They are trained to generate summaries or to create prompts for other models that generate, edit, and manipulate other media.

Unless, of course I am horribly behind the times and they are stuffing all kind of other tasks in the LLM themselves.