Eric Eaglstun · AI
A working notebook on AI & machine learning.
Almost nobody fully understands how modern AI works - not even the people building it. That makes it easy to fear, easy to hype, and hard to think clearly about.
I started this to fix that for myself - then figured it might help anyone else trying to make sense of AI. The heart of it is a glossary - short, plain-language definitions of the terms I kept tripping over while running models on my own machine, explained the way you’d explain them to a friend. Around it sit practice notes, blog posts, and deep dives into the machinery underneath. No hand-waving: because you can’t have a sane opinion about something you don’t understand.
Latest
- blog My Karaoke Machine Throws Away Every Word It Hears
I built my band a karaoke video maker, and the trick that makes it work is refusing to trust the one part everyone assumes you'd trust: the transcription. The machine listens to the singing, mishears most of it, and I keep only its sense of timing - never its words. A small lesson in using a model that lies, plus why the fakery around the edges is what makes it feel real.
- deep dive I Went Looking for Metal and Found a Broken Lion
I forked bitsandbytes to build an AI-assisted native Metal backend for Apple Silicon. The backend stayed in the fork. The first correctness test found a Lion optimizer bug in three existing backends, and that smaller finding made it upstream. This is the useful shape of AI-assisted open-source work: build the ambitious thing in public, then separate the part you can prove, test, and hand to a maintainer without handing them your whole experiment.
- practice I Gave Five Agents One Marble and Told Them to Stay in Their Lane
MARBLE MAGNIFICENCE has specialist agents for level shape, block geometry, rendering, criticism, and music. The useful part isn't what each one knows. It's what each one is forbidden to touch - and why the critic has to leave the bug broken when it finds one.
Pick a thread
The site makes more sense as a few arguments than as one long reverse-chronological pile.
Why does the machine sound so certain?
What does local AI actually take?
Blog →
Notes and posts on AI/ML - local inference, models, and whatever I’m tinkering with.
Practice →
How I actually use AI day-to-day, and the techniques that have earned a permanent place in my workflow.
Deep Dives →
Longer, hands-on walkthroughs - how the pieces fit together, with the details left in.
Glossary
All 42 terms →Short, plain-language definitions of the AI/ML terms I keep bumping into - formats, GPU backends, architectures, and the concepts underneath.