The elephant in the room is that we blindly trust the binary that is allowed to read our private keys and perform critical actions. The trust boundary should be at the bottom of the stack, at the kernel, and we need to verify that binaries (such as sshd) that have access to our private keys haven’t been tampered with. Read more ›
We’re on a journey to advance and democratize artificial intelligence through open source and open science. Read more ›
Prism: An Impure Functional Language With Typed Effects This is going to be a very nerdy post so bear with me. Here is a function. Read it the way you would read any other function, and then tell me its type. fn fib(n) = var a := 0 var b := 1 repeat(n) fn let t = a + b a := b b := t a That is a mutable loop. There is a var, there is assignment, there is a temporary so the swap does not eat itself. It is, line for line, the fib you would write in Python after deciding that recursion was a youn... Read more ›
The definitive ranked board for open-weight AI models — all 21, graded on six axes, filtered by hardware tier, category, and license. Updated June 2026. Read more ›
A surprising alignment quirk I learned the hard way: adding 4 bytes of struct padding makes Go's array clearing 49% faster on Intel, all thanks to REP STOSQ. Read more ›
The future of development is commanding an army of AI agents, orchestrating Gemini’s multi-modal power. Read more ›
"After 27 years of Unreal Engine, and 12 years at Epic Games and Unreal Engine I have decided to move on." Read more ›
Why two identical VMs on the same host can perform 20% differently — and what NUMA topology in Xen-based virtualization actually costs you. Read more ›
I’m working on a Kubernetes operator and an API server for it to interface with. Both of these are crates in the same Rust workspace. The motivation for this was that I wanted to define all of the types used by the two services in one place, and keep the services tightly-coupled. When writing the operator, I wrote a protocol to define the HTTP requests the operator sends to the server: Read more ›
Compiler infrastructures such as MLIR rest on a set of design principles: IR abstractions, interfaces, match-and-rewrite, flow analysis, type conversion, staged lowering, and so on. These concepts have proven themselves in practice. Good designs typically arrive through engineering knowledge, intuition and experience. Many of them, however, have correspondences in formal theory. MLIR's match-and-rewrite engine has correspondence to a \emph{term-... Read more ›
The fastest WebGPU runtime on the web. Run models in the browser, add a cloud gateway, and build with one simple Sipp client. Read more ›
Proper TCP socket splicing reduces the load on userspace processes and enables more efficient data forwarding. We realized that Linux Kernel's SOCKMAP infrastructure can be reused for this purpose. Read more ›
Tips and tricks in Python, Bash, Clojure and other languages/environments. Read more ›
I’ve been working on Emacs performance on macOS for a couple of months - not continuously but for couple of months nonetheless. During that time I’ve been occupied with hooking up instrumentation and creating benchmarks. I also gave multiple LLMs the codebase and asked for it to search for specific things. Usually the results were poor. After analyzing patches, either the impact was minimal or the problem was misunderstood - something completely expected. Read more ›
How good are local LLMs at translation, and do you actually need the cloud? A reproducible benchmark of 24 on-device, self-hosted, and cloud models translating into English, with the low-resource case (Afrikaans) front and centre. The headline: on Afrikaans→English a local 18 GB model lands in a statistical tie with frontier cloud. Same blinded Tatoeba sentences, same prompt, greedy decoding, scored multi-reference with COMET (meaning) and chrF++ (surface). Built to pick a translation model f... Read more ›
Safety property means “nothing bad happens during the run of a program”; liveness property means “the program eventually does something good”. In this post, we walk through a simple proof of a liveness property in Veil, using a basic consensus protocol as an example. Read more ›
Spawning a worker is easy. Making it reliable means designing a protocol, packaging the worker file, proxying callbacks, handling crashes, and shutting down cleanly. Read more ›
The Epic Games CEO discusses his vision for "Team Open," his objections to Steam's AI disclosure requirement, and the huge problems facing AAA game development. Read more ›
In May 2026, the Bun team did something the software industry has been whispering about for years: they rewrote their entire runtime from Zig to Rust. Not over the course of a year with a dedicated team. In six days. Using AI agents. At nearly a million lines of code, Read more ›