> We’re also building tools to help detect misleading content such as a detection classifier that can tell when a video was generated by Sora.

I am curious of how optimised their approach is and what hardware you would need to analyse videos at reasonable speed.