Published on July 28, 2026 at 18:01 CEST (UTC+2)
7.1 Earthquake in Japan (626 points by krembo)
7.1 Earthquake in Japan – This article reports a magnitude 7.1 earthquake that struck Japan, as recorded by the Japan Meteorological Agency. The page provides detailed seismic intensity data by prefecture and municipality, along with a map replicated with permission from the Geospatial Information Authority of Japan. The high score (626 points) reflects the immediate news value and public interest in natural disasters.
New HIV vaccine shows unprecedented success in preclinical study (263 points by codebyaditya)
New HIV vaccine shows unprecedented success in preclinical study – Scientists at La Jolla Institute for Immunology, Scripps Research, and IAVI have developed an HIV vaccine that generates high numbers of broadly neutralizing antibodies in non-human primates. The vaccine trains the immune system to overcome HIV’s defenses, achieving the best antibody response ever seen in primates. Human trials have already begun, marking a major milestone in HIV prevention.
Harmony Explained: Progress Towards a Scientific Theory of Music (2012) (17 points by surprisetalk)
Harmony Explained: Progress Towards a Scientific Theory of Music (2012) – This arXiv paper by Daniel Shawcross Wilkerson proposes a scientific theory of music harmony, deriving the Major Scale, Standard Chord Dictionary, and the difference between Major and Minor triads from physical and computational principles. It critiques traditional music theory as superstitious and aims to explain harmony through the lens of human perception and machine processing of sound. The paper claims original derivations for chord dictionaries and triad emotions.
Kimi Linear: An Expressive, Efficient Attention Architecture (145 points by ronfriedhaber)
Kimi Linear: An Expressive, Efficient Attention Architecture – The Kimi team introduces a hybrid linear attention architecture that outperforms full attention in fair comparisons across short- and long-context scenarios. This architecture aims to improve efficiency while maintaining expressiveness, challenging the dominance of standard softmax attention in large language models. The paper represents a notable advance in making transformer models more computationally scalable.
Show HN: Formally verified 3D CSG: Trust 93 lines spec, not 1000 lines AI code (69 points by permute)
Show HN: Formally verified 3D CSG: Trust 93 lines spec, not 1000 lines AI code – This project implements a formally verified 3D mesh intersection operation for constructive solid geometry (CSG) in Lean 4. The author emphasizes that only 93 lines of specification need to be trusted, contrasting with the opaque nature of AI-generated code. It claims to be the first formally verified CSG operation, highlighting the value of mathematical proof over statistical code generation.
Europe's Ultra-Rich Could Fund a Substantial Part of the EU's Budget (27 points by robtherobber)
Europe's Ultra-Rich Could Fund a Substantial Part of the EU's Budget – This policy analysis argues that taxes on ultra-high net worth individuals (over €100 million), crypto assets, and financial trades could generate significant revenue for the EU budget. It criticizes current reliance on labour-based contributions and proposes EU-level levies to capture mobile wealth. The piece presents revenue estimates and discusses political feasibility.
Solving Fermat: Andrew Wiles (36 points by 1970-01-01)
Solving Fermat: Andrew Wiles – A NOVA interview with Andrew Wiles recounts his lifelong obsession with Fermat’s Last Theorem, which he first encountered as a ten-year-old. Wiles describes his 1993 announcement of a proof, the subsequent discovery of an error, and his eventual successful correction. The story underscores the interplay between mathematical persistence, genius, and public scrutiny.
Italy Blocks Reproductive Health Websites Women on Web and Women Help Women (21 points by miniBill)
Italy Blocks Reproductive Health Websites Women on Web and Women Help Women – The Open Observatory of Network Interference (OONI) reports that at least six Italian ISPs began blocking access to Women on Web and Women Help Women in February 2026 using DNS tampering. These sites provide abortion information and reproductive health care access. The blocking was detected via OONI Probe tests, with bogon IP addresses used for censorship.
GrapheneOS Defends Data-Wiping Function That Blocked US Border Search (46 points by pseudolus)
GrapheneOS Defends Data-Wiping Function That Blocked US Border Search – GrapheneOS, a privacy-focused Android OS, defends its data-wiping feature that prevented a US border agent from searching a device. The feature automatically erases data after repeated failed unlock attempts, which activists argue is essential for protecting sensitive information at checkpoints. The article highlights tensions between border security powers and digital rights.
DMARC Has Been Public Since 2012. 68.4% of Domains Still Don't Enforce It (82 points by adulion)
DMARC Has Been Public Since 2012. 68.4% of Domains Still Don't Enforce It – An analysis of 67,336 domains found that only 29.7% of those with DMARC records enforce a reject policy, while 45.1% lack any DMARC record at all. Despite being a free, standard email authentication protocol for over a decade, most domains still fail to prevent spoofing. The report calls for industry-wide adoption to reduce phishing and email fraud.
Formal Verification as a Counterpoint to AI-Generated Code – The 3D CSG project explicitly contrasts a 93-line formal specification with “1000+ lines of AI-written code,” signaling a growing skepticism about relying on LLM-generated code for safety-critical tasks. This trend matters for AI/ML development because it highlights the need for rigorous validation methods alongside generative models. Potential implication: increased investment in tools that combine formal verification with AI assistance, such as Lean Copilot or proof synthesis.
Linear Attention Architectures Gain Ground – Kimi Linear’s claim of outperforming full attention under fair comparisons indicates that efficient attention mechanisms (e.g., linear, hybrid, or state-space models) are maturing. This is crucial for scaling LLMs beyond current compute limits, enabling longer contexts and lower inference costs. Expect more production deployments and further research into hybrid architectures that blend linear and softmax attention.
AI in Scientific Discovery Crosses into Immunology – The HIV vaccine success relies on computational modeling of immune responses (epitope design, antibody evolution), a domain where AI/ML is accelerating vaccine and drug development. This trend matters because it demonstrates that AI can directly contribute to solving long-standing medical challenges, not just optimize existing processes. Implication: biotech firms will increasingly adopt generative models for antigen design and trial simulation.
DMARC Adoption Gap Highlights ML’s Role in Cybersecurity – The 68.4% enforcement gap shows that even simple, rule-based standards are poorly deployed; this creates fertile ground for ML-based anomaly detection and automated policy management. For AI/ML development, this suggests a market for tools that analyze email flows and recommend DMARC configurations, or for models that detect spoofing in real time. The implication: AI can help bridge the human adoption gap through automated compliance and threat detection.
The “AI Code vs. Verified Code” Debate Reflects Broader Trust Issues – The tension between trusting AI-generated code versus formally verified specifications (article 5) mirrors broader concerns about reliability, accountability, and explainability in AI. This is a key trend as AI coding assistants become widespread—developers must weigh speed against correctness. Potential takeaway: formal methods may become a complementary layer for safety-critical AI applications.
Censorship Detection Leverages Open Data and Probes – OONI’s report on Italy’s DNS blocking uses distributed measurements to track internet censorship, a technique that can be augmented with ML to detect patterns or predict new blocks. For AI/ML, this trend points to using anomaly detection on network measurements to identify censorship events in near real-time. It also raises ethical considerations for AI tools used to bypass or counteract censorship.
Wealth Tax and Crypto Tax Proposals Have Indirect AI Implications – While not directly AI/ML, the policy discussion on taxing ultra-wealth and crypto transactions could affect AI funding: large tech companies and investors (often targets of such taxes) might redirect resources from R&D. Additionally, AI-driven financial analytics could be used by governments to identify taxable wealth or by crypto traders to minimize liabilities. The trend matters because AI development is capital-intensive, and changes in tax policy could shift investment landscapes.
Analysis generated by deepseek-reasoner