$ grep -ri notes/
9 matches
I removed a learned behavior from a tiny AI model. Adding it to another model was harder.
A small mechanistic-interpretability result: subtracting one internal direction mostly removed a learned behavior, but adding matching directions to a separately trained model did not install it.
What I Automated with OpenClaw (and What I Couldn't)
I set up OpenClaw to handle my dev workflow. Some of it felt like magic. Some of it broke in ways I didn't expect.
What Micrograd Taught Me That PyTorch Hides
Building autograd from scratch changed how I think about every loss.backward() call. Here's what the framework doesn't show you.
Why Your Typing Speed Doesn't Matter (for Code)
I type 90 WPM on MonkeyType. I was mass slower writing actual code. That gap is why I built CodeSprint.
Final Report – Planetary Exploration Project↗
Engineering a six-wheeled rover prototype for Proxima Centauri B exploration — design, construction, testing, and what the data actually tells us about a real mission.#fusion
PEP Assignment – Proxima Centauri B↗
Why Proxima Centauri B is humanity's most realistic interstellar mission target — and what it would actually take to get there and survive.#fusion
Building ChatGPT and Adding My Own Twist
Rebuilding GPT-2 from scratch, wiring in RAG, and shipping LittleGPT as a local Streamlit app.#fusion
Reading Reddit’s Market Mood: A Minimal Streamlit Sentiment Dashboard
A single meme can move a stock twenty percent before the coffee finishes brewing....
How I built a Hacker News sentiment analyzer more neutral than Switzerland — then managed to fix it
### Building a Hacker News Sentiment Analyzer in 30 Minutes: Solving the “Everything is Neutral” Problem ### The 30-Minute Challenge This morning I ga...