michael_nielsen

It's best to make cards as atomic as possible.

An example: when learning about soft linking I initially had a question "How to create a soft link?" with answer "ln -s filename linkname".

This was too complicated --- I always stumbled on the order of filename and linkname.

I broke the question in two: "What is the command and option to use to create a soft link?" A: "ln -s". And "What is the ordering of filename and linkname"

This sounds silly and obvious, but the improvement was very considerable: the two cards became trivial.

(There's something quite deep about memory in this example, which I don't understand.)

Think of it as a search problem: Given A, retrieve the appropriate B. The way it does that is by mining A for contextual cues that it uses to restrict the set of possible Bs and then retrieves the one with higher likelihood of being the correct B.

Problem with wide questions is not enough restriction of that set while concrete questions provide not only more direct cues (you end up with a better set of possible answers) but possibly also more cues in general.