maybe the biggest practical claim in the Beyond Language paper (beyond modality scaling laws): you don't need Janus. At least, not for vision. Unified representation suffices. Of course, the vision of Janus is a bit more expansive, but maybe it holds in the general case.