Mason  🏃‍♂️ ✂️

I should note that there are (probably broadly trainable) strategies for massively improved rote memorization, but the very hacky way this works makes me very skeptical that information stored this way can be accessed for much more than basic call-and-response

The first sentence is just plain wrong. But the implied question, when fixed up, is a good one: what are the opportunity costs for SR? I've experimented intensively with this question in mind. AFAICT, SR is most useful when learning a little outside your expertise.

Inside one's field or fields of expertise, SR is still occasionally useful, with little niggling things you keep having to look up. But for the most part I agree: one's natural interest and ways of encoding knowledge are powerful signals which do most of the trick.

(Caveat: within one's field there are genuinely valuable ways of using SRS: http://cognitivemedium.com/srs-mathematics  But that piece is principally about a very useful method of analysis; the use of SRS is subsidiary, & has some problems associated to opportunity cost, as discussed at the end.)

You told me once that you have things like paintings you like in your SRS — this seems lovely and right. Intuitively, it feels less hacky/risky to apply SR principles to unlabeled lovely things, rather than flashcard-style "view input, produce output, check output against record"

SRS gives you control over what you remember. Turns out, granted that power, most people don't know what to do with it. I think that's the fundamental reason underlying the risk, & can cause high opportunity costs.

Eg, I keep meeting people who use SRS to memorize lists of capitals of countries they don't care about, or dates in history they don't care about, or minutiae of APIs they almost never use. Ask them why, and their models are very poor.