1- Each model looks for different deepfake signatures. By design, the models do not always agree, which is the goal. We are much more concerned with false negatives, and we target a min of 95% accuracy for our model of detection models.
2 - The challenge is educating users about results without requiring a PhD. Our platform is targeted for use by junior analysts in cyber security or trust and safety.
3 - This is a good suggestion. We are exploring how we can offer an unlimited plan that can cover our high compute costs (we run our multiple models in realtime).
Why would you be more concerned about false negatives? Wouldn't false positives erode trust and value in your product, and considering the applications you're targeting, possibly open you up to lawsuits if you start accusing innocent people of being deepfakes (which, IMO, currently seems unlikely)?
We provide a probabilistic percentage result that is used by a trust and safety team to set limits (ie. flag or block content) so it is not a binary yes/no. We search for specific deepfake signatures and we explain what our results are identifying.
So... that percentage sort of 'return type' allows the people using your service to decide how aggressive they want to be? Smart. It also could possibly turn your service into more of a tool and less of something that someone could blame incorrect results on.
