Manifesto

You can watch ten hours of video about neural networks and still not know what a weight does. Not because the videos are bad, but because watching is passive. Understanding starts the moment you change something and see what breaks.

So every idea here has a thing you can move. Drag a weight and the decision line turns. Raise the learning rate and the ball overshoots the valley. Change one token and the attention map lights up somewhere else.

Most technical Arabic on the internet is a machine translation of an English page, written by nobody, read by nobody twice. That is not a language, that is a residue.

Every lesson here is authored twice. The Arabic is written as Arabic, with its own phrasing and its own examples, by people who think in it. Right to left is the default, not a stylesheet afterthought. Numbers, dates and plurals follow the language, not the code.

There is no notebook to install and no GPU to rent. The TypeScript in a lesson runs in your browser tab, in a sandbox, and prints its output next to the text.

This is a constraint we chose. Anything that cannot be made to run in a tab is either explained without code or left honestly unfinished.

A lot of AI writing hides the gaps. Why a large model learns in context, why scaling keeps working, what a feature really is inside a layer: these are open questions.

When a lesson reaches one, it says so and marks it. Knowing where the map ends is part of the map.

Chapters, visuals, quizzes, exams, the playground and the placement test are free, and they stay free. Writing them is a fixed cost, and a fixed cost does not belong on a meter.

What costs money every month is memory: syncing your progress across devices, storing your bookmarks, running the tutor. That is what Premium pays for. Nothing you learn from is ever behind it.

Masfofah is built in Saudi Arabia, for anyone who wants to understand this technology in their own language. If it helps one student see a gradient for the first time, it was worth writing.