Comments by "Tiago D\x27Agostini" (@tiagodagostini) on "" video.

@DefensePoliticsAsia In fact no. I work with this type of AI. When the image is very very uncommon it matches very little natural patterns it is easy to get AI generated. A burned baby is something I am SURE the Adversary network was never trained with. So As a specialist I can say, you CANNOT trust an adversary network to make such classification. Be very careful with you statement. I know you have good intention but you do not understand the math concepts required to say somethign with the strength of statement you said. No You CANNOT say that is or is NOT AI generated using an adversarial network. These networks are trained in tandem with a generative network and they are fed by a trueth knowing part if the image they were fed with was or was not generated by the generative network (After they have tried to guess) . They are fed with a mix of things from the generative Ai and a mix of real life images. THere is a natural bias that exists that whenever the network is fed with a something very weird, to guess it was AI is a statistically likely option that is true. The result is that these networks are only good at detecting AI works in fields where the images overlap in patterns and content with things they were fed upon their training. Again, NONE of these networks were fed with burned baby images.
2