A little over a year ago it was presented to the public Nvidia GameGAN, a neural network that had been able to create its own version of Pac-Man simply by watching videos of the game. Now they are trying to do the same with Grand Theft Auto.
With a demo published in GitHub for the enjoyment of connoisseurs and curious, Nvidia has shown that your neural network can learn everything you need to know about GTA V to recreate a small portion of the game including all its details and mechanics.
From the control of the vehicle to often imperceptible details such as the reflection of the sun on the rear window or the shadow on the asphalt, the most impressive of GAN Theft Auto It is not that it manages to replicate a game like this with a very nappy look, but that it is able to do so without generating any type of code by a human.
That it still has a long time to become a completely useful tool at the development level is more than evident, of course, but it is no less impressive to have in your hands the demo of a game that has been created by an AI.
Advances in artificial intelligence have already left us with sweet potato paste on more than one occasion with GTA V as the protagonist. Just a few weeks ago we saw how Intel managed to give the game a photorealistic look no human intervention, but this from Nvidia seems even crazier.
