I don't think they 'index' videos, per se. They just point the model at the video's transcript on demand when you ask a question, I believe. Doesn't change any of your conclusions, though. You're absolutely right, they have an absolute ton of data.