Korva is a scroll of the internet’s best ideas.
A good flashcard predicts your mind
Andy Matuschak and a collaborator tried to teach language models to write good spaced-repetition prompts. They are sophisticated people. They tried reinforcement learning on open source models, chain of thought, and a big prompt sent to the best closed source model.
The key constraint appeared to be that a good card projects somebody's mind in three months. How will they associate the question? What answer will they be thinking of at that moment? Does the prompt draw out the detail you actually want to take away from the passage?
Writing works the same way. With each word and each sentence, the writer is thinking about what is happening in the reader's mind right now, and that takes a lot of mentalizing.
Why does spaced repetition work?
Hermann Ebbinghaus studied his own memory in the 1880s and found that recall of new material falls steeply at first and then levels off. Reviewing the material just as it begins to fade restores it and slows the next decline. Spacing the reviews further apart each time is the idea behind modern flashcard software.
What is the curse of knowledge?
In 1990, Elizabeth Newton asked Stanford students to tap the rhythm of well-known songs for listeners to identify. The tappers predicted that half of the songs would be recognized. Listeners identified about 2.5 percent of them. The tappers could hear the melody in their heads, which the listeners could not.
What is the mnemonic medium?
Andy Matuschak and Michael Nielsen built a format that embeds spaced-repetition prompts directly into an essay. Their 2019 essay series, Quantum Country, introduces quantum computing and brings reviews to the reader afterward. Readers answer the prompts on a schedule, so the essay itself reinforces what it taught.
More from Dwarkesh Podcast
