Sunday — September 7, 2025
The US is urged to ban AI surveillance due to significant privacy risks, a portfolio website mimicking Windows XP has been created, and researchers have developed Contemplative AI, which incorporates principles from mindfulness and non-duality for more resilient AI systems.
News
AI surveillance should be banned while there is still time
The rise of AI chatbots poses significant privacy risks, as they can collect and exploit highly personal information, including thought processes and communication styles, to manipulate users commercially and ideologically. To mitigate these risks, it is essential to establish protected and anonymous AI services, as some companies like DuckDuckGo are already offering, and for Congress to enact AI-specific federal legislation to prevent the entrenchment of bad privacy practices.
Show HN: I recreated Windows XP as my portfolio
Mitch Ivin is a graphic designer who has created a portfolio website designed to resemble the Windows XP operating system, complete with a mock desktop, start menu, and taskbar. The site allows visitors to explore his work and experience his portfolio in a unique and interactive way, with various "applications" and links to his social media profiles and resume.
GLM 4.5 with Claude Code
GLM-4.5 and GLM-4.5-Air are two flagship models that leverage a Mixture-of-Experts (MoE) architecture and are designed for agent-oriented applications, with capabilities in reasoning, coding, and tool invocation. These models have achieved significant breakthroughs in performance, parameter efficiency, and cost-effectiveness, outperforming other models in various benchmark tests and demonstrating strong competitive advantages in real-world scenarios, particularly in tool invocation reliability and task completion rate.
GigaByte CXL memory expansion card with up to 512GB DRAM
GIGABYTE's website uses various cookies to enhance user experience, including necessary cookies for basic site functions, functional cookies for features like social media sharing, and analytical cookies to track visitor interactions. These cookies, set by GIGABYTE and third-party services like Google and LinkedIn, have varying durations and are used to store information such as user preferences, session data, and consent preferences.
OpenAI set to start mass production of its own AI chips with Broadcom
OpenAI is set to start mass production of its own AI chips with Broadcom, although the details of the article are not provided as it is locked behind a subscription paywall. The Financial Times website offers various subscription options to access the full article and other content, starting at $1 for a 4-week trial.
Research
Contemplative Artificial Intelligence
Traditional alignment strategies for artificial intelligence may be insufficient due to the complexity and unpredictability of intelligent systems, but incorporating principles from contemplative wisdom traditions, such as mindfulness and non-duality, can lead to more resilient AI systems. Implementing these principles, referred to as Contemplative AI, has shown promising results, including improved performance and cooperation in certain tasks, and may be further developed through active inference and embodiment.
How Developers Wield Agentic AI in Real Software Engineering Tasks
Software Engineering Agents (SWE agents) can struggle with complex real-world tasks, so they are designed to collaborate with developers, allowing for interactive problem-solving. A study of 19 developers using an in-IDE agent found that those who actively collaborated with the agent and iterated on its outputs were more successful in resolving issues, but faced challenges in trusting the agent's responses and collaborating on debugging and testing.
Reverse Designing Ferroelectric Capacitors with ML-Based Compact Modeling
Machine learning-based compact models are used to introduce two reverse-design algorithms that can quickly identify device parameters, such as layer thicknesses, needed to achieve specific electrical characteristics. These algorithms offer a substantial efficiency advantage over traditional methods, such as phase-field modeling, by significantly reducing computation time while maintaining accuracy.
The Future of Memory: Limits and Opportunities
Traditional large, shared memory architectures are limited by scaling and signaling challenges, so an alternative approach is proposed where memory is broken into smaller slices closely tied to compute elements. This design, leveraging 2.5D/3D integration, enables faster and more energy-efficient access to data through private local memory and in-package memory elements, while still using DRAM for larger datasets.
A Constructed Closure of the Bering Strait Can Prevent an AMOC Tipping
An artificial closure of the Bering Strait could potentially prevent the collapse of the Atlantic Meridional Overturning Circulation (AMOC) under climate change by extending its safe carbon budget. According to an Earth system model, a closure of the Strait can provide an additional carbon budget of up to 500PgC if done early enough, when the AMOC is still strong, making it a feasible climate intervention strategy.
Code
A Software Development Methodology for Disciplined LLM Collaboration
The Disciplined AI Software Development methodology is a structured approach to collaborating with AI on development projects, addressing issues like code bloat and architectural drift through systematic constraints and validation checkpoints. The methodology involves four stages - AI configuration, collaborative planning, systematic implementation, and data-driven iteration - to ensure consistent and maintainable code quality, with the goal of reducing debugging time and improving overall project efficiency.
Lisp Is Not Lists
ChrysaLisp is a 64-bit, MIMD operating system designed for parallel and distributed computing, featuring a self-hosting environment with its own toolchain, virtual processor, and graphical user interface. It supports various platforms, including macOS, Windows, Linux, and Raspberry Pi, and offers advanced features such as dynamic binding, object-oriented assembler, and a unique Lisp interpreter, making it suitable for developing complex distributed computing systems.
Show HN: I Built Open Source Claude Code Alternative, but Better at API Testing
Ani Code is an open-source alternative to Claude Code, designed for better API testing, providing features such as QA testing, security scans, API exploration, and persistent agent memory. It supports multiple AI providers, including OpenRouter, OpenAI, Anthropic, and Google Gemini, and offers a range of tools and commands for testing, security, and development, all wrapped in a user-friendly interface.
Anti AI/LLM scraper bot firewall
Fantasma Cero is a web firewall that checks incoming HTTP requests to distinguish between human visitors and AI scraper bots, using methods like Proof-of-Work challenges and token verification. Once configured, it allows only verified visitors to pass through to the backend, while deterring or deceiving AI scrapers with techniques like trap endpoints and decoy content.
Using Small Language Models to Predict Psychological Vulnerabilities in Cybersec
The Cybersecurity Psychology Framework (CPF) is a psychoanalytic-cognitive model that identifies pre-cognitive security vulnerabilities, addressing the human factors that contribute to 85% of breaches. The framework consists of 100 indicators across 10 vulnerability categories, utilizing ternary scoring and attack vector mapping to assess risk, and provides a foundation for understanding and mitigating psychological vulnerabilities in cybersecurity.