Experts are feeding their AIs with anything they can get their hands on. From YouPorn search queries to Harry Potter books, it's simply impossible to predict the results. The latest developments in AI and neural networks lead us to the Twitter profile of Robbie Barrat, who decided to train his generative adversarial network with thousands of nudes to create similar art...
So, how well does it do? Let's just say that eroticism is conspicuously absent, and the works of this neural network are worthy of a lava lamp turned on at four in the morning. Barrat, who recently finished school and is currently interning under Nvidia, explained that sometimes networks fall into the trap of 'local minima', meaning that the generator and discriminator networks end up fooling each other without improving the final results. In this specific case, the generator network discovered that it can easily fool the discriminator with lumps of deformed flesh and greasy appearance.
What do you think? Personally, I'd do a hard reset and let it try again.
