3 ms·
> * Trustworthiness of content (text and soon photo and video) is and will be a fundamental problem. This is the basic attack vector for national and global pol
by zelly 7y ago
> * Trustworthiness of content (text and soon photo and video) is and will be a fundamental problem. This is the basic attack vector for national and global politics in the last few years. Anyone with a good solution would help the world immeasurably and have their pick of customers.
The solution to this is just to train and use only the discriminator used in deepfake GANs. The good guys will inevitably have more compute power than the bad guys, and the discriminator will be better than the generator.
- woodson 7y agoUnfortunately, it’s not the solution – the discriminator is only going to work well on deepfakes produced by that particular GAN model, not the plethora of other possible models.
- somatic 7y agoIf the fake is "pixel-perfect", then how would there be any way to tell? Your fundamental assumption is that the resolution inherent to the transmitted medium is greater than the "resolution" of the faker.