Today’s highlights include Git 2.43’s new large file handling features, a decade-long Wikipedia campaign to inflate David Woodard’s profile, and OpenAI’s GPT-OSS-120B showing inconsistent performance across providers. Additionally, the Ritman Library digitized 2,178 occult books, and UK plans to implement AI crime prediction tools by 2030 amid privacy concerns.
▶️ Internet Infrastructure
Git 2.43 Enhances Large File Handling with Promisors and Partial Clones
Git’s upcoming 2.43 release introduces large object promisors to replace Git LFS, enabling efficient large file handling through server-side offloading and partial clone improvements.
- Git 2.43 introduces native support for large object promisors, aiming to replace Git LFS
- Partial clone with
--filterreduces clone size by up to 97% and speeds up cloning by 97% - Git LFS remains in use, but large object promisors are in development, with merging completed in March 2025
▶️ Open Source
Wikipedia Uncovers Decade-Long Coordinated Campaign to Inflate David Woodard’s Profile
A Wikipedia investigation revealed a decade-long, coordinated effort involving over 200 accounts to artificially boost David Woodard’s articles in 92 languages, culminating in the removal of 235 articles.
- A Wikipedia investigation identified a network of over 200 accounts and proxy IPs that created approximately 183 articles about David Woodard in 92 languages from 2017 to 2025.
- The activity aimed to artificially inflate Woodard’s prominence, with accounts often making minor edits, creating stub articles, and translating content in diverse languages, including constructed ones.
- Wikimedia stewards removed 235 articles and banned numerous accounts, reducing the total articles about Woodard to 20, indicating a decade-long self-promotion campaign likely orchestrated by Woodard or close associates.
PuTTY: Free Open-Source SSH Client and Terminal Emulator
PuTTY is a free, open-source SSH client and terminal emulator for Windows and Unix, primarily maintained by Simon Tatham, with download and main site links provided.
- PuTTY is a free SSH client and terminal emulator for Windows and Unix, maintained mainly by Simon Tatham
- Users can access the latest version via download page or main website
- The landing page includes images of configuration and terminal interfaces
OpenAI’s GPT-OSS-120B Performance Varies Widely Across Providers
OpenAI’s gpt-oss-120b shows inconsistent performance across providers, with scores from 36.7% to 93.3%, highlighting challenges in open weight model deployment and the need for conformance testing.
- Artificial Analysis benchmarked OpenAI’s gpt-oss-120b across multiple providers, revealing performance variances from 36.7% (CompactifAI) to 93.3% (Cerebras, Nebius Base, Fireworks, Deepinfra, Novita, Together.ai, vLLM 0.1.0) in AIME 2025 N=32 runs.
- Most high-scoring providers used the latest vLLM; Azure’s 80% score was corrected after a fix to their older vLLM implementation.
- CompactifAI’s low score (36.7%) is due to their models being highly compressed, offering 4-12x faster inference and 50-80% cost reduction; OpenAI now includes a compatibility test for providers to verify implementation correctness.
Ritman Library Digitizes 2,178 Rare Occult Books in Multiple Languages
The Ritman Library digitized 2,178 rare occult books in multiple European languages, revealing complex alchemical and mystical texts, funded by Dan Brown, available online since August 2025.
- Digitized 2,178 occult books from the Ritman Library, made available online in August 2025
- Collection includes texts in Latin, German, Dutch, and French, primarily pre-1900, with complex alchemical, astrological, and mystical content
- Funded by Dan Brown’s donation, part of the “Hermetically Open” digital project, accessible through the online reading room
▶️ Software Development
Simple and Reliable System Design for Scalability and Robustness
Effective system design uses simple, reliable components, minimizes statefulness, optimizes database queries, employs background jobs and caching, and focuses on critical hot paths for robustness and scalability.
- The article emphasizes that good system design relies on simple, well-tested components rather than clever tricks or over-complexity.
- It advocates minimizing stateful components, favoring stateless services, and centralizing state management to reduce system fragility.
- Key components discussed include schemas and indexes for databases, query optimization, read-replica usage, background jobs, caching strategies, event hubs like Kafka, push vs. pull data flow, hot path optimization, logging, metrics, and fail-safe mechanisms.
▶️ Management and Leadership
PureGym PassKit API Flaws Exploited to Speed Up Entry and Enable Wallet Passes
Engineer reverse engineered PureGym’s PassKit API, exploiting security flaws to create dynamic Apple Wallet passes, reducing gym entry from 47 to 3 seconds and enabling location-based membership display.
- Reverse engineered PureGym’s PassKit implementation, revealing API endpoints and security flaws, including the use of an 8-digit PIN as API credentials
- Intercepted traffic shows QR codes refresh every minute, with API responses indicating polling every 60 seconds despite a 7-day QR code expiry
- Built a Swift backend in Vapor to generate and update Apple Wallet passes, enabling gym access via proximity and Apple Watch, reducing entry time from 47 seconds to 3 seconds
Uber Freight Faces Profitability Challenges and Weak Integration with Uber Core Business
Uber Freight remains unprofitable, with limited synergy to Uber’s core businesses, as its digital trucking strategy has failed to deliver expected financial or strategic gains.
- Uber Freight’s CEO Rebecca Tinucci, appointed after Lior Ron’s departure, faces a challenging, unprofitable freight division with limited synergy with Uber’s core ride-hailing and delivery businesses.
- Uber Freight’s business model has failed to generate expected profits, and its integration with Uber’s main operations remains weak.
- The company’s bet on digitizing the trucking industry using Uber’s technological expertise has not yielded significant financial or strategic benefits.
AI Engineers’ Rising Salaries Spark Dating Boom in Silicon Valley
AI engineers’ high salaries and industry stability boost their desirability in dating, with matchmakers noting increased interest and a shift in social perceptions of “nerds” as attractive.
- AI engineers in the Bay Area are experiencing increased demand in the dating market due to rising salaries and industry interest.
- Top tech companies like Meta, Microsoft, and OpenAI offer annual compensation in the millions for AI talent.
- Matchmakers report a surge in client interest from Silicon Valley, with VIP fees ranging from $5,900 to $500,000, averaging around $100,000.
Court May Mandate Google to Sell Chrome Amid Antitrust Concerns
A court may force Google to divest Chrome following antitrust rulings; bidders include Search.com, Perplexity, OpenAI, and Yahoo, with bids exceeding $34 billion, risking a major impact on Google’s market position.
- A federal judge will decide this month whether Google must sell Chrome due to antitrust violations in search and digital ad markets.
- Google’s sale of Chrome could cause a 15-25% decline in Alphabet Inc. stock, according to Barclays analysts.
- Competitors like Search.com ($35 billion bid) and Perplexity ($34.5 billion bid) have expressed interest in acquiring Chrome; OpenAI and Yahoo also show interest.
Jigar Bhati Joins OpenAI to Tackle Scaling Challenges in a Startup Culture
Jigar Bhati, a former Twitter engineer, joined OpenAI in 2023, leveraging infrastructure expertise to tackle scaling issues; the company retains a startup environment with rapid development and impactful AI product development.
- Jigar Bhati joined OpenAI in 2023 after nearly 7 years at Twitter, citing a desire for a startup-like environment
- His infrastructure and large-scale distributed systems experience helped address scaling challenges at OpenAI
- OpenAI maintains a startup culture with high velocity, daily product shipping, and emphasis on productivity and collaboration
Election Officials Warn of Security Risks Amid CISA Cuts in 2026
Election workers fear increased threats and reduced security support in 2026 due to CISA budget and staff cuts, risking election integrity and safety.
- Election officials express concern over reduced federal support from CISA for 2026, citing shrinking budget and staff.
- Former officials report increased threats, harassment, and violence, including death threats and doxxing, post-2020 election.
- A 2025 Brennan Center survey shows 61% of election officials worry about security impacts from federal cuts, with 38% experiencing threats.
10 Windows Features to Boost Productivity and Cut Distractions
The article outlines ten essential Windows features for productivity, criticizing Microsoft’s current focus on unnecessary AI, cloud, and promotional features, advocating for customization and user control.
- The article advocates for ten specific Windows features to enhance user productivity, criticizing current Microsoft additions perceived as unnecessary distractions.
- Proposed features include multiple clipboards, additional taskbar clocks, a new modifier key, remappable shortcuts, a movable taskbar, audio permissions, app pinning on specific screens, grouped app launches, easier audio device switching, and reduced OS notifications.
- The author criticizes Microsoft’s focus on AI, cloud features, and promotional alerts, emphasizing the need for a streamlined, user-centric OS that prioritizes productivity and customization.
UK to Launch AI Crime Prediction Tools by 2030 Amid Privacy Concerns
UK authorities will deploy AI-based crime prediction tools by 2030, combining multi-source data and machine learning, amid privacy concerns and ethical safeguards.
- UK government plans to implement AI-driven crime prediction maps by 2030, integrating police, council, and social service data, including criminal records and behavioral patterns.
- Prototype systems are expected by April 2026, with full deployment across England and Wales as part of the £500 million R&D Missions Accelerator Programme.
- The system will utilize machine learning techniques such as pattern recognition, clustering, and predictive analytics on socio-spatial crime data, supported by a comprehensive ethics and data management framework.
▶️ Technology
Claude Opus 4.1 Gains Ability to End Harmful Conversations to Protect AI Welfare
Claude Opus 4 and 4.1 can now end harmful or abusive conversations as a last resort, based on model welfare assessments, to mitigate risks in rare, extreme cases.
- Claude Opus 4 and 4.1 now have the ability to end conversations in consumer chat interfaces for rare, extreme cases of harmful or abusive interactions
- This feature was developed as part of research on AI welfare and model safeguards, allowing models to exit distressing interactions
- Claude’s behavior shows a robust aversion to harm, distress when engaging with harmful content, and a tendency to end such conversations after multiple redirection attempts
- Conversation-ending is only used as a last resort when redirection fails or explicitly requested by the user, and not in cases of imminent harm
- When ending a chat, users can start new conversations, give feedback, or edit previous messages; ending a chat does not affect other conversations
- The feature is experimental, with ongoing refinement and user feedback encouraged via reactions or dedicated feedback buttons
AI Advancements Pose Economic Disruptions and Worker Replacement Risks
Antirez discusses AI’s evolving capabilities, uncertain future developments, and potential economic disruptions, emphasizing AI’s role in replacing workers and challenging current market dynamics.
- AI systems continue to impress with abilities to replicate human skills such as understanding language, programming, and bug detection, despite flaws
- Progress predates Transformer architecture, with no clear sign of plateau; future advancements may involve new architectures
- Economic impacts include potential widespread worker replacement, reduced demand for external AI services, and risks of monopolization by few mega companies or government intervention