People reason in ways far richer than language. Images, spatial relationships, feelings, half-formed intuitions, all translated into words only when we need to communicate. Machines reason in the opposite fashion: they take in text, break it into tokens, and build everything they know from predicting what comes next.

We don't actually know what the medium of higher thinking is, and we can't replicate it faithfully even between two humans. Maybe we can tap into this plane, but if not, perhaps our whole definition of intelligence is too anchored to how humans happen to do it. Machines could find entirely different ways to reason if we stopped insisting they borrow ours.

ATOM is where I explore that question, building frameworks to test whether genuinely new forms of machine reasoning are possible.

If you've tried to learn anything recently you know the problem. Endless guides, tutorials, and AI-generated content, and somehow all of it makes learning harder, not easier. You consume more and understand less.

I started paying attention to the revealing power of one's own consumption. The things you naturally save, return to, and spend time with reveal genuine sparks of curiosity. Lumin is built to harness those sparks toward building mastery. You capture what interests you, and the system challenges your understanding through Socratic dialogue. Not the most efficient path to competence, but the most artisanal one.

This started because I needed to learn math and computer science for my own projects and couldn't find the right tool. Everything was either a textbook or a tutorial. What I wanted was a well-designed sequence of questions that let me discover the ideas myself. No answers given, just the right question at the right time to push me forward.

Cove is that experiment. An adaptive curriculum where you learn by working through problems, not by being told solutions. I'm testing new question types and features to find out what actually helps people build intuition from the ground up.