Perhaps in the future there could be a way where vector database are used to distill large datasets and enable the training of smaller, more efficient models