Cloze deletion-based prompts are typically much easier and faster to write than full question/answer prompts. But Cloze deletion prompts seem to produce less understanding than question-answer pairs in spaced repetition memory systems. One interesting “middle path” may be to use language models to generate question/answer prompts from cloze deletions. Initial experiments with Ozzie Kirkby (2021-06-10) showed surprising promise.
Trying again with GPT-4 on 2023-06-14, I had immediate success with this prompt: 20230614095308. Also worked pretty well with GPT-3.5-turbo, though results were slightly less good.
See also: Cloze deletions can be used to create typical two-sided flashcards.
Idea emerged in conversation with Ozzie Kirkby, 2021-06-11