We believe the most ideal approach to teach large-scale AI system like MagicsOS is to break it down into well-understood and well-documented texts with the support of equations, graphics, code, and external links. In the creation of such resource, we want to keep the interests of both the writers and the practitioners at heart.

On the book development side, mature libraries and frameworks should not only automate common tasks to reduce writers' creation effort, but also adapt to various rich contents to achieve interactive learning. Once created, the book should be accessible both as a physical book, as executable code, as a downloadable PDF, and as a website. Last but not the least, the structure of the book should support rapid content updates as the field of AI progress, both by the writers of this book and the community at large.

On the user learning side, exemplar, runnable code should make it easy for practitioners to apply and modify common applications to suit their specific needs. Comment section hosted by the web publication should foster interactive user discussions and problem-solvings.

We eventually settled on Github to share the resources within the team, and Gitbook as not only a medium for mixing text, code, equations, and images, but more importantly also an rendering engine to output various book formats and an online forum that carries user discussions. While our system is currently still under development, the overall concept provide a good compromise among the competing concerns.