This has, incidentally, made many of my matrix-based Anki cards rather ugly. I'll probably update a bunch of answers with cut-and-paste from this essay.
Not to put too fine a point on it, but it's pretty bad I'm using raw LaTeX in this way - I'm sure it hurts my understanding! On the other hand, I've written so much LaTeX over the years that reading it seems really natural.
Even in something as basic as language learning, it would allow to commit idioms and common collocations (words "often used together") to memory once you know the original words - a nice "next step" after pure vocab.
1. learn the concept
2. break it into atomic question/answer pairs
3. define some dependence relation for all these pairs (X is foundational to Y and Z but not A)
4. input them into your SRS of choice
Each of these steps require significant amounts of time and brainpower. Also it would be harder if you were still learning the concept since you'd continually need to go back and redefine relations as your understanding grew.
On the one hand all this would be a great way to understand a subject. On the other, one could argue that it's very inefficient, and it would be better to not bring SRS into it all all. Compare this to say a vocabulary card, which you make, enter into SRS, then you're done. So far this (the vocab thing I mean) is the only SRS use-case that makes sense for me.
The hard part might be the UI for finding and selecting the dependencies for a new item.
There's actually an addon to help out with that - one I'm personally really satisfied with.
https://ankiweb.net/shared/info/1199027445
Long story short: Take notes in a LaTeX document (or take notes somewhere else and then put them in a LaTeX document, this will be your ground truth). If you format your document based on the addon's guidelines, you can automatically "compile" your document to Anki flash cards. That way if you have to change stuff, just change the document and recompile.
It seems he and I approach SRS very differently. I see it as a place to drop bits of information I want to remember, and once the info is in there, I no longer have to think about it, outsourcing the effort of remembering to the algorithm. He seems to use it as a tool to mentally digest a subject, constantly revising his understanding and readjusting his cards. That process isn't foreign to me, but I prefer to use something besides SRS for it. If I had to constantly revise my cards, that would seem to defeat the purpose of using SRS anyway.
Very interesting to get a different perspective on this though, so I'll have to think more about his approach. I'm also interesting in the creation details of his math-cards. Does he use LaTex? Maybe he goes into it in one of his earlier posts...