You got a point there! I haven't given it too much thought really -- I def need to check out ada.

But wouldn't the ChatGPT embeddings still be better? Given that they're cheap, why not use the better option?

You can use the embeddings to search through documents. First, create embeddings of your documents. Then create an embedding of your search query. Do a similarity measurement between the document embeddings and the search embedding. Surface the top N documents.

Will we be able to generate embeddings using the ChatGPT API?