Very similar, it also learns from context of a conversation what is likely in a document, a page, video, audio, etc. This alleviates the need to parse media.