Parker Conley's Shortform
post by Parker Conley (parker-conley) · 2024-11-07T20:44:54.703Z · LW · GW · 1 commentsContents
1 comment
1 comments
Comments sorted by top scores.
comment by Parker Conley (parker-conley) · 2024-11-07T20:44:54.846Z · LW(p) · GW(p)
What are the actual use cases of memory systems like Anki?
Epistemic status: spent 30min cleaning up some notes from my Obsidian I jotted down yesterday. This ontology is rough and a bit illegible but potentially useful for narrowing down the actual use cases of memory systems.
Inspired by @Saul Munn [LW · GW]'s recent short form: Active Recall and Spaced Repetition are Different Things [LW(p) · GW(p)]. The concepts of active recall and spaced repetition apply pretty well here, but I saw Saul's post after writing most of the text below.
Roughly, there are types of knowledge in domains (recalling from Scott Young's Ultralearning, I might be slightly off):
- Proceduralized knowledge
- Conceptual knowledge
- Factual knowledge
I think spaced repetition systems are useful for three types of domains based on the nature of the cues in the domains:
- Domains where you are already cued on the learnings often (e.g., key business procedures in your full-time job). You naturally get spaced repetition in these domains, so memory systems like Anki are less useful.
- Domains where you are not cued on the learnings often or at properly spaced intervals. This may include domains where you were an expert at one point and then stopped being an expert.
- Domains where you need to understand cues before being effective. (e.g. understanding a new language or technical domain like quantum mechanics; see Quantum Country.)
Combining the two frameworks above, Anki is useful for:
- Proceduralized knowledge where you can appropriately cue yourself and you're not cued often on.
- Cueing procuduralized knowledge is hard, though. For example, if I wanted to review my knowledge of 'how-to-play-beach-volleyball' on a regular basis, I have to spend a time coordinating or attending an event.
- Though, some proceduralized knowledge is possible to cue: for example, I have been using Anki to review LeetCode problems.
- Conceptual and factual knowledge that you're (2 above) not cued often on or (3 above) need to meet a criteria of knowing.
- Conceptual and factual knowledge is much easier than procedural knowledge because these types of knowledge neatly fit in flashcards. I'm still not sure about the strict boundary between conceptual and factual knowledge, though.
- And I think the general heuristic that conceptual knowledge is harder to cue than factual knowledge is true: medical students are obsessed with Anki while students in other vocational schools (e.g. law) don't seem to be
Accordingly, memory systems may not be useful for:
- Domains where you are cued on information often enough to get natural spaced repetition.
- Are highly procedural in a way that can't be Ankified and therefore active recall is hard.