Comment by johnsmith1840

The study was to use Q&A that are neither discoverable or guessable.

Pretty simple without training the one shot the model never guessed even under training the pepperoni answer

I mean, it rediscovers it because it stored it hidden from our view. Use whatever words you like though.