this is somewhat true but i'm not sure how load bearing it is. for one, i think it's going to be a while until 'we asked the model what bob said' is as admissible as the result of a database query