Hacker News Top 10
- English Edition

Published on March 20, 2026 at 18:01 CET (UTC+1)

1. VisiCalc Reconstructed (54 points by ingve)

   VisiCalc Reconstructed: This article details a personal project to rebuild a minimal clone of VisiCalc, the pioneering 1979 spreadsheet software. The author explains the core data model, formula evaluator, and simple UI needed, highlighting how VisiCalc's efficient design allowed it to run on limited 16K RAM machines. The piece serves as both a historical appreciation of a foundational "killer app" and a technical guide to understanding spreadsheet internals.

2. ArXiv declares independence from Cornell (605 points by bookstore-romeo)

   ArXiv declares independence from Cornell: This article reports on the arXiv preprint server's move to become an independent entity separate from Cornell University, which has hosted it since its inception. This shift aims to ensure the long-term sustainability and governance of the critical open-access repository for scientific papers, particularly in physics, computer science, and related fields.

3. Delve – Fake Compliance as a Service (58 points by freddykruger)

   Delve – Fake Compliance as a Service: This investigative article accuses "Delve," a compliance platform, of systematically fabricating audit evidence and misleading clients about their adherence to regulations like HIPAA and GDPR. It claims the company uses offshore "certification mills" and violates auditor independence rules, leaving customers exposed to significant legal and financial risk despite believing they are compliant.

4. The Los Angeles Aqueduct Is Wild (117 points by michaefe)

   The Los Angeles Aqueduct Is Wild: This educational article (a video transcript) explores the history and engineering of the Los Angeles Aqueduct, which transports water over 300 miles from the Sierra Nevada. It describes the aqueduct's pivotal role in enabling LA's growth, the dramatic moment of its 1913 opening, and its ongoing status as a vital yet controversial piece of infrastructure that reshaped California's water politics.

5. Entso-E final report on Iberian 2025 blackout (125 points by Rygian)

   Entso-E final report on Iberian 2025 blackout: This presents the final report from the European network of transmission system operators (Entso-E) on a major blackout in Spain and Portugal in April 2025. The report concludes the cause was a complex combination of voltage oscillations, reactive power control gaps, and uneven generator stabilization, leading to cascading failures. It issues recommendations for improved grid monitoring, coordination, and adapted regulatory frameworks to prevent future incidents.

6. Parallel Perl – autoparallelizing interpreter with JIT (14 points by bmn__)

   Parallel Perl – autoparallelizing interpreter with JIT: This presentation abstract discusses a project to integrate AI into the Perl programming language, specifically through an autoparallelizing interpreter with a Just-In-Time (JIT) compiler. The author, who has worked on AI with Perl for decades, frames this as "letting AI do Perl," suggesting the use of machine learning techniques to optimize and potentially transform code execution.

7. The Social Smolnet (48 points by aebtebeten)

   The Social Smolnet: This blog post proposes leveraging existing decentralized tools—blogs and email—as a social network, avoiding the need for new protocols. It introduces new "share" and "reply" commands in the Offpunk gemini/gopher browser that automate sharing links and contacting authors via email, arguing that this simple approach can recreate core social networking functions with established, user-controlled technology.

8. France's aircraft carrier located in real time by Le Monde through fitness app (50 points by MrDresden)

   France's aircraft carrier located in real time by Le Monde through fitness app: Based on the title and context, this article reveals a security breach where the real-time location of a French aircraft carrier was compromised due to data from a fitness tracking app (like Strava). It highlights how aggregated data from consumer wearables and apps can pose significant national security and operational security (OPSEC) risks by inadvertently revealing sensitive movements.

9. Video Encoding and Decoding with Vulkan Compute Shaders in FFmpeg (92 points by y1n0)

   Video Encoding and Decoding with Vulkan Compute Shaders in FFmpeg: This technical blog explains the integration of Vulkan compute shaders for video processing within FFmpeg, a major multimedia framework. It addresses the need for high-performance, flexible software encoding/decoding for professional workflows (like 8K film scans) where dedicated hardware may be insufficient or unavailable, marking a shift towards open, cross-platform GPU acceleration.

10. HP realizes that mandatory 15-minute support call wait times isn't good support (193 points by felineflock)

    HP realizes that mandatory 15-minute support call wait times isn't good support: This article reports on HP's now-abandoned policy of intentionally imposing a mandatory 15-minute wait message on support calls in several European countries, regardless of actual queue length. The stated goal was to push customers toward digital self-service options, but the policy was widely criticized as a misguided and customer-hostile approach to support cost reduction.

AI/ML Insights & Trends

1. Trend: AI for Systems Optimization & Legacy Modernization

   • Why it matters: Articles 1 (VisiCalc rebuild) and 6 (Parallel Perl) demonstrate a trend of applying AI/ML to understand, optimize, or reinvent foundational software systems. This moves AI from being just an application feature to a core tool for systems programming and compiler design.
   • Implication: We'll see more AI-assisted development tools that can refactor, parallelize, or translate legacy code, and more research into AI-driven JIT compilation and runtime optimization, potentially breathing new life into older languages and paradigms.

2. Trend: The "Data Integrity Crisis" in AI Training and Compliance

   • Why it matters: Articles 3 (fake compliance) and 8 (fitness app leaks) highlight a dual crisis: synthetic/fabricated data (Delve) and unintentionally leaked sensitive data (Strava). For AI, training on corrupted or synthetic data without provenance risks model collapse, while leaked data creates privacy and security nightmares.
   • Implication: There will be increased demand for AI-powered data provenance, verification, and synthetic data detection tools. Furthermore, privacy-preserving ML techniques (like federated learning) will become essential to learn from data without centralizing sensitive information.

3. Trend: Compute Evolution: From Specialized Hardware to Flexible GPU Acceleration

   • Why it matters: Article 9 (Vulkan in FFmpeg) showcases a move towards using general-purpose GPU compute (via Vulkan) for tasks traditionally handled by specialized ASICs. This reflects a need for flexibility and open standards in high-performance computing, which directly parallels trends in AI where frameworks seek to leverage diverse hardware beyond just NVIDIA's ecosystem.
   • Implication: The AI infrastructure stack will continue to diversify. Efforts like OpenXLA and SYCL will gain importance, allowing models to run optimally across different accelerators (GPUs, NPUs, custom ASICs), reducing vendor lock-in and spurring innovation.

4. Trend: AI for Critical Infrastructure Modeling and Resilience

   • Why it matters: Articles 4 (aqueduct) and 5 (power grid blackout) underscore the complexity and fragility of human-engineered systems. AI and ML are uniquely suited to model these complex, non-linear systems (water networks, smart grids) to predict failure points, optimize flows, and simulate disaster scenarios.
   • Implication: Growth in "digital twin" technology powered by AI for infrastructure. ML models will be deployed for real-time anomaly detection in grid operations and predictive maintenance of physical assets, making critical infrastructure more resilient to climate change and unexpected events.

5. Trend: The Decentralization Counter-Narrative in Social Tech

   • Why it matters: Article 7 (Social Smolnet) presents a minimalist, decentralized alternative to monolithic AI-driven social networks. It represents a growing skepticism towards centralized platforms that use user data for AI training and engagement algorithms, advocating for protocols that return control to users.
   • Implication: While large platforms dominate, there's fertile ground for AI applications that serve decentralized networks—think personal AI agents that manage communications across email/blog protocols, or on-device ML for content recommendation/filtering in a user-owned data context.

6. Trend: AI-Human Interaction and the Support Paradigm Shift

   • Why it matters: Article 10 (HP support) illustrates the failure of a blunt, automated gatekeeping strategy. The future lies in sophisticated AI that can genuinely resolve issues or seamlessly triage to humans, not in fake barriers. Customer support is a prime domain for LLM-powered agents, but the HP case is a cautionary tale.
   • Implication: Successful AI integration in customer-facing roles requires a focus on augmenting human agents and providing real, immediate value via chatbots (e.g., true problem-solving), not just deflecting contact. The measure of success shifts from "calls deflected" to "first-contact resolution" and customer satisfaction.

7. Trend: Open Science and AI Research Infrastructure

   • Why it matters: Article 2 (arXiv independence) highlights the enduring need for stable, community-governed infrastructure for knowledge dissemination. For AI/ML, this extends to preprint servers (like arXiv), but also to datasets, model hubs (like Hugging Face), and code repositories. Their independence and sustainability are critical for transparent, collaborative progress.
   • Implication: The health of the AI research ecosystem depends on supporting non-profit, open infrastructure. This trend will see increased discussion about funding models and governance for the platforms that host the foundational elements (papers, code, data, models) of AI innovation.

---

Analysis generated by deepseek-reasoner

Deutsch