The model represents a significant breakthrough in making powerful AI coding assistants accessible to individual developers without relying on expensive cloud APIs or subscriptions.

Identical pendulums, released almost together, that refuse to stay in step.

MicroDuck biped PPO policy running as an ArduPilot (ArduRover) task, MuJoCo as SITL plant, MAVProxy control.

Rust terminal UI for working with Git history from a topology-first point of view. It is built with ratatui, crossterm, and libgit2, and is designed for exploring, filtering, and operating on large repositories without leaving the terminal.

To remove a background, drop a photo here and the subject is cut out as a transparent PNG. Free, with no watermark, and the AI runs on your device, so the photo is never uploaded.

I asked this question to AI image generators for each country: Midjourney in 2023, and Grok Imagine, GPT Image and Gemini in 2026, separately for a man and a woman. Pick the model at the bottom of the map.

• AI-Generated: These images are generated by AI.
• Not Representative: They do not represent the diversity and complexity of each country.
• Just for Fun: This is a fun exploration of how AI sees the world.

Timestral, Mistral’s first family of time-series foundation models, delivers state-of-the-art probabilistic, multivariate forecasts with just 209M parameters, outperforming larger models on key benchmarks while supporting past and future covariates. Available in four sizes (5M to 209M parameters), it balances accuracy and efficiency, running efficiently on CPUs and excelling in zero-shot forecasting for diverse domains. The model combines synthetic pre-training, covariate-aware attention, and long-context learning, with fine-tuning options for domain adaptation via Mistral Forge.

How a broken pascal compiler, a 300 baud modem, and a 1970s terminal helped create the (in)famous text editor.

A lab releases a better model. The weights are public, the benchmarks look good, and developers start figuring out what they can build with it.

Then they check the memory requirements.

I think this is going to become one of the harder problems for the open-source AI community. Models can become more efficient at generating tokens while becoming harder for individuals and small teams to host. We could gain access to increasingly capable weights while losing the practical ability to run them ourselves.

The rapid integration of AI code agents tools like Claude, Codex, and various internal LLM deployments into our development workflow promises unparalleled productivity. These agents, capable of reading, analyzing, and modifying complex codebases, are revolutionizing how we build software.

However, this power comes with a critical security question: How do you prevent an autonomous, powerful AI agent from looking at your secrets, reading unrelated projects, or even accessing sensitive files outside the scope of the project you asked it to work on?

In his ground breaking 2011 book, “Thinking Fast and Slow”, Daniel Kahneman introduced the idea that your brain has two different modes of thinking, what he called System 1 and System 2. System 1 is the auto-pilot: fast, automatic, instinctive, and effortless. It runs continuously in the background and makes judgement calls: he looks angry, the road is blocked, this sounds off, etc.

System 2 is the opposite. Slow, deliberate, thoughtful, rational. It follows the rules, makes calculations, explains itself, and pays attention. Unfortunately, attention is limited and System 2 is lazy so it often follow the lead of System 1.

The current generation of LLMs, like Claude, ChatGPT, Gemini, Grok, etc fall cleanly into the System 2 category. They work through the prompt systematically to figure out what to output. They excel at producing long form text, work through complex problems, translate across multiple languages, etc. They fall neatly into the System 2 category.

Despite these strengths, many of the pipelines that leverage LLMs in production revolve around quick decision making. Which team should handle this ticket? Is this summary correct? What tool should be called? System 1 type problems.

PDFCube is a PDF presentation viewer with animated transitions, automatic slide timing, and keyboard-controlled zoom. It brings cube rotations, page flips, shattering glass, and smooth fades to slide decks from LaTeX Beamer and office suites.

A website that was dedicated to the human endeavor of reverse engineering, “decompiling,” and modifying video games is shutting down because people using AI to clone and decompile games have fully overrun the community and are far outpacing what human software engineers alone can do. The news comes as AI-coded “clones” of popular enterprise software like the Adobe creative suite and Microsoft Office have gone viral and are seemingly popping up every day, and as people in video game modding communities warn that 'AI decomps' are everywhere.

If AI is so great, why aren’t we seeing an explosion of new, amazing software products? If anyone can turn their idea into working software with coding agents, where is all that software? People have been asking this question for years. As far as I can tell, the answer is that it’s spread out over a lot of modestly successful projects.

“It’s just timestamps”: the decades-old mystery of the origins of one of the most popular cryptographic standards has been explained.

We pursued this long-standing open question of the cryptographic community to settle concerns about ECDSA, which is one of our most widely used cryptographic standards.

Readers should take this as a fun closing chapter and the end of years of speculation. There is no action to be taken or things to do differently. Most researchers trusted the curves already. Enjoy.

TreeTime is a general purpose information manager.

Trustmux lets you monitor and interact with your tmux or Byobu sessions from any browser — privately, over your Tailscale network. No relay server. No cloud. Point to point. Your data is always private and TLS encrypted.

From mechanical design and electronics to firmware, control and learning.
Built to be understood. Open to be built upon.

Tool-augmented language models are bounded by the APIs humans bothered to write; existing tool-creation systems patch this by prompting a frozen LLM at inference time, leaving the model that writes a tool decoupled from the one that uses it, with no signal that the schemas it produces are schemas it can actually invoke. We propose SMITH (Schema-grounded Multi-task Iterative Tool Honing), a reinforcement learning framework that jointly trains tool creation and tool use inside a single policy.

Minecraft: Java Edition 26.3 rewritten from scratch in Rust, checked against vanilla with 281 differential harnesses. Get it with the Horizon launcher.

Where directories are code and drive space is infinite.

OpenSSH creator Theo de Raadt was pushed out of NetBSD, lost a DARPA grant and fought half of open source. Now his paranoid code guards the world's servers.

grep for images – local, offline image search for the terminal.

Flyline is a readline replacement that provides an enhanced line editing experience.

Terminal-first calendar, todo, and journal manager with iCalendar (RFC 5545) support and CalDAV sync — CLI + TUI, local SQLite storage.

I live in two places: a terminal running Claude Code and Neovim. For months I bounced between them. Edit in Neovim, alt-tab to the terminal, paste a file path, ask Claude something, alt-tab back, apply the change by hand. The friction was small but constant, and constant friction is the kind that wears a groove in your day.

Image editing, vector illustration, video, photography, PDFs, motion graphics, page layout, interface design, word processing, audio, presentations, CAD and spreadsheets. Native, open-source apps from the ArtCraft team, built in Rust and free to use.

CLIs are a fantastic way to build products. Unlike web applications, they take a small fraction of the time to build and are much more powerful. With the web, you can do whatever the developer programmed. With CLIs, you can easily mash-up multiple tools together yourself to perform advanced tasks. They require more technical expertise to use, but still work well for admin tasks, power-user tasks, or developer products.

A curated collection of open-source TUI apps for developers, operators, and power users.

Reverse.horse is my newest project, an attempt to understand and explain the new AI model Jev, by using your feeble human brain as the model. Place yourself in-request and respond as best as you can to incoming questions.

If you want to jump right in, go try it. The horse will put you through your training on how to be the AI model and then cut you loose to do the job of the AI model. You may come back with a new appreciation of what these models are doing.

The UEFI Shell command set, rewritten from scratch

Files, disks, UEFI variables, drivers and devices, PCI/SMBIOS/ACPI, memory, network, editors. UEFI Shell options are accepted too. More than 80 commands are documented in the user manual.

cosh doesn't impose a philosophy: attach or detach its built-in tools and choose your model to shape the agent you want. Works with Claude, OpenAI, Gemini and many more providers. Boots in milliseconds.

I have been using DeepSeek 4.1 Flash for about a month, heavily, across a dozen projects. It is super capable, and orders of magnitude cheaper than the "frontier" models. When I'm mid-session, if I don't look at the model name, I honestly could not tell you if I'm using DeepSeek or Opus. Whether it's our conversations, the work, or the speed, I don't notice a difference. I don't care that there is no 4.1 "Pro". I treat this like a frontier model because it behaves like one. I'm coming at this from my subjective usage experience, but you can see more complete benchmarks here if that floats your boat.

Flyline: a Bash plugin to replace readline for a modern line editing experience: syntax highlighting, agent integration, rich prompts, tooltips, fuzzy history search, and more!

Securing and modernizing software is a top priority today. Tech debt and memory-safety bugs remain the largest sources of vulnerabilities, which is why teams increasingly turn to Rust for its reliability and memory safety. But rewrites are hard, scary, and carry a lot of risk. Mainmatter assembled our experience from helping clients successfully modernize some of the worlds most important software into this book.

When we made our strategic bet on Rust back in 2022, we focused on Rust for backends and cloud systems. At the time, that was not an obvious choice. Rust was mostly seen as a systems language only, a safer replacement for C and C++ in operating systems, embedded devices, or browser engines. Suggesting to use it also for the kind of web backends that teams would typically build with Rails, Django, or Node.js required some explaining.

Let’s debunk the myth of Vim and learn how it’s possible to remember all the shortcuts using the specific Vim language. We’ll see how to move with vim motions, and I’ll share what I’ve learned so far, and why you might give Vim a try as well.

The internet allowed us to publish and link-together an enourmous amount of information, how to find anything in this vast ocean of interlinked documents is still a problem. To build a better internet we need to work on its content's discoverability.

500 things you can make with JavaScript and a single canvas element. No WebGL, no libraries, just a 2D drawing context and a loop. Open any demo to play with it, then copy a prompt that gets your coding agent started on your own version.

Every image a computer shows you, whether it's a game, a film effect, a map or the text you're reading right now, ends up as a grid of colored dots. This book is about how those dots get their colors.

We start with the simplest thing there is: setting one dot. From there we draw lines, fill shapes, mix colors, and then step into 3D, building a complete renderer from scratch in plain JavaScript, followed by a ray tracer and a path tracer that follows light the way physics says it really behaves. Only when every stage is familiar do we hand the work to the GPU, the specialized chip built to do exactly what you've just written, millions of times faster. The next parts chase the path tracer's realism in real time and bring things to life with animation and simulation. The last part covers the techniques behind today's real-time renderers, from temporal antialiasing and clustered lights to volumes, ray tracing in compute shaders, and scenes learned from photographs.

This past year, OpenAI, Anthropic, and other labs have announced breakthroughs on numerous long-standing mathematical problems, in some cases pushing well beyond what researchers expected current systems to be capable of — including resolving one of the famous Millennium Prize problems. But in classic Silicon Valley style, AI labs are moving fast and breaking things, barreling through the discipline with all the grace of a runaway bulldozer. Results that might normally have been celebrated have instead sparked backlash. AI labs say they are learning from earlier mistakes. Whether those promises bear fruit remains to be seen.

Something shifted over the last year or so, and it's been hard to put into words.

Open a PR on a popular repo today and observe what happens. Within seconds, a bot leaves an AI review summarizing your own change back to you. Another bot posts a preview deployment. Another posts a "walkthrough" with a sequence diagram nobody asked for. Sometimes a fourth shows up to respond to the third. By the time a human actually looks at it, the thread is a wall of generated text, and the real conversation (the human part) is buried somewhere in the middle.

The details vary, but the shape is the same: maintainers are no longer just deciding whether a contribution is good, they’re deciding whether they can afford to find out.

There are not many computer file formats I would trust to still be readable fifty years from now, but plain text is one of them. That may sound like faint praise. A text file cannot embed a spreadsheet, preserve elaborate page layouts, run a presentation, or provide many of the conveniences we expect from modern applications. What it does instead is store text in one of the simplest and most widely understood forms in computing, and that simplicity is a large part of why it has lasted.

Ansi.md is my brain dump for creating modern command line apps. My intent is to gradually expand this guide with hard-won tricks, recommended apps and homegrown tools.

You have to split your code so numerically intense parallel routines are offloaded to the GPU and serial work remains on the CPUs. The GPU is a distinct device, and you have to move data back and forth between these devices. The latter takes energy, the former takes money. Perhaps with GenAI coding assistants, this hybrid computing model can be made easier to implement.

Stop for one second and ask yourself a simple question. Where do your words come from?

When you speak, what comes first, the idea or the word? Do you first feel a thought inside you, and only after that go searching for the right word to wrap around it? I think we all do. The word is never the start. The word is just the skin. The idea, the consciousness, is the thing sitting under it.

Now ask the same question about an LLM. For an LLM, it is exactly the opposite. And I think this one small difference explains almost everything about where we are heading.

• User wants to do X.
• User doesn't know how to do X, but thinks they can fumble their way to a solution if they can just manage to do Y.
• User doesn't know how to do Y either.
• User asks for help with Y.
• Others try to help user with Y, but are confused because Y seems like a strange problem to want to solve.
• After much interaction and wasted time, it finally becomes clear that the user really wants help with X, and that Y wasn't even a suitable solution for X.

PTY-driven test framework for bash/shell scripts and TUI applications.

We show you colors. You recreate them from memory. Challenge friends to beat your score. It's harder than you think. Play free at dialed.gg.

Weekend at Bernie’s showed that a good chunk of the most-depended-on open source packages are dead, and there are a lot of different ways for a project to end up that way.

In the event of an imminent nuclear apocalypse, we suspect that many people who have access to private jets will immediately take to the skies and escape city centers. This site tracks this indicator in realtime. The current emergency level is reported on a scale of 1 to 5, with 5 being an indicator of a likely imminent apocalypse.

Email is like those creaking old Terminators from the ’70s which continue to function without complaining. Designed for a world that doesn’t exist anymore, it has optional encryption, no built-in auth, three⁺ retrofitted security layers bolted on top, an unstandardized filtering layer and many more quirks. Yet billions of emails arrive correctly every single day.

Email is not elegant but nonetheless it is Lindy7. In the new age of agentic AI8, we can only expect it to metamorphose into another dimension.

In April 2026, Andrej Karpathy published a GitHub gist. Not code. Not a library. A markdown document describing a pattern: an LLM-maintained folder of wiki pages that compounds across sessions and beats RAG.

LuaJIT finishes a computational benchmark in 23.29 seconds. C finishes in 22.29. That's a 4.5% gap between a dynamically typed scripting language and the fastest compiled language on earth. Python, doing the same test? 416.55 seconds. Eighteen times slower.

I thought I was seeing fewer arXiv papers on the front page of Hacker News (HN) these days, and I wanted to check if that was real.

So I asked Claude to run a quick analysis: track the share of arXiv stories on HN over time. It queried the BigQuery HN dataset, bucketed the stories by month.

This is the 16th year we’ve been teaching the Stanford Lean LaunchPad class. This year, from the first hour of the first class, we realized we were seeing something extraordinary happen. It was both the end and beginning of a new era.

Teams showed up to the first day of class with MVPs (Minimal Viable Products) looking like finished products that previous classes had taken weeks or months to build. After the class, as the instructors sat processing what just happened, we realized there’s no going back.

I’ve been writing about how AI is going to change startups, but the shock of seeing 8 teams actually implementing it was mind blowing. And not a single team thought they were doing anything extraordinary.

I’ve been getting more and more curious about the risk from Anthropic’s Claude Mythos Preview. So I pulled the system card, a whoppingly inefficient 244-page document that devotes just seven pages to the claim that the model is too dangerous to release. In fact, the 23MB of PDF I had to download was 20MB of wasted time and space. Compressing the PDF to 3MB meant I lost exactly nothing.

Foreshadowing, I guess.

Spoiler alert: the crucial seven pages out of 244 do not contain the word “fuzzer” once. That’s like a seven page vacation brochure for Hawaii that leaves out the word beaches.

Also, the crucial seven pages out of 244 do not contain the expected acronyms CVSS, CWE or CVE, they do not have comparison baseline, an independent reproduction, or the word “thousands.” I’ll get back to all of that in a minute.

The flagship demonstration document turns out to be like the ending of the Wizard of Oz, a sorry disappointment about a model weaponizing two bugs that a different model found, in software the vendor had already patched, in a test environment with the browser sandbox and defense-in-depth mitigations stripped out. Anthropic failed, and somehow the story was flipped into a warning about its success.

An autonomous robot ping-pong player dubbed Ace has achieved a milestone for AI and robotics in Tokyo by competing against and sometimes defeating top-level human players at table tennis, a feat that could presage an array of other applications for similarly adept robots.

Ace, created by the Japanese company Sony's (6758.T), opens new tab AI research division, is the first robot to attain expert-level performance in a competitive physical sport, one that requires rapid ‌decisions and precision execution, the project's leader said. Ace did so by employing high-speed perception, AI-based control and a state-of-the-art robotic system.

If you’ve ever spent any length of time in a tall building—either because you live or work in one—you probably know this feeling: The elevator’s cars always seem to go in the wrong direction. If you want to go down, there’s an elevator going up, and vice versa.

The phenomenon intrigued physicists George Gamow and Marvin Stern in the mid-1950s.

Shellscape is an online web app that simulates a terminal environment for learning Linux shell commands. It has 31 levels of increasing difficulty that work entirely on the frontend without needing any virtual machines or installations.

In Serbia, we say “zimi-zami-zum”. When “zum” is said, everyone shows either a fist (like rock) or a relaxed palm (like paper). This is played by 3 players. The odd one out (eg. the lone palm vs. 2 fists) is “it”. We can scale this up to more players by deciding that the majority is eliminated and the minority moves to the next round. Repeat this process until there are fewer than 3 people remaining.

Mozilla says it used an early version of Anthropic's Claude Mythos Preview to comb through Firefox's code, and the results were hard to ignore. In Firefox 150, the team fixed 271 vulnerabilities identified during this effort, a number that would have been unthinkable not long ago. Instead of relying only on fuzzing tools or human review, the AI was able to reason through code and surface issues that typically require highly specialized expertise.

A curated list of modern alternatives to classic command-line tools.

Faster, prettier, smarter replacements for the Unix utilities you use every day.

TUI SSH client Suite: list-based host selection, parallel sessions, multi-hop proxying, and port forwarding. SSH/SCP/SFTP etc. Written in Go.

Claude Code has gotten extremely good at finding security vulnerabilities, and this is only the beginning.

The landscape of AI is not merely filled with news. It is filled with teams. You have the doomers, the accelerationists, the skeptics, the it’s-a-bubble oracles, the anti-bubble counter oracles, and so on. It would be convenient for my sanity—and, perhaps, the sanity of my readers—if I simply joined one team and never removed the jersey. But I don’t think any aforementioned tribe has a monopoly on good arguments. I think the doomers are right about the risk of the technology, and the accelerationists are right about the promise of the technology, and the skeptics are right that the doomers and accelerationists can both overstate their cases.

In 2022, I made a New Year’s resolution to switch from Chrome to Firefox, and from VS Code to Neovim.

My goal was to reduce my dependence on GAFAM tools, and it turned out to be a good decision considering this.

It took some time to adjust, but I am now a happy Firefox user on both desktop and mobile.

That said, it still has some issues, for example, the tab system on Android. I wish there were an easier way to search through tabs or select multiple tabs to close them, instead of closing everything at once.

However, my experience with Neovim was very different. I can say I really tried to adopt it, as I used it for four years before deciding to abandon it.

My first session with Claude Code was practically magical. I was speaking to my computer, telling it with natural language what I wanted it to do, and it was able to just do it. It did ( and still does ) feel like a completely new form of input, a new way to control my machine. I have misgivings about using AI in this way, but I still think this is a great tool for sufficiently low-level tasks. I’m waiting eagerly for the day that I can spin up a local LLM that can perform this function as well as Claude Code does.

I'm as anti-genAI as it gets. And yet, this past month, I have used generative coding to complete a project. It works. I hated making it.

These days, Wandering Thoughts has some hacked together HTTP request rate limits. They don't exist for strong technical reasons; my blog engine setup here can generally stand up to even fairly extreme traffic floods (through an extensive series of hacks). It's definitely possible to overwhelm Wandering Thoughts with a high enough request volume, and HTTP rate limits will certainly help with that, but that's not really why they exist. My HTTP rate limits exist for ultimately social reasons and because they let me stop worrying and stop caring about certain sorts of abuse.

One of the biggest problems with measuring AI progress is the ambiguity of measuring intelligence itself.

AGI is treated as a milestone we have yet to cross, but there is no central definition of AGI.

Depending on who you ask, AGI is achieved when a system:

• can fool humans into thinking it is one of them, in other words, pass a Turing Test
• demonstrates creativity (Springer)
• can develop new skills (DeepMind)
• solves unfamiliar tasks (DeepMind)
• is generally capable across domains (IBM)
• is superior to humans in intelligence (Scientific American)
• outperforms humans economically (OpenAI Charter)
• can independently solve complex problems without human oversight (DeepMind)

Even with the lack of consensus, I can confidently say we have AGI, because the criteria above has been met.

Retrieval-Augmented Generation (RAG) has become the dominant paradigm for grounding Large Language Model (LLM) agents in domain-specific knowledge. The standard approach requires selecting an embedding model, designing a chunking strategy, deploying a vector database, maintaining indexes, and performing approximate nearest neighbor (ANN) search at query time. We argue that for domain-specific knowledge grounding --- where the vocabulary is predictable and the corpus is bounded --- this entire stack is unnecessary. We present Knowledge Search, a two-layer retrieval system composed of (1) grep with contextual line windows and (2) cat of pre-structured fallback files. Deployed in production across 20 specialized LLM agents serving three knowledge domains (Traditional Chinese Medicine, Christian spiritual classics, and U.S. civics), our approach achieves 100% retrieval accuracy with sub-10ms latency, zero preprocessing, zero additional memory footprint, and zero infrastructure dependencies.

Scientists and educators are concerned about students using artificial intelligence to shortcut their learning. But there are also opportunities, especially when it comes to teaching neuroscience students how to code.

Visualizing machine learning one concept at a time.