Train real neural networks. Battle architectures side-by-side. Understand how language models work — from first neuron to full deployment.
Whether you're here to learn, build, or fund breakthrough research — everything happens in one place.
10 interactive modules from zero to expert. No prerequisites, no command line. Watch models train, see attention patterns, understand how it all works.
Train your own models on real GPUs. Pick an architecture, choose your data, tune the knobs. Watch it learn in real time, then battle it against others.
Fund novel AI architecture research. Watch live training of ChaosSat — a fractal-topology LLM that challenges the transformer paradigm. Your support, our science.
Every architecture is matched at ~12.8M parameters. Same data, same compute. Pure design difference.
From "what is a neural network?" to "build your own LLM." Spiral pedagogy — you see results first, understand machinery after.
Join the community building, battling, and understanding neural networks from the ground up.
10 modules from zero to architect. Each module builds on the last, but you can jump to what interests you. Every concept appears at increasing depth — no walls, no prerequisites you didn't already see.
Configure and launch your own training run. No command line needed — just pick, tune, and go.
Each model is ~12.8M parameters. Same compute budget, different design philosophy.
Past battles will appear here.
Hands-on tools for building intuition. No code, no math — just drag, adjust, and watch. Each playground is also embedded in relevant curriculum lessons.
Live window into the ChaosSat architecture research. Your support funds novel AI science — here's what it's producing.
Track your progress, manage your models, and customize your experience.
Create an account to save your training runs, track curriculum progress, and earn badges.