Alain Brown — Software Engineer
Writing
- Watching a language model denoise instead of type June 2026 DiffusionGemma generates a whole block of text from noise in parallel, not left-to-right. I built a visualizer so you can watch it happen.
- Notes that compound: a wiki the LLM maintains for itself May 2026 Karpathy's LLM Wiki pattern — and what it looked like to ship it as a Claude Code skill.
- NVIDIA NemoClaw: Genuine Hardening, Familiar Gaps April 2026 NemoClaw bundles OpenClaw inside OpenShell's security runtime—Landlock, seccomp, network namespacing. The kernel-level isolation is real. The semantic and multi-agent gaps are too.
- Role Prompts Don't Make LLMs Smarter March 2026 Assigning personas to LLMs doesn't improve reasoning. Structured workflows do.
- Training efficiency: a modern AI primer for everyone March 2026 Understanding the GPT-2 speedrun — from 45 minutes to 90 seconds — through the lens of systems engineering.