Not sure if this is already in the roadmap, but one thing AI will boost productivity of the organization is the ability to use AI to answer questions from the knowledge store in the database. Currently, Fibery AI can only take context from the active entity. Is it possible to let it look at the entire database with permission from the user, to answer questions that are related to it?
Not currently possible, but definitely in plans.
In theory, the beta semantic search feature will retrieve the entity that most closely matches the question.
So to build this Fibery Q&A AI you would just need to 1) retrieve the entities that most closely match the question and then 2) give them to an LLM like GPT-4 to generate the answer.
The implementation to get an accurate and verifiable answer isn’t that easy, however.
using the DB to train a private AI model would be great. so that the AI has the same knowledge as the DB and can answer without reading through all the DB over and over again. will that be something that can be done?
Notion’s Q&A is basically doing this. I like that they designed it to be somewhat independent-- you can trigger Q&A using the desktop app without opening the actual Notion page.