The interesting conclusion from this seems to be that the AI then could only get better when the underlying models improve, right? So we'd still need to keep w

The interesting conclusion from this seems to be that the AI then could only get better when the underlying models improve, right? So we'd still need to keep working on them anyway, or not?