Animated videos about how AI models work, and the projects I build with them.
Diffusion models learn to turn random noise into new samples, one small step at a time. A young beaver loses his dam to a flash flood and learns to rebuild it the same way: by learning what the noise is, then removing it, nudge by nudge.
A faster model decodes a few tokens ahead, so the powerful model can check all of them in a single pass. In the original paper, this made models two to three times faster.
An AI-driven choose your own adventure game. A Director keeps the hidden side of the story: secrets, character arcs, stats and pacing. A Writer turns each beat into prose, and a Producer illustrates every turn and keeps the characters looking consistent. Each turn you answer the Director's questions, and risky choices are rolled against a success chance. Tell the Director precisely what kind of story you want, pick one of its ideas, or let it lead you without input. Play alone or with friends in a party.
Play Labyrinth
An AI language tutor. Describe what you want to learn in a sentence, like "I speak English and want to learn Libyan Arabic for a trip to Tripoli", and it builds the course. Tutoring sessions are made mostly from words you already know, with a few you're still learning, and every line is voiced and explained after you answer. It also writes stories in the language and teaches grammar using your own vocabulary. No streaks or XP.
Open LanguagesRepositories I've worked on, past and present, including the Quart template I use for the backend of all my web projects, and TensorFlow implementations of machine learning models.
github.com/manicman1999
