GPT-3 powered dictionary

How would one go about trying to utilize the base models or embeddings to create a dictionary. I ask because I think a smart dictionary would give more concise and informative definitions to certain words. I’m thinking there are two possibilities a live dictionary which would generate definitions in real time, or a frozen model which would be just like a normal dictionary in which all the words are defined ahead of time.

I was thinking of giving example definitions as the prompt, but that feels like too much of an influence on the output, so from that perspective it seems like the embeddings would need to be used, but then again maybe more influence would be good in certain contexts.