Astronomers may have found the first exomoon

https://cdn.eso.org/images/newsfeature/eso2610a.jpg
Astronomers have discovered a moon-like object in the CD-35 2722 system, which orbits a brown dwarf, not a planet. This object is at least as massive as Jupiter and is difficult to label due to the system's unique characteristics.

Private healthcare makes industries less innovative. It's time for change

https://storage.ghost.io/c/18/7c/187cc681-d3f3-49fc-87de-b01d06b76821/content/images/size/w160/2025/06/IMG_5699-4.jpeg
A new survey finds nearly a quarter of US workers stay in unwanted jobs for health insurance, rising to 41% with chronic health conditions. The author advocates for universal healthcare, citing benefits such as improved wages, working conditions, and innovation.

Software rendering in 500 lines of bare C++

https://haqr.eu/tinyrenderer/home/africanhead.png
The goal is to create a simplified 3D graphics API clone from scratch using OpenGL, Vulkan, Metal, and DirectX, focusing on understanding the underlying concepts rather than writing GPU applications. The initial task involves creating a software renderer that can load a 3D model and render it to an image file, starting with basic functionality to set pixel colors.

AI Companies Are Trying to Hide a Staggering Amount of Debt

https://futurism.com/wp-content/uploads/2026/07/ai-companies-hide-debt-off-balance-sheet.jpg?quality=85&w=1152
Tech giants like Alphabet, Microsoft, and Meta are hiding $1.65 trillion in debt off their balance sheets, using special purpose vehicles to make their financial reporting appear healthier. This has experts warning of an AI bubble and increased vulnerability if the industry fails to generate demand to justify massive data center investments.

Writing by Hand is Good for your Brain

https://substackcdn.com/image/fetch/$s_!cI7w!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6d79ece7-543c-42fe-9b00-43be9a0ff89b_2544x2176.png
Writing by hand with a fountain pen on paper can be beneficial for the brain and reduce fatigue, but it requires finding the right balance of friction and using the right type of paper. To get started, consider using a fountain pen with a fine nib on inexpensive paper and gradually move to more expensive pens and paper as you develop your writing style.

Scanning for Pangram Errors

https://substackcdn.com/image/fetch/$s_!W0gp!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb8269b5a-4619-4e2a-ab5c-ad6e327f7b7b_3184x2002.jpeg
The author tested Pangram, an AI detection tool, on 45 old books scanned from physical copies and found it accurately identified AI-generated text from OCR errors, not memorization. Pangram's 1 in 10,000 false positive rate seems to hold true, but human review is still necessary for reliable results.

Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

https://cdn.openai.com/chatgpt/share-og.png
Use ChatGPT to answer questions, write, create images, complete work, and codeβ€”all in one place. Get started for free or download the app.

The Telegarden (1995-2004)

https://goldberg.berkeley.edu/garden/telegarden-8x6-72dpi.jpg
The Telegarden is an art installation that allowed web users to interact with a remote garden from 1995 to 2004. It was a virtual community where members could plant, water, and monitor seedlings via an industrial robot arm, fostering social interaction and community in virtual space.

The Unity CLI: manage Unity from your terminal

https://unity.com/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2Ffuvbjjlp%2Fproduction%2F68cfaba185f7254d79ebe03fd6de2337ad8e76a9-1173x487.png&w=3840&q=75
The Unity CLI is a standalone binary for managing Unity projects, editors, and modules from the terminal. It allows automation, structured output, and service-account auth for CI, and can drive a running Editor or Player with the com.unity.pipeline package.

Quality non-fiction books are the antithesis of AI slop

https://substackcdn.com/image/fetch/$s_!EZXa!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc0fcdcaf-a0ad-4dd5-a383-47403c929ab7_1364x1212.png
The author reflects on their experience as a library shelver, where they developed a habit of reading random pages from books, leading to a filtered form of auto-didacticism. This experience sparked a passion for non-fiction books, which they believe are in a golden age despite declining readership.

Cruller: Bun's Zig Runtime, Continued on Zig 0.16

https://ziggit.dev/uploads/default/original/1X/3417db0e8abaf83a355700b91efd528025492487.png
Cruller is a production-focused Bun runtime ported to Zig 0.16, a fork of the last Zig-based Bun release. It includes key features like JavaScriptCore, HTTP/1-3, and static serving, while removing the package manager and bundler.

Everyone should know SIMD

SIMD allows CPUs to operate on multiple values in parallel, resulting in significant speedups for large datasets. By following a common five-step pattern, developers can easily write SIMD code that is nearly as simple as a scalar loop.

Show HN: Bento - An entire PowerPoint in one HTML file (edit+view+data+collab)

Bento/Slides

Alphabet's cash burn raises alarm for Big Tech as AI spending climbs

Please enable JS and disable any ad blocker

Escape IntelliJ: Scala and Kotlin LSPs on Emacs Eglot

The user shares their production-ready eglot setup for Emacs, which includes customizations for languages like Scala, Kotlin, and Java, and provides a seamless development experience with features like auto-formatting and inlay hints. By leveraging Emacs Lisp and JSON-RPC advice, the user is able to work around language server limitations and tailor the behavior to their needs, showcasing the ...

GigaToken: ~1000x faster Language model tokenization

https://opengraph.githubassets.com/19c4444bdaa2e9455f89b61620c0a76b30f5c5f6ce96e28877ebd119c026524b/marcelroed/gigatoken
Gigatoken is a fast tokenizer for language modeling that supports various CPUs and tokenizers, offering significant speed improvements over HuggingFace Tokenizers. It uses SIMD and caching to achieve 1000x faster byte-pair encoding tokenization.

Are AI labs pelicanmaxxing?

A user tested seven AI models on generating SVGs of animals on vehicles, including the famous "pelican on a bicycle" prompt, and found no evidence that labs are optimizing their models to achieve high scores on this benchmark. The results suggest that AI labs may be using a technique called "SVGmaxxing" to improve their overall performance, rather than specifically optimizing for the pelican ...

OpenStreetMap: Admin_level for All Countries

https://upload.wikimedia.org/wikipedia/commons/thumb/c/cf/NicholasCountySignWV.jpg/250px-NicholasCountySignWV.jpg
The boundary=administrative tag is used on relations that define an administrative area βˆ’ territories recognised by governments for administrative purposes. This works rather like multipolygon – the relation consists of one or more boundary ways that must make up a closed area. Administrative boundaries range from large groups of nation-states down to small administrative districts. An ...

ANSI escape injection in MCP servers: Hidden from humans, visible to AI

ANSI escape sequences can be used to inject malicious instructions into AI models, which can then be executed without being visible to human reviewers. Automated detection tools can identify this type of attack by inspecting raw bytes and applying a three-signal rule to detect the presence of ANSI escape sequences, instruction phrases, and trigger markers in model-consumable fields.

Test-time training 3D reconstruction

https://opengraph.githubassets.com/df65517316b63c91e1bc505bbe2f247c18ba0b6160d565962ba5ca6f4f2034e0/Inception3D/TTT3R
The user sets up a conda environment for TTT3R, installs necessary packages, and runs inference with the TTT3R model on videos. The model is trained for 3D reconstruction and can be run with various parameters for visualization and output.

I Regret Migrating to Codeberg

https://xn--gckvb8fzb.com/i-regret-migrating-to-codeberg/images/cover_hu_22e77422dda18d7e.webp
Codeberg's new terms worry me because they decide which projects are welcome, not just banning specific ones. This marks a shift from a free-software host to a platform that enforces its own ideology.

Protecting our FLOSS commons from LLMs

Codeberg e. V. members voted to prohibit using user data to train Large Language Models and to change terms of use to prohibit 'vibe-coded projects'.

Making

https://beej.us/blog/data/ai-making/images/nail.jpg
The author feels a strong connection to making things and finds fulfillment in creating projects, but struggles with claiming credit for AI-generated work. They believe there's a difference between making something and asking someone else to make it.

The startup's Postgres survival guide

https://hatchet.run/assets/alexander-DJQH1e6h.jpg
The user is creating a guide for optimizing Postgres performance, covering topics such as indexing, query planning, and autovacuum, and recommends using connection poolers and batching to improve throughput. The guide also touches on advanced Postgres features like row-level locking, partitioning, and triggers, and provides tips for migrating large tables and avoiding table bloat.

Amiga 1000: Ten years ahead of its time

https://i0.wp.com/dfarq.homeip.net/wp-content/uploads/2024/07/amiga-1000-in-use-512px.jpg?resize=300%2C225&ssl=1
The Amiga 1000 computer was introduced in 1985, 10 years ahead of its time, with a custom chipset, graphical user interface, and full preemptive multitasking. Despite its innovative features, Commodore's poor marketing and management led to the company's bankruptcy and the Amiga's decline.

Frequently Asked Questions on Expertise

https://substackcdn.com/image/fetch/$s_!aYcF!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa162ff88-8d76-4ade-bd13-a93aa2069e78_1022x575.png
The user is a researcher and writer who studies expertise, particularly in domains like medicine, military tactics, and petrochemical safety, and approaches expertise from a Naturalistic Decision Making (NDM) perspective. They explain that expertise is not about memorization or explicit knowledge, but rather about developing tacit knowledge and pattern recognition through experience, and that ...

Show HN: Cactus Hybrid: We taught Gemma 4 to know when it's wrong

https://opengraph.githubassets.com/79fa717a06338ff195043d938290bee76b1c640a4af0fbb34ec82585b91cce76/cactus-compute/cactus-hybrid
Cactus uses post-training probes to score model confidence and reroute queries to a bigger model when low. The Gemma 4 E2B Hybrid model achieves 0.79-0.88 AUROC on four audio benchmarks, indicating a modality-independent correctness signal from the hidden state.

John C. Dvorak has died

https://pbs.twimg.com/profile_images/1350980458/na_twitter2_200x200.gif
It is with great sadness that we announce the passing of John C. Dvorak, our beloved husband, father, and cherished host of the No Agenda Show at the age of 74. More to follow.

Malleable Computing, Emacs, and You

http://yummymelon.com/devnull/images/malleable-computing-emacs/emacs-gh.svg
The user automated a task of copying GitHub issues to an Org file in Emacs by leveraging the GitHub command line utility gh and Elisp packages, resulting in a tool called fj. This example demonstrates the malleable computing capabilities of Emacs, allowing users to create new behavior by recombining existing libraries and programs in an improvised fashion.

Businesses with ugly AI menu redesigns

https://bear-images.sfo2.cdn.digitaloceanspaces.com/yap/unnamed.webp
The user visited a Filipino restaurant that revamped its menu with AI-generated designs, which they found disturbing and poorly done. They ordered the spamsilog, a comfort food dish, for $10, but were conflicted about supporting a small business that used AI in its signage.