This is really interesting and something I've been anticipating for a while. An obvious beneficiary would be youtube, whose transcription of every video could w

This is really interesting and something I've been anticipating for a while. An obvious beneficiary would be youtube, whose transcription of every video could with some effort be transformed into a lot of interesting features, same with the chapter markings often supplied by creators. Currently they already detect swearing and nudity, but only early on in videos, which suggests that the costs of running these kinds of models over large corpora of videos is prohibitively expensive. As an aside, is it me or does it make very little sense to display a stacked bar chart on a log scale? They probably should have re-thought that one.