I use GPT-3.5 for grading a deck of ~1,600 Korean/English review flash cards. I’ve noticed that a small percentage of cards have an unusually high API error rat

I use GPT-3.5 for grading a deck of ~1,600 Korean/English review flash cards. I’ve noticed that a small percentage of cards have an unusually high API error rate to the point that I will just delete them and vary the card slightly to not deal with the error.