Shane Caldwell

The Tests All Pass

Intro to GPUs For the Research Oriented

All Reduce Across the Atlantic: Bandwidth in Decentralized Training

Twenty Billion Tokens of What, Exactly?

Pretraining at home: 20B tokens from 222 hours to 12

Offsec Evals: Growing Up In The Dark Forest

DiLoCo: Data Parallelism for the Datacenter Poor

RL Needed LLMs Because Agency Requires Priors

GPT-5 is Good, Actually: The Agony and Ecstasy of Public Benchmarks

The Religious Devotion of Haskell

The Input Sanitization Perspective on Prompt Injection

Infosec's Data Problem

Deep Reinforcement Learning for Security: Toward an Autonomous Pentesting Agent

An ML Eng's Review of OSCP