I think people underestimate how the nature of current generative AI products is simply based on what extremely large corpuses of data are available. Text, photos, video, audio. This is the easy stuff, the first step.