What you describe is it what the models can do today, but what they might be able to do with more work. It is possible conceptually now, but getting it to work on a product or MVP level - still risk there.