But these models don't work that well even for text when you gave them a huge context. They're reasonably good at summarization, but if you ask them to "continue the story" they will write very inconsistent things (eerily similar to what a sloppy human writer does, though.)