Not really. The training methodology opens up whole new mechanisms that'll make it much easier to train non-language models, which have been very much neglected. Think robot multi-modal models; visual / video question answering; audio processing, etc.