The Future of Multimodal Models

With the amount of data that's been required to train these large language models, would we expect a plateau in the intelligence of these systems? And how might that slow down or restrict the advancement? We're not hitting the data limits for multimodal models yet. But think about images and video. Multimodal models will be much smarter than models that just train on language alone. They'll have a much better idea of how to deal with space, for example. The most expensive lecture I've ever paid for, but I think it was worthwhile.