Finally some insight, thanks 
Basically, if i understand, we currently need a database on our side
- Generate embeddings from pdfs, and whatnot, and save to a local database
- Generate embedding from question, and do a query against local database, and return original content
- Use original content as input to pre-feed prompt & add original question
- get result, and pay mucho tokens ^^
I hope OpenAI integrates something like this on their server backend. We need a way to simply add our content to be indexed by GTP.
Once companies can simply have a marketing or documentation team feed the GTP database, and have another team build a frontend that is used internally as a AI Assistant, then this will go to the moon 
Imagine simply feeding it your whole website archive, and then have a search assistant on your site. This would rock on sites that rely heavily on reading materials. QA, marketing, education, ā¦
Here is a simple use-case for the company i work for
They do a lot of collaboration with Airbus industries, much to do with safety regulations, electrical standards, assembly/supply chain management., and training courses/consultancies in these matters
So we have domain experts & instructors (airbus has Service Now as an internal tool)
Now if i could just feed GTP all kinds of technical docs, standards, etc, that would make for a great interactive wikipedia