Newsletter

Thoughts on AI, development, and technology. Published on Substack.

Note: Trying a new format with the sort of sparsely-lined paragraphs - I think it may read a little better here on substack than my normal ranting long form drivel....

I’ve been running Qwen3.6-35B-A3B-MTP on my RTX 3070 (8GB VRAM) for a while now and previously documented getting ~55 tok/s....

PR #22673 just landed in upstream llama.cpp, and it is a big deal....

If you’ve been following the local LLM space this year, you’ve probably been tracking parameter counts, quantization methods, and hardware requirements like a stock trader watching candlesticks....

Let’s talk about a paper that quietly dropped on arXiv this week and absolutely deserves more attention than it’s getting....

May 15, 2026
John Paul Wile

Living with ADHD as an adult in 2026

I hate the fact that I can sit down and wrestle with complex problems for hours....

Google has quietly installed a 4GB local LLM into Chrome, and billions of users have no idea it is there....

and it quietly reveals something deeper about knowledge, logic, and how natural thinking really works...

May 24, 2024
John Paul Wile

Megafortress

A trip down memory lane...

January 6, 2024
John Paul Wile

Aunt Michele

This just made my whole year....

A small piece on why we use monitors, EQ's and other things in order to get a flat frequency response....

The correct way to set the cutoffs on your studio gear....