One of the problems with recalling Anki cards is a lot of them are coded in through the specific format of the card leading to context retrieval. It would be cool if we could get a LLM to rewrite the card in a different way every time the card comes up. That way I actually have to read the card and recall rather than contextualize. No idea how this would be done software wise.
I completely agree and had the same idea! Wonder how many others did too. I’ve noticed that, at least with the prompts I’ve tried, GPT-4o does not provide good results. However, in the little testing I’ve done with GPT-o1, it actually does provide decent results.
The main challenge I’ve found is that I’m asking GPT to rephrase a card but without modifying anything inside clozes - i.e., not only keep the same concepts clozed, but the exact same phrases, too. This severely limits the options, although it’s still usually doable.
Perhaps the most difficult part in this all is simply in deciding which cards should be rewritten… any ideas on how to detect these sort of pseudo-leech cards?
Duplicate of AI Card Variations - to prevent card memorisation