Most of us aren't playing with multi-modal models yet. Text can describe any world, but when it starts coming AI watching live video feeds for learning, can it base itself in reality with that?