I like the comparison with books, but: running Whisper AI to transcribe YouTube as a whole would give us a enormous amount of data.

Or would it? Are there estimations abut this?