Published on September 08, 2026 at 06:02 CEST (UTC+2)
I've factored the RSA keys of a Certificate Authority from the 90s (155 points by ahlCVA)
The article describes how the author successfully factored RSA encryption keys used by a Certificate Authority from the 1990s, demonstrating that early web security relied on now-insecure key lengths (e.g., under 1024 bits). Using publicly archived browser root certificates and modest computing resources, they showed these outdated keys are easily breakable today. The piece contextualizes this within historical cryptographic standards, export restrictions of the era, and evolving security practices. It also touches on looming threats from quantum computing to even modern 2048-bit RSA keys.
TALA Is Open-Source (124 points by alixanderwang)
TALA, Terrastruct’s AutoLayout Algorithm for software architecture diagrams, has been open-sourced under the MPL-2.0 license. Unlike traditional DAG-based layout engines like Dagre or ELK, TALA uses an orthogonal layout approach that better mimics hand-drawn whiteboard diagrams. It incorporates multiple aesthetic objectives—symmetry, node clustering, flow—and supports hybrid layouts where some nodes have fixed positions. The author highlights its utility in AI-assisted diagram generation, where models can place nodes while TALA handles complex edge routing.
Jellyfin 12.0 (146 points by 0xC0ncord)
Jellyfin 12.0 is a major stable release focused on performance improvements stemming from a completed database conversion, along with long-awaited support for books and comics. The upgrade requires users to be on Jellyfin 10.10.7 or 10.11.x beforehand and insists on full manual backups due to irreversible schema changes on first boot. The release also introduces case-sensitive usernames and emphasizes careful pre-upgrade preparation to avoid data loss or migration issues.
Watch Los Angeles get built, one building at a time (1880–2026) (245 points by rustywasm)
This interactive visualization tracks the construction history of every existing building in Los Angeles County from 1880 through 2026, showing urban growth one structure at a time. Each building appears in the year it was built, though demolished structures are excluded, presenting only the surviving built environment. Users can explore by decade, height, and footprint, with exaggerated heights when zoomed out for better visibility, offering a compelling temporal-spatial view of LA’s development.
John Margolies' photographs of roadside America (43 points by duck)
The article showcases John Margolies’ photographic archive of American roadside architecture from the 1970s–2000s, now in the public domain via the Library of Congress. Margolies documented whimsical, vernacular structures like giant animal-shaped diners and eccentric motels that emerged alongside the interstate highway system. Contrary to critics who saw these as kitsch or blight, Margolies celebrated their grassroots creativity and cultural significance as expressions of local identity and entrepreneurial spirit.
Leaving VMware just got harder after Broadcom pulled VDDK downloads (129 points by josephcsible)
Broadcom has removed public access to VMware’s Virtual Disk Development Kit (VDDK), complicating customer efforts to migrate away from VMware’s vSphere platform. VDDK is a foundational library used by numerous third-party migration tools—including Microsoft Azure Migrate, Red Hat’s toolkit, and Nutanix Move—that enable workload transfers to alternative hypervisors or clouds. This move is seen as an anti-competitive tactic that imposes technical and logistical barriers on organizations seeking to exit VMware post-acquisition.
WeatherNext 3 (263 points by matthieu_bl)
Google DeepMind’s WeatherNext 3 is an advanced AI weather forecasting model that delivers hourly global predictions by directly processing raw satellite imagery. Unlike traditional numerical weather prediction models, it’s faster, more detailed, and integrated into Google products like Search, Maps, and Gemini. As an ensemble model, it provides high-resolution, actionable forecasts to help individuals and businesses make better decisions amid increasingly extreme weather events.
Trusting-Trust Attack against an Entire Linux Distribution (174 points by signa11)
Researchers demonstrate a practical “trusting-trust” attack—inspired by Ken Thompson’s classic compiler-backdoor concept—against an entire Linux distribution (NixOS) using binary manipulation of GNU strip, a common build utility. By tampering with a single binary in the bootstrap seed, they propagate a hidden payload through successive builds, ultimately backdooring nearly all binaries in a complete graphical installer. This proves such attacks aren’t limited to compilers and can compromise full OS distributions through seemingly innocuous tools.
Scientists observe Einstein's gravity in the quantum world (170 points by mudil)
Scientists at the University of Oxford report the first experimental observation of Einstein’s general relativistic gravity effects in a quantum system. While specific details aren’t provided in the preview, the breakthrough likely involves measuring gravitational interactions between quantum particles or superpositions, bridging two fundamental but historically incompatible theories of physics. This could pave the way for new insights into quantum gravity and unified field theories.
216M Spy TVs – The LG Smart TV Problem [video] (601 points by treve)
A viral video exposes serious privacy concerns with LG Smart TVs, alleging that over 216 million devices may function as covert surveillance tools. The investigation reveals how built-in microphones, cameras, and telemetry systems—often enabled by default—can collect and transmit user behavior without meaningful consent or transparency. The title “Spy TVs” underscores fears about consumer IoT devices harvesting sensitive data under the guise of smart features and ad personalization.
AI is increasingly embedded in operational infrastructure beyond chatbots
WeatherNext 3 exemplifies how AI models are becoming core components of essential services like weather forecasting, replacing or augmenting decades-old physical simulation approaches. This shift matters because it demonstrates ML’s transition from experimental or consumer-facing applications to mission-critical backend systems. Implication: Developers and organizations must prioritize reliability, interpretability, and integration robustness when deploying AI in infrastructure roles.
Human-AI collaboration is driving new UI and layout paradigms
TALA’s design explicitly accommodates generative AI agents that can position diagram nodes but struggle with edge routing—a nuanced division of labor between human-like spatial reasoning and algorithmic optimization. This reflects a broader trend toward “agentic workflows” where LLMs or multimodal models handle creative framing while specialized engines manage technical execution. Takeaway: Future tooling should expose modular, composable interfaces that let AI delegate subtasks to deterministic algorithms.
Security threats now target the ML and software supply chain itself
The trusting-trust attack against NixOS shows that AI/ML systems running on compromised OSes or toolchains could inherit undetectable backdoors. As ML pipelines depend on complex open-source ecosystems, integrity of the entire build stack becomes a prerequisite for trustworthy AI. Action item: Adopt reproducible builds, SBOMs (Software Bill of Materials), and binary provenance verification to harden ML deployment environments.
Legacy technical debt creates exploitable vulnerabilities as compute scales
The factoring of 1990s-era RSA keys illustrates how advancements in commodity hardware and algorithms continuously erode assumptions about cryptographic longevity. For AI systems that rely on secure data transmission, model signing, or federated learning, outdated crypto primitives pose real risks. Insight: ML practitioners must integrate crypto-agility principles and regularly audit dependencies for deprecated security practices.
Privacy erosion in consumer AI/IoT demands urgent regulatory and architectural responses
The LG Smart TV exposé highlights how pervasive sensing in AI-enabled devices enables mass surveillance under opaque consent mechanisms. As edge AI expands into homes and workplaces, current opt-in frameworks prove insufficient. Implication: Developers should adopt privacy-by-design—e.g., on-device processing, minimal data retention, and clear hardware indicators—to rebuild user trust and preempt regulatory backlash.
Generative media and interactive visualizations are becoming primary storytelling formats
Projects like the LA skyline timelapse and WeatherNext 3 leverage real-time rendering and geospatial data to communicate complex phenomena intuitively. In AI/ML, such visual narratives help demystify model outputs (e.g., forecast uncertainty, urban growth patterns). Trend: Invest in visualization literacy and interactive dashboards as essential complements to model development—not afterthoughts.
Open-source remains a double-edged sword for AI innovation and security
While TALA’s open-sourcing accelerates adoption and community scrutiny, the VMware VDDK takedown shows how dependencies on corporate-controlled OSS can create strategic vulnerabilities. For AI projects relying on open-source libraries, licensing stability and maintainer continuity are as critical as code quality. Recommendation: Diversify critical dependencies or maintain forks for high-risk components; advocate for truly neutral foundations in core AI infrastructure.
Analysis generated by qwen/qwen3-max