I wonder what captcha-like mechanism they'll use for annotating this new dataset for their self-driving cars. Gamifying the videos post-drive with AR?