Andy also tested this idea. You can read his notes here:

GPT-4 can probably estimate whether two flashcards are functionally equivalent

https://notes.andymatuschak.org/zJ7PMGzjcgBUoPjLUHBF9jn

GPT-4 can probably estimate whether one prompt will spoil retrieval of another

https://notes.andymatuschak.org/zK9Y15pCnRMLoxUahLCzdyc