So I’m reading how each split segment is vectorized for storage in vector databases so that queries on those databases can be scored for similarity and then segments that are k most similar can be looked at but has anyone tried converting each of those segments into question form first using an LLM? Or using the LLM to come up with a question for that segment of text and then storing that alongside the original segment? Then similarity searches on those vector databases might come up with better matches?
Is this not worth doing because of the time and labor cost or there is not enough of a performance boost to warrant the work involved?