Gemini's Deep Research mode also searches videos in addition to sites. From what I can tell, Ki doesn't seem to. While proper video input would probably be too expensive, and Google has access to YouTube's database itself for efficient querying, it would be nice if the model could look up and use transcripts of videos.
There is some knowledge that is largely contained within videos rather than written form, or there exist videos that are higher quality than their written equivalents. An example would be wanting to learn about different perspectives on a certain issue (e.g. a political/economic policy).
