I have been playing around with generating embeddings for some documents, storing those in a Postgres column (pg_vector), generating emeddings for a search term

I have been playing around with generating embeddings for some documents, storing those in a Postgres column (pg_vector), generating emeddings for a search term, finding the closest match using pg_vector and sending the results off to GPT3 to be summarized and it seems to work incredibly well. It seems to stay on topic, I’ve been trying to “jailbreak” it and get it to say crazy things for a while. I suppose this is more of one off question answering thing than a chatbot, though (previous questions and results are not fed back in). I have only tried with super simple “documents” so far, I am going to try some more complex ones when I get a chance to get a pipeline setup.