And you think that it's all or nothing? Like they can't create a story or cut-scenes, but use AI to generate the gameplay? How do you know the levels would be endless? How do you know that restrictions can't be added in place? The article was about how it generated existing content.
The point is that you're interacting with a video, and AI is becoming better at generating videos. The "AI clone" of the game is not rendering via traditional graphics engines or even via a game engine. It's not performing physics or anything like that. Assets were also injected using images rather than being fully modeled.
They're showcasing the fact that their AI is taking user input in near-realtime and then generating an image that closely reflects what the real game's video had looked like. If you instruct an AI how a game should look and play, and can inject content via pictures instead of modeling them, then the production cost of a game would be quite significantly reduced.