 18% to $76.4 billion, driven by cloud and AI, with Azure revenue up 34%, and full-year revenue increased 15% to $281.7 billion.
- Microsoft Q4 FY25 revenue reached $76.4 billion, up 18% year-over-year; net income increased 24% to $27.2 billion; EPS rose 24% to $3.65
- Cloud and AI drove growth, with Azure surpassing $75 billion in revenue, up 34%; Microsoft Cloud revenue was $46.7 billion, up 27%
- Segment revenues: Productivity and Business Processes at $33.1 billion (+16%), Intelligent Cloud at $29.9 billion (+26%), More Personal Computing at $13.5 billion (+9%)
Legendary Satirist Tom Lehrer Dead at 97
Tom Lehrer, a mathematician and satirical songwriter famous for darkly humorous songs like “Poisoning Pigeons in the Park,” died at 97, leaving a legacy of sardonic lyrics set to cheerful music.
- Tom Lehrer, Harvard-trained mathematician and satirist, died at age 97 in Cambridge, Mass., confirmed by friend David Herder.
- Known for darkly memorable songs like “The Vatican Rag” and “Poisoning Pigeons in the Park,” performed with sardonic lyrics to cheerful melodies.
- Gained popularity in 1950s-60s on college campuses and Greenwich Village, with recordings sold in the hundreds of thousands, often via mail order.
Google Partners with Energy Dome to Enable 24-Hour CO2 Energy Storage
Google’s partnership with Energy Dome involves deploying CO2 gas bag energy storage systems, enabling over 24 hours of renewable energy storage, with proven Italian facilities and potential grid or onsite use.
- Google and Energy Dome announced a deal for deploying CO2-based energy storage gas bags and making a strategic investment, amount unspecified.
- Energy Dome’s long-duration energy storage (LDES) system uses pressurized CO2 in a closed-loop cycle, with a 20MW/200MWh plant operational for over three years.
- The system stores excess renewable energy by compressing CO2 into a liquid and later heating it to generate electricity, providing energy for over a day, surpassing typical lithium-ion battery duration.
CISA to Release 2022 Telecom Security Report Amid Political Pressure
CISA will release the 2022 US Telecommunications Insecurity Report after years of withholding, amid political pressure from Senator Wyden, revealing critical security vulnerabilities in US telecom infrastructure.
- CISA agreed to release the unclassified 2022 US Telecommunications Insecurity Report, developed under the Biden administration, but did not specify a release date.
- The report highlights poor security practices in US telecom networks, with implications for national security and foreign espionage.
- Senator Ron Wyden (D-OR) has blocked CISA’s nominee Sean Plankey until the report’s release, with legislation passed requiring its disclosure within 30 days of enactment.
Microsoft to Block Public Folder Migrations from Old Exchange Servers in 2025
Microsoft will block public folder migrations from Exchange Server 2010 and earlier starting October 1, 2025, requiring customers to upgrade first; this aims to phase out legacy systems and enhance long-term service stability.
- Support for public folder migrations from Exchange Server 2010 and older will be deprecated starting October 1, 2025, including ongoing migrations.
- Customers must complete migrations before October 1 or migrate to a newer Exchange version first, then to Exchange Online.
- Microsoft states the deprecation aims to reduce reliance on outdated systems and improve service reliability, despite Exchange Server 2010 support ending in 2020.
UK MoD Launches International Defence Esports Games to Boost Cyber Skills
The UK MoD is launching the 2026 International Defence Esports Games to improve cyber and digital skills, leveraging esports to prepare servicepeople for modern warfare challenges.
- UK Ministry of Defence (MoD) is establishing a new esports tournament, the International Defence Esports Games, scheduled for late 2026, organized by British Esports.
- MoD recognizes esports as a military sport to enhance cyber understanding and digital literacy; the Royal Navy installed an esports facility aboard HMS Prince of Wales.
- Lt. Gen. Sir Tom Copinger-Symes states esports and serious games improve warfighting readiness by developing team coordination, rapid decision-making, and cyber skills, citing Ukraine’s drone and cybersecurity training initiatives.
IBM Report Warns of Growing AI Security Risks and Lack of Governance
IBM’s 2025 report reveals organizations’ neglect of AI security and governance, with 13% experiencing AI-related breaches, mainly via supply chain, risking data, operations, and trust.
- IBM’s Cost of a Data Breach Report 2025 analyzes 600 organizations globally from March 2024 to February 2025
- 13% of surveyed organizations reported security incidents involving AI models or applications, with 97% lacking proper AI access controls
- Nearly one-third of breaches caused operational disruption, unauthorized data access, or financial loss; supply chain compromise, especially third-party SaaS, was the primary source
- 87% of organizations have no governance for AI risk; two-thirds do not conduct regular risk audits, and over 75% do not perform adversarial testing
- The report highlights risks from shadow AI, which is used unofficially without oversight, increasing vulnerability to exploitation
- IBM warns that the lack of basic access controls exposes sensitive data and models, emphasizing AI security as foundational amid rapid AI adoption driven by organizational fear of lagging behind
Italy’s AGCM investigates Meta over AI integration in WhatsApp without user consent
Italy’s AGCM investigates Meta for integrating AI into WhatsApp without user consent, potentially violating EU competition law by leveraging market dominance and imposing AI features.
- Italian Competition Authority (AGCM) investigates Meta for adding AI services to WhatsApp without user request, placing it prominently in the interface.
- The AGCM suspects Meta’s AI integration may violate Article 102 of the TFEU, potentially abusing market dominance and harming competitors.
- Meta claims AI features are optional, not constantly active, and asserts compliance with law; investigation includes an inspection at Meta’s Italian subsidiary.
Palo Alto Networks Acquires CyberArk for $25B to Boost Identity Security
Palo Alto Networks is acquiring CyberArk for $25 billion to strengthen identity security and privileged access management, supporting its platform strategy amid rising machine identities and AI integration.
- Palo Alto Networks announced a $25 billion cash-and-stock acquisition of CyberArk on July 30, 2025
- The deal is the company’s largest to date and one of the year’s most expensive, behind Google’s $32 billion purchase of Wiz
- CyberArk specializes in identity security and privileged access management, addressing the needs of verifying human and machine identities amid AI proliferation
- CyberArk investors will receive $45 in cash and 2.2005 Palo Alto shares per share, with closing expected in H2 fiscal 2026
- The acquisition aims to enhance Palo Alto’s platform strategy by integrating identity security into its cloud, network, and endpoint security products
- CyberArk reports machine identities outnumber human identities by 40:1, a ratio expected to increase with AI agent adoption
- CEO Nikesh Arora emphasizes the importance of privilege controls for all identities in AI-driven security environments
Trump CMS Announces Private-Partner Digital Health Ecosystem Amid Privacy Concerns
The Trump administration and CMS plan to create a digital health ecosystem with private partners to enhance patient-controlled medical data sharing, amid privacy and security concerns.
- Trump administration and CMS announced development of the CMS Digital Health Ecosystem to improve patient access to medical records, involving private partners including Epic, Oracle, Amazon, Apple, Google, Microsoft, OpenAI, UnitedHealth, and others.
- The initiative aims to enable apps for AI assistants, paperless forms, and chronic disease management, with a long-term goal of patient-controlled data sharing via opt-in systems, avoiding centralized government databases.
- The system’s development is in early stages; no timeline announced. Concerns include privacy risks, data security, and potential misuse, despite assurances of strict opt-in and no centralized database.
▶️ Technology
Google Deploys ML Age Estimation in US to Protect Minors and Enforce Parental Controls
Google is launching ML-powered age estimation in the US to enforce protections for minors, analyzing signals like search and viewing history, with verification options and privacy safeguards.
- Google is deploying an ML-based age estimation model in the US to identify users under or over 18.
- The model analyzes signals such as search history, YouTube viewing categories, account longevity, and user-provided data.
- If estimated under 18, protections include YouTube Wellbeing tools, disabling Timeline in Maps, restricting personalized ads, and blocking minors from adult apps on Google Play.
- Users receive email notifications; verification options include government ID, digital ID, phone lookup, credit card, or on-device selfie.
- The approach avoids collecting additional data or sharing granular user info; rollout begins in the next few weeks to a small user subset, with monitoring before wider deployment.
- Age estimation is already used successfully in other countries, with parental controls recommended for accounts used by children.
OpenAI Launches Study Mode in ChatGPT to Enhance Active Learning
OpenAI introduced study mode in ChatGPT, a guided learning feature with interactive prompts, personalized feedback, and scaffolding, designed to support deeper understanding and active learning.
- Study mode in ChatGPT launched on July 29, 2025, for logged-in users on Free, Plus, Pro, Team, with ChatGPT Edu coming soon.
- Powered by custom system instructions developed with educators, scientists, and pedagogy experts to promote active participation, cognitive load management, metacognition, curiosity, and supportive feedback.
- Features include Socratic questioning, hints, self-reflection prompts, scaffolded responses, personalized support, knowledge checks, and toggleable study mode.
Unified Rust GPU Framework Supports Cross-Platform Compute Without Shader Languages
A Rust GPU project enables a single Rust codebase to run on all major GPU platforms, using compile-time kernel translation to SPIR-V, PTX, or native code, with full no_std support and platform-specific abstractions.
- Demonstrates a unified Rust codebase running on CUDA, SPIR-V, Metal, DirectX 12, WebGPU, and CPU targets
- Uses Rust features and conditional compilation to target multiple GPU backends without shader languages
- Kernels are compiled at build time with
rustc_codegen_spirvandrustc_codegen_nvvm, embedded into binaries, enabling cross-platform GPU compute in Rust
Old MacBook Pro M2 Runs 44GB Quantized AI Model to Generate Space Invaders Code
A 2.5-year-old laptop (MacBook Pro M2 with 64GB RAM) successfully generated working Space Invaders code using the 44GB 3-bit quantized GLM-4.5 Air model (details), demonstrating advanced local AI coding capabilities.
- Used the 44GB 3-bit quantized GLM-4.5 Air model (106 billion parameters, 205.78GB) on a 64GB MacBook Pro M2
- Generated functional Space Invaders HTML/JavaScript code with no edits, response included 4193 tokens, peak memory 47.687 GB
- Ran model via
mlx-lmlibrary with specific commit support, usinguvfor execution, and downloaded model weights from Hugging Face
Meta invests $72B in AI and recruits top talent to build superintelligence
Meta aims to develop superintelligence by leveraging $72 billion in AI infrastructure and recruiting top talent, focusing on multi-gigawatt clusters and small research teams.
- Meta CEO Mark Zuckerberg states the company has “all the ingredients”—compute and talent—to build superintelligence.
- Meta is recruiting top AI researchers with nine-figure offers and investing $72 billion in AI infrastructure, including multi-gigawatt clusters the size of Manhattan.
- The company is shifting to small, talent-dense teams to optimize research on superintelligence, following a lukewarm reception to its Llama 4 model.
The Power of Speed: Boosting Productivity and Focus with Fast Software
Fast software enhances productivity, focus, and user experience by reducing latency and complexity, with future AI tools expected to prioritize optimization for new capabilities.
- Emphasizes the importance of software speed in changing user behavior and productivity, with examples like Superhuman’s sub-100ms response time and Mercury’s instant bank transfers
- Fast software reduces cognitive friction, signals prioritization and focus, and often requires stripping non-essential features, exemplified by Linear versus Workday
- Achieving speed involves complex behind-the-scenes optimizations, handling background messiness, and will become a focus in future AI development for lower latency and better user experience
AI HUDs Over AI Copilots: Enhancing Human Awareness and Senses
The article argues for AI HUDs over copilots, emphasizing interfaces like HUDs that extend human senses and awareness, aligning with Weiser’s 1992 vision of invisible, background computing.
- Advocates replacing AI copilots with AI HUDs inspired by aircraft Head-Up Displays
- Weiser’s 1992 critique emphasized designing interfaces that enhance human awareness without distraction
- Examples include spellcheck and custom AI debugging tools functioning as HUDs, not virtual assistants
Native Browser API and CSS Transitions Outperform SPAs in Performance and Accessibility
Modern CSS and browser APIs like View Transitions and Speculation Rules enable real, native page transitions, outperforming SPAs by reducing JavaScript, improving performance, SEO, and accessibility.
- Modern browsers, specifically Chromium-based ones, support native, declarative page transitions via the View Transitions API, enabling CSS-based animations between documents without JavaScript.
- CSS transitions, shared element animations, and speculation rules allow for smooth, real page navigation, shared element animations, and instant preloaded navigation, reducing reliance on JavaScript-heavy SPAs.
- Native browser features like bfcache and declarative transitions outperform traditional SPAs in performance, SEO, and accessibility, with examples showing 0KB JavaScript bundles and faster TTI (~1s) compared to SPAs (~3.5–5s).
Meta Aims to Develop Personal Superintelligence for 2030 Empowerment
Meta plans to create personal superintelligence to enhance individual agency, integrate into daily devices, and promote societal progress, emphasizing safety and broad accessibility by 2030.
- Meta envisions developing personal superintelligence to empower individuals, enabling goal achievement, creativity, and personal growth.
- The technology aims to be integrated into personal devices like glasses that understand context through sight and sound.
- Meta emphasizes broad sharing of superintelligence benefits while addressing safety risks and cautious open sourcing.
Meta invests billions in GPU Data Centers and AI Superintelligence Development
Meta is investing heavily in GPU infrastructure and AI research, aiming to develop and distribute personal superintelligence despite limited progress and unclear technical definitions.
- Meta invests tens of billions of dollars into GPU data centers, including a 2.2 GW AI supercluster in Louisiana and multiple multi-gigawatt datacenters planned through 2026.
- Zuckerberg announced efforts to develop AI superintelligence, claiming glimpses of self-improving AI systems, with little clarity on its precise definition or capabilities.
- Meta’s AI projects, including Llama 4 models, have underperformed expectations, leading to the cancellation of a 2-trillion-parameter model; the company plans to deploy over 1.3 million GPUs and spend up to $72 billion in 2025 on infrastructure.
Alibaba Reverts to Separate Models for Improved Qwen3 Performance
Alibaba abandoned hybrid “thinking” mode in Qwen3 models, focusing on separate instruct and thinking-tuned models with up to 2.8x performance gains and a 256k token context window.
- Alibaba’s Qwen 3 models initially featured a hybrid “thinking” and “non-thinking” mode introduced in April, which degraded performance and quality.
- The company has reverted to training separate instruct and thinking-tuned models, emphasizing quality over convenience.
- The latest models, including Qwen3-235B-A22B-Instruct-2507 and Qwen3-30B-A3B, show significant improvements in reasoning, math, and problem-solving benchmarks, with up to 2.8x gains in some cases.
- The context window has been expanded from 32k tokens to 256k tokens, enhancing long-term memory and reasoning capabilities.
- Alibaba plans to release updated versions, including code-tuned models, and continues research on hybrid thinking modes for future deployment.