From 300KB to 69KB per Token: How LLM Architectures Solve the KV Cache Problem
How the KV cache gives every AI conversation a physical weight in silicon, and what happens when the memory runs out.
The latest blog posts to take off on Hacker News. 80,808 posts.
How the KV cache gives every AI conversation a physical weight in silicon, and what happens when the memory runs out.
In my last post, I showed you to use FireWire on a Raspberry Pi with a PCI Express IEEE 1394 adapter. Now I'll show you how I'm using a new FireWire HAT and a PiSugar3 Plus battery to make a portable MRU, or 'Memory Recording Unit', to replace tape in older FireWire/i.Link/DV cameras. The alternative is an old used…
Every wall, floor, barrel, and imp is a div — DOOM rendered entirely in CSS. Using 3D transforms, CSS math functions, @property, clip-path, anchor positioning, and SVG filters to build a fully playable 3D first-person shooter in the browser without Canvas or WebGL.
AI eval mistakes and the data science practices that fix them.
A personal blog, by a programmer and IT expert. Essays, Articles, Guides, and Recipes. As well as Code, Quotes, and Links.
TruffleRuby started as my internship project at Oracle Labs in early 2013. It is an implementation of the Ruby programming language on the JVM, using the Graal dynamic compiler and the Truffle AST interpreter framework . TruffleRuby can achieve peak performance well beyond that possible in JRuby at the same time as…
I’ve taken agency in the treatment of my bone cancer (osteosarcoma in the T5 vertebrae of the upper spine). After I ran out of standard-of-care treatment options and there were no trials available for…
And why you would want to exec a cpio
EDIT: removed a couple of false positives based on comments from people. 2/04/2026: added a GitHub repository with data for this post
How the commercial spreadsheet reshaped America
The official White House Android app has a cookie/paywall bypass injector, tracks your GPS every 4.5 minutes, and loads JavaScript from some guy's GitHub Pages.
People close to Trump are trading based on national secrets
Executives love AI while ICs stay skeptical. The divide comes down to non-determinism: who is evaluated on judgment versus deterministic execution.
UA researchers find 3,500-year-old loom that reveals key aspects of textile revolution in the Bronze Age
Natalie Shapira 1 Chris Wendler 1 Avery Yen 1 Gabriele Sarti 1 Koyena Pal 1 Olivia Floody 2 Adam Belfki 1 Alex Loftus 1 Aditya Ratan Jannali 2 Nikhil Prakash 1 Jasmine Cui 2 Giordano Rogers 1 Jannik Brinkmann 1 Can Rager 2 Amir Zur 3 Michael Ripa 1 Aruna Sankaranarayanan 8 David Atkinson 1 Rohit Gandikota 1 Jaden…
Alongside the various bugs you get, one of the issues of upgrading to MacOS 26 is that it has one of the most notorious inconsistency issues in window corners. I'm not sure what exactly pushes product designers to like …
Before we explore FreeBSD jails, it is worth refreshing our understanding of how Linux solved the same problem with LXC (Linux Containers).
Today marks 15 years of Waterfox!
Personal website of Lara Aigmüller. Thoughts about web frontend development, music, and more…
Over the past fifty years, the U.S. economy built a giant rent-extraction layer on top of human limitations: things take time, patience runs out, brand familiarity substitutes for diligence, and most people are willing to accept a bad price to avoid more clicks. Trillions of dollars of enterprise value depended on…
Many people know, or have heard of, the Motorola 68000 architecture. These processors have been used in many machines: in the first generations of Apple Macintosh computers, in the ever-rivals Amiga and Atari ST home computers, in many workstations built by Sun , HP and NeXT (to name only a few), but also in many…
A fabulously gay blind Romance author.
There are a lot of challenges in the way QA and engineering work together. Many eng leaders think you shouldn't have QA. I dig in.
A warning about rising prices, vanishing consumer choice, and a future where owning a computer may matter more than ever as hardware, power, and control drift toward data centers and away from people.
We have significantly improved how code reviews work in Lubeno. We think code review should do more than catch bugs, it should also help teams reduce comprehension debt. In this post, I want to explain how that idea shaped our recent work on code review in Lubeno.
An introduction to the jsongrep tool, a technical explanation of its DFA-based search engine, and performance results against popular JSON query tools.
What if you could let Claude and Codex work together as pair programmers, talking to each other directly? One of them as the main worker and the other as a reviewer.