Stories by qwertyforce

Claude Fable 5 will sabotage "frontier LLM research" tasks

I (Spiritually) Won Comma.ai's Compression Challenge

Flash Attention Is Not Always Faster for Short Sequences

Fast Attention for Short Sequences

Forecast: Stack Overflow Has About 2 Years Left

Math and Me

Forecasting the Death of StackOverflow

Google AI Studio is now sponsoring Tailwind CSS

Flash Attention from Scratch

Learn Cutlass the Hard Way

QK Norm and the Curious Case of Logit Drift

Engineering the highest cited cat, Larry

ICLR Blogposts 2024

Tutorial on Diffusion Models for Imaging and Vision

A Probabilistic Interpretation of Regularization (2016)

ICLR Blogposts

Why Batch Norm Causes Exploding Gradients [2020]

Tensor Core Performance: The Ultimate Guide (2020) [pdf]

The Stilwell Brain [video]

Optimal power limit for deep learning tasks on RTX 3090