I was just watching the vid! I was like, oh wow all of these levels look really familiar… it’s not imagining new “Doom” locations, its literally a complete memorization the levels. Then I saw their training scheme involved an agent playing the game and suddenly I’m like oh, you literally had the robot navigate every level and look around 360 to get an image of all locations and povs didnt you?
and yet, with zero evidence to support the claim, the paper’s authors are confident that their model can be used to create new game logic and assets:
Today, video games are programmed by humans. GameNGen is a proof-of-concept for one part of a new paradigm where games are weights of a neural model, not lines of code. GameNGen shows that an architecture and model weights exist such that a neural model can effectively run a complex game (DOOM) interactively on existing hardware. While many important questions remain, we are hopeful that this paradigm could have important benefits. For example, the development process for video games under this new paradigm might be less costly and more accessible, whereby games could be developed and edited via textual descriptions or examples images. A small part of this vision, namely creating modifications or novel behaviors for existing games, might be achievable in the shorter term. For example, we might be able to convert a set of frames into a new playable level or create a new character just based on example images, without having to author code.
the objective is, as always, to union-bust an industry that only recently found its voice
Which is funny, as creating new levels in an interesting way is very hard. What made John Romero great is that he was very good at level design. He made it look easy. People have been making new levels for ages but only few of them are good. (Of course also because you cannot recreate the experience of playing doom for the first time, so new experiences will need to be ‘my house’ levels of complexity.
I was just watching the vid! I was like, oh wow all of these levels look really familiar… it’s not imagining new “Doom” locations, its literally a complete memorization the levels. Then I saw their training scheme involved an agent playing the game and suddenly I’m like oh, you literally had the robot navigate every level and look around 360 to get an image of all locations and povs didnt you?
and yet, with zero evidence to support the claim, the paper’s authors are confident that their model can be used to create new game logic and assets:
the objective is, as always, to union-bust an industry that only recently found its voice
Which is funny, as creating new levels in an interesting way is very hard. What made John Romero great is that he was very good at level design. He made it look easy. People have been making new levels for ages but only few of them are good. (Of course also because you cannot recreate the experience of playing doom for the first time, so new experiences will need to be ‘my house’ levels of complexity.