HN Daily | July 23, 2026

Today's tech landscape is a fascinating mix of the analog and the digital, the philosophical and the practical. We have celebrated authors defending the fountain pen, astronomers spotting a potential exomoon, and the U.S. Air Force flying an AI-controlled F-16. Meanwhile, the debates around open-source AI, the economics of LLMs, and the sheer speed of new tools like GigaToken show that the software world is moving faster than ever. Let's dive in.

AI & Machine Learning

1. Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample — A fascinating look at a world-class mathematician using ChatGPT to digest a complex counterexample to the Jacobian Conjecture. It's a powerful demonstration of LLMs as a thinking partner, not just a code generator.
2. Startup founders urge U.S. government not to shut off Chinese open weight AI — A coalition of founders argues that restricting access to open-weight AI models from China would harm American innovation. The letter highlights the tension between national security and the open-source ethos that drives much of tech.
3. Understanding the AI Economy — Google's ATLAS study analyzes 15 million AI interactions to map how people actually use AI at work and home. Key finding: AI use is broad but shallow, with most interactions being collaborative rather than fully automated.
4. Are AI labs pelicanmaxxing? — A clever investigation into whether AI labs are secretly training their models to ace Simon Willison's famous "pelican riding a bicycle" SVG benchmark. The evidence suggests they aren't, but the methodology is a joy to read.
5. The arguments against open source AI are bad — A spirited defense of open-source AI, arguing that attempts to suppress it are historically doomed and that the real threat is not open models but centralized control. A must-read for anyone following the policy debates.
6. Protecting our FLOSS commons from LLMs — Codeberg, a European Git hosting platform, has voted to prohibit "vibe-coded" projects and promise not to train LLMs on user data. It's a principled stand against the resource drain and philosophical shift that LLMs represent for open-source communities.

Open Source & Tools

1. GigaToken: ~1000x faster Language model tokenization — A drop-in replacement for HuggingFace tokenizers that is up to 1000x faster. If you process large amounts of text for LLMs, this is the kind of performance leap that changes your infrastructure costs overnight.
2. Building on ATProto — A thoughtful critique of Bluesky's AT Protocol from someone who wants to build a decentralized review app. The author finds the protocol promising but worries it's not on track to become the open standard we need.
3. git's –end-of-options Flag — A deep dive into a little-known Git flag that solves a real security problem: argument injection. The history of how -- and --end-of-options differ is a great lesson in the unintended consequences of design decisions.
4. The startup's Postgres survival guide — A practical, battle-tested guide to running Postgres in production, from schema design to query planning to autovacuum. Every startup engineer should bookmark this.
5. Software rendering in 500 lines of bare C++ — A classic tutorial that builds a software renderer from scratch, teaching how OpenGL and Vulkan actually work under the hood. Perfect for anyone who wants to understand graphics programming at a fundamental level.

Science & Research

1. Astronomers may have found the first exomoon — ESO's VLT has detected what could be the first moon outside our Solar System. The catch? It orbits a brown dwarf, not a planet, blurring the lines between moon, planet, and star.
2. A solid-state “atomic channel” for separating rare earth elements — Researchers have developed a cleaner, electrochemical method to separate rare earth elements using layered manganese oxide. This could make the production of magnets, LEDs, and EVs much more environmentally friendly.
3. DARPA, U.S. Air Force fly AI-controlled F-16 — The VENOM program has successfully flown an AI agent piloting a standard F-16. The goal is to develop trusted autonomous combat AI, with human pilots remaining in the loop to monitor and override.

Business & Startups

1. What happened to TheNumbers.com — The definitive film industry data site was taken down by a bot-driven DDoS attack, likely from AI training crawlers. It's a cautionary tale about how the web's infrastructure is being strained by the AI gold rush.
2. Show HN: Palmier Pro – Open-source macOS video editor built for AI — A native macOS video editor that integrates AI generation directly into the timeline. It's open-source and lets Claude or Codex handle the grunt work of editing, from rough cuts to transitions.

Culture & Commentary

1. Writing by hand is good for your brain — Neal Stephenson, who has written thousands of pages by hand, explains why the practice recruits more of your brain and offers practical tips for avoiding writer's cramp. A beautiful defense of analog tools in a digital age.
2. So Reddit has decided that plain HTML is unsafe — Reddit now requires a login to view old.reddit.com, citing "safety" concerns. The author argues this is really about preventing scraping and controlling access to user-generated content, not security.
3. Making — Beej (of Beej's Guide fame) reflects on the joy of making things by hand, whether it's code, woodworking, or art. He argues that using AI to generate things at your request robs you of the fulfillment that comes from true creation.

History & Engineering

1. The Beam Engine — An interactive, beautifully illustrated guide to how steam engines powered the Industrial Revolution. Starting from first principles, it explains steam, pistons, and Newcomen's engine with hands-on figures you can manipulate.

Closing Thought

Today's stories remind us that technology is never just about the new shiny thing. Whether it's the quiet satisfaction of a fountain pen on paper, the raw power of a beam engine, or the dizzying speed of an AI tokenizer, the best tools are the ones that make us think more deeply about what we're building and why. See you tomorrow.