I want to know if adding information to the knowledge base (as tables) pre-trains the model? Or is it same as passing the same data as a prompt while directly making a call to GPT? Context: I have like 500 lines of data in CSV format that I want to run query against. So I wanted to better understand the pros/cons between using the two options?

The Botpress KB is a vector database, if that answers your question. The advantage to using a table as a KB is that it's globally accessible and reduces token usage and is cached.