-
Total Annihilation, rebuilt for modern computers (412 points by jttnr)
Nanolathe is an open-source, MIT-licensed reimplementation of the Total Annihilation RTS engine for modern computers. It offers a browser-playable demo with three original missions and native beta installs for Windows, Mac, and Linux that use the player's own game files for the full game. The engine adds modern GPU rendering, lighting, glow, zoom, and an optional classic visual mode. It emphasizes documented formats, clean-room research, and moddability so developers can study and extend it.
-
Fingerprints are the new cookies (156 points by royfish)
The article argues that browser fingerprinting has become a powerful tracking method that survives clearing cookies, using a VPN, and sometimes private browsing. It demonstrates assigning visitors a persistent nickname based on device/browser characteristics, then explains how that identifier can link to accounts, ad profiles, and purchased data. The author warns that advertisers, data brokers, and governments can use such fingerprints for surveillance, often without meaningful consent. It frames privacy as a structural problem because combining persistent identifiers with “anonymized” data can re-identify people.
-
Whatever became of holography? (2011) (92 points by cainxinth)
This 2011 American Scientist article revisits the stalled promise of holography. It asks why holograms never became an everyday display or imaging technology despite decades of excitement. The article discusses technical and commercial obstacles such as coherent light sources, recording media, resolution, cost, and viewing requirements. It also notes the niches where holography persisted, including security, art, microscopy, data storage, and scientific visualization.
-
Cartographers of North Korea (2019) (205 points by edent)
Cartographers of North Korea is an interactive web project/visual essay about how one of the world's most closed countries gets mapped. It appears to combine narrative, maps, and likely satellite or remote-sensing sources to show the people and methods behind North Korean cartography. The Korean subtitle suggests a focus on mapmakers themselves, not just the resulting borders and terrain. It highlights maps as political and technical constructions shaped by secrecy, geopolitics, and limited access.
-
X Is Bad, Though (4 points by ayaros)
Lex on Tech's “X Is Bad, Though” responds to John Gruber's argument that X remains valuable despite Elon Musk. The author admires Gruber but disagrees that X's dominance, returning executives, and interesting users justify overlooking Musk's behavior or the platform's problems. It uses receipts to argue that platform success does not settle the ethical question of whether to use or support X. The piece explores the tension between network effects, utility, and moral/community responsibility on social media.
-
Calculus with Julia (22 points by subset)
“Calculus with Julia” is an open set of notes for learning calculus through the Julia programming language. It follows the “rule of four” by connecting graphical, numerical, algebraic, and verbal perspectives, using Julia mainly for numerical and computational exploration. The notes contrast Julia with symbolic computer algebra systems such as Mathematica, Maple, and Sage. They are designed for students who want to install Julia and work through calculus interactively.
-
Tattoos you can turn on and off (22 points by MortyDev)
The Live Science article describes new research into tattoos whose color can be turned on, off, or changed under different types of light. The advance relies on light-responsive chemistry or inks that alter their appearance in a reversible way. Potential applications include dynamic body art, medical or diagnostic indicators, anti-counterfeiting, and skin-mounted displays. It is a materials-science story about making tattoos programmable rather than permanent and static.
-
The gap in shared understanding of LLM capability is widening (305 points by albertzeyer)
Andrej Karpathy argues that the gap in shared understanding of LLM capabilities is widening into a sharp funnel. He estimates that around 6 billion people have barely encountered LLMs, while 1–2 billion are casual free-tier users who see derpy chatbots or better search/writing aids. Only about 20 million people (0.2%) professionally use frontier models for math and code, where agents can complete complex projects that once took weeks or months. This creates radically different intuitions about what AI can do, with public discourse and policy often shaped by people outside the frontier-use tier.
-
Illustration Beyond Proof (43 points by smilelamp)
“Illustration Beyond Proof” is a Terence Tao blog guest post by Kate Stange and Edmund Harriss about a workshop on mathematical illustration. The workshop asked what it would mean for an illustration to be rigorous and how computer graphics and other media contribute to mathematical meaning. The discussions led to an article tied to the Leiden Declaration and accepted in the Notices, amid broader debates about human mathematical practice and AI. It argues that visualization can go beyond decoration to become part of proof, explanation, and mathematical understanding.
-
Terence Tao: Math 2.0 [pdf] (346 points by Anon84)
Terence Tao's “Math 2.0” slides, presented at Caltech in 2026, address how computation, formal verification, AI, and large-scale collaboration are reshaping mathematical practice. The title signals a transition toward a new phase of mathematics where software, data, and machine reasoning are first-class tools. Based on Tao's recent work, the talk likely covers proof assistants, machine-assisted theorem proving, LLM reasoning, and human-AI collaboration. The slides are a high-profile signal that AI and formal methods are moving toward core mathematical infrastructure.
AI/ML Insights & Trends
- The LLM capability funnel is widening.
- Trend/insight: Karpathy's breakdown shows a tiny frontier-use tier, mostly math/code professionals using agentic models, a large casual tier using free chatbots, and billions with almost no LLM contact.
- Why it matters for AI/ML development: Benchmark discourse, safety policy, and product strategy can easily assume the wrong capability level, leading to mispriced risk, bad UX, and public misunderstanding.
-
Implications/actionable takeaways: Build tier-aware evaluations, separate casual chatbot metrics from agentic frontier metrics, invest in AI literacy, and design safety controls for both low-context and high-power users.
-
Mathematics is an early proving ground for AI co-reasoning and formal verification.
- Trend/insight: Tao's “Math 2.0,” “Illustration Beyond Proof,” and “Calculus with Julia” all point to math adopting computation, visualization, proof assistants, and AI-assisted reasoning.
- Why it matters for AI/ML development: Math offers verifiable ground truth, making it ideal for testing LLM reasoning, formalization, and multimodal explanation.
-
Implications/actionable takeaways: Prioritize AI4Math benchmarks, formal proof pipelines, human-in-the-loop verification, and tools that make AI-generated proofs and illustrations auditable.
-
Fingerprinting turns privacy into an AI/data-fusion problem.
- Trend/insight: Cookies are declining, but device/browser fingerprinting and linked data can persistently identify users, and ML accelerates matching and inference.
- Why it matters for AI/ML development: LLMs and data brokers can combine chat logs, behavior, and fingerprints, increasing deanonymization and surveillance risk.
- Implications/actionable takeaways: Adopt
Analysis generated by deepseek-reasoner