How to Evaluate a Video GPT Model?

Video GPT is what the name says it's a GPT for video models. You can't just throw a straight GPT transformer at it at a pixel level. So how do you evaluate a video model? There are metrics in general evaluating generative models is really hard. We're not even at a point in video generation where a human would find it hard to save whether this video is from AI or from a human until we get there.