do you need to write your own spaced repetition prompts?

There is a notion that writing your own prompts may be more beneficial than using canned ones.

There has been pushback on that idea from some serious practitioners, though. In tweet - Dwarkesh custom spaced repetition card creation tooling, Dwarkesh Patel feels that the process of creating the cards is often rote, especially for his primary use, which is caching the ideas in his head for his next interview.

In my hacker news comment about canned prompts I detail Andy Matuschak and Michael Nielsen's evolving thoughts on the topic, where they seem to think that the process of generating high quality prompts is so hard to do well, that a person could benefit from having an expert create the cards for them.

Most people write sub-optimal prompts. You see them everywhere in the wild. Writing effective prompts is really hard. See Andy's blog post about writing SRS prompts for the myriad considerations. Even just keeping the 5 lenses he suggest in mind while creating prompts is challenging.

I suspect that an author who is well-versed in learning science and has deliberately practiced getting good at prompt generation would be able to make prompts that would help individuals as well (or perhaps even better) than their own prompts would.

Also, that author's efforts will be amortized across all the people who benefit from the prompts. Seems like a good trade!

The one thing to consider, however, is that externally authored spaced repetition prompts have some specific challenges compared to self-authored ones. Specifically, it's a problem if you either already know the material or you don't care about specific prompts. The problem is simple to overcome if you can maintain the deck of prompts. The problem is harder to address for sites that integrate spaced repetition prompts directly into them, like blog post - quantum country, which presume to know exactly which prompts you need to answer, when, and don't give the user much control over the experience.

Related

The questions around the effective assimilation of externally-generated material are also relevant to LLM wiki. I consider that in keeping externally-generated material directly in your notes.