Arkiv

Grandparents are glued to their phones, families are worried [video]

In a recent essay for The Atlantic, writer Charlie Warzel explored why so many older adults are spending more time on their digital devices — and why their children and grandchildren are increasingly uneasy about it. But is this shift actually worth worrying about? Or are younger people just projecting their own anxieties about screen […]

Write up of my homebrew CPU build

Hello! This is the third part of the series about WCPU-1, my homebrew 8-bit computer. Check out Part 1 and Part 2! Recap/Intro # Let’s just get this over with: Don’t look it directly in the eye! So in Part 2 I simulated the whole design in Logisim-Evolution and tweaked it until I was happy. […]

Stop Sloppypasta

Sharing raw AI output is like eating junk food: it’s easy and may feel good, but it’s not in your best interest. You’ll negatively influence your relationship with the recipient, and do yourself a disservice by reducing your own comprehension. Before LLMs, writing took effort. Authors spent time and effort considering and selecting their words […]

The most brilliant move in corporate history?

An allergy to centralized computing Based on Tweets by MilkRoadAI The 3rd? most valuable company on Earth watched as its rivals lit $650 billion on fire and did nothing. The biggest cash bonfire in history, by far, eagerly fed by all the usual suspects but one. It’s either the dumbest or the most brilliant move […]

LLM Architecture Gallery

Popular 230B coder that opts for a classic architecture instead of the newer hybrid-attention ideas. Scale 230B total, 10B active Date 2026-02-12 Decoder type Sparse MoE Attention GQA with QK-Norm Key detail Deliberately avoids sliding-window or linear-attention hybrids while keeping a 10B active path. Related concepts

In Memoriam: John W. Addison, my PhD advisor

John Addison (1930–2026) died last summer, 2025, at the age of 96. He was my PhD advisor at UC Berkeley, and I count myself extraordinarily lucky to have worked under his guidance. When I arrived in Berkeley in 1966, I had no clear idea what I wanted to study. Some fellow Canadian newcomers strongly recommended […]

UMD Scientists Create ’Smart Underwear’ to Measure Human Flatulence

A new nationwide study is recruiting volunteers to map the full spectrum of human flatulence.  Hall’s team demoing a Smart Underwear Prototype. Credit: University of Maryland. Scientists at the University of Maryland have created Smart Underwear, the first wearable device designed to measure human flatulence. By tracking hydrogen in flatus, the device helps scientists revisit […]

Show HN: GDSL – 800 line kernel: Lisp subset in 500, C subset in 1300

Pages for posting about my work until I decide to tackle websites. 2/14/26 Early GDSL Seeds 3/14/26 Compilers are enormous these days, millions of lines spanning frontend to backend to turn text to process on a system. Any small compiler gets small by making tradeoffs, it constrains itself to a shape small enough that it […]

Pretraining Language Models via Neural Cellular Automata

Research Paper Blog What if the path to smarter language models doesn’t require more text — but synthetic data from abstract dynamical systems? 6% Perplexity gain 1.6× Faster convergence 164M NCA tokens used 01 — The Problem We’re running out of text Large language models are hungry. They require exponentially more data to keep improving, […]

Learning athletic humanoid tennis skills from imperfect human motion data

Zhikai Zhang1,3*, Haofei Lu1,3*, Yunrui Lian1,3*, Ziqing Chen1,3, Yun Liu1,3, Chenghuai Lin3, Han Xue1,3, Zicheng Zeng3, Zekun Qi1,3, Shaolin Zheng3, Qing Luan3, Jingbo Wang5, Junliang Xing1, He Wang2,3, Li Yi1,4† * Equal contribution † Corresponding author 1Tsinghua University, 2Peking University, 3Galbot, 4Shanghai Qi Zhi Institute, 5Shanghai AI Laboratory