For example, as an AI researcher, I can't consider Imagen/Parti to be the state of the art if all we have are cherry-picked examples and we can't verify anything. For all practical intents and purposes, they are just vaporware, and the state of the art are models like Stable Diffusion or DALL-E.
Of course they are free to keep them that way, but they risk losing their reputation as AI/ML/NLP leaders.