206 ·GitHub
What is this?
A free, interactive online course that teaches how world models work in AI, covering prediction, latent dynamics, planning, JEPA, video models, and where they fail.
A free, interactive online course that teaches how world models work in AI, covering prediction, latent dynamics, planning, JEPA, video models, and where they fail.Is this a tool I run, or just a website to read?It is a Next.js web app with nine chapter pages, each containing interactive visualizations, a quiz, and a downloadable PDF version.
It is a Next.js web app with nine chapter pages, each containing interactive visualizations, a quiz, and a downloadable PDF version.How is the course organized, and where should I start?There are nine chapters ordered for learning rather than history, starting with 'What Is a World Model?' and ending with 'What Is Still Broken in World Models?'; you read them online or download a chapter PDF.
There are nine chapters ordered for learning rather than history, starting with 'What Is a World Model?' and ending with 'What Is Still Broken in World Models?'; you read them online or download a chapter PDF.Do I need any background in ML to follow it?Not stated explicitly, but each chapter cites underlying papers and builds up concepts (latent space, dynamics models, JEPA) progressively, suggesting basic ML/RL familiarity helps.
Not stated explicitly, but each chapter cites underlying papers and builds up concepts (latent space, dynamics models, JEPA) progressively, suggesting basic ML/RL familiarity helps.Are the claims in the course fully verified?Only claims that can be independently verified are stated as fact; everything else is explicitly attributed to its source, and every embedded YouTube video is checked against the oEmbed endpoint before shipping.
Only claims that can be independently verified are stated as fact; everything else is explicitly attributed to its source, and every embedded YouTube video is checked against the oEmbed endpoint before shipping.Can I reuse or adapt this course material?Yes, under CC BY-SA 4.0, but you must credit the author and the course, note any changes, and keep your adaptation under the same license.
Yes, under CC BY-SA 4.0, but you must credit the author and the course, note any changes, and keep your adaptation under the same license.How long does the whole course take?About two hours across the nine chapters, with no signup required.
About two hours across the nine chapters, with no signup required.Why does the course bother defining 'world model' so carefully?Because the term is used for at least five different kinds of systems (renderers, simulators, controllers, representation models, implicit models), and the course argues you need to know which one is meant before evaluating any claim about it.
Because the term is used for at least five different kinds of systems (renderers, simulators, controllers, representation models, implicit models), and the course argues you need to know which one is meant before evaluating any claim about it.What is JEPA and why does the course cover it?JEPA (Joint Embedding Predictive Architecture) is presented as an alternative to predicting raw pixels, predicting in latent space instead; Chapter 07 is dedicated to explaining why this avoids the 'impossible blur' problem of pixel prediction.
JEPA (Joint Embedding Predictive Architecture) is presented as an alternative to predicting raw pixels, predicting in latent space instead; Chapter 07 is dedicated to explaining why this avoids the 'impossible blur' problem of pixel prediction.How do I generate the PDF versions of chapters myself?Run the project's pdf script (node scripts/make-pdf.mjs), which writes chapter PDFs to the public/pdf/ folder; a chapter page auto-links its PDF once the file exists.
Run the project's pdf script (node scripts/make-pdf.mjs), which writes chapter PDFs to the public/pdf/ folder; a chapter page auto-links its PDF once the file exists.What are similar projects?le-wm, transferlearning, ai-engineering-from-scratch, Book-Mathematical-Foundation-of-Reinforcement-Learning, d2l-zh, maths-cs-ai-compendium, d2l-en, fastai
le-wm, transferlearning, ai-engineering-from-scratch, Book-Mathematical-Foundation-of-Reinforcement-Learning, d2l-zh, maths-cs-ai-compendium, d2l-en, fastaiWhich topics is it in?self-supervised-learning, world-models, mdx, course, reinforcement-learning, video-generation, nlp, nextjs, book, artificial-intelligence, learning, tutorial
self-supervised-learning, world-models, mdx, course, reinforcement-learning, video-generation, nlp, nextjs, book, artificial-intelligence, learning, tutorial