The AI video description generator doesn't take context into account, it keeps generating descriptions that are vaguely related to one or two words in the script. That entire AI feature has to be rethought from the ground up.