ChatGPT-4 - Use PDFs for data exclusively

[micheal.killarney]

Is it possible to use GPT-4 api and have it answer queries based exclusviely on links to PDFs or just train it on the information in those PDFs?

[georgecollier]

Sure.

[ambroisedlg]

@micheal.killarney I would suggest looking into the Assistants API

[agiledood]

Yes, I do this on my site.

You can use this plugin ChatGPT Toolkit (with streaming) Plugin | Bubble

or build it with API connector.

[micheal.killarney]

To add to this, can this data not be used to train the OpenAI models? As it would contain sensitive information

[micheal.killarney]

With the api connector, do I have to change anything in how I set the api up or is it a workflow?

[moabe.leite]

You can train a Pinecone vector database, with your own information and use gpt4 to respond those questions

Upload and Search Your Own Documents With Bubble, OpenAI, and Pinecone (Part 2)

[agiledood]

@micheal.killarney

With the api connector, do I have to change anything in how I set the api up or is it a workflow?

You’d have to implement your own AI if you’re using the API connector and the biggest annoyance is (as of now), the Assistant’s API has no webhooks or ability to ping you when it’s finished a response so you have to use a ‘do every XX seconds’ and wait for the status to be completed.

If you have sensitive info, I think the Assistants API keeps it private, I’m not sure.