New AI Model 'Learns' How To Simulate Super Mario Bros. From Video Footage

Researchers have used a model called MarioVGG to generate video of Super Mario Bros. in response to user inputs, following the success of Google's GameNGen AI model with Doom. The results show impressive physics and gameplay dynamics inferred from studying video and input data, despite glitches and slow gameplay. The researchers aim to create a reliable and controllable video game generator, potentially replacing game development and engines. The MarioVGG model is a step towards this goal, using generalized image diffusion techniques similar to GameNGen. The pre-print paper was published by Virtuals Protocol, a crypto-adjacent AI company. The model's current limitations include apparent glitches and slow gameplay, but it demonstrates the potential for AI models to understand and replicate game mechanics. In the future, such models could revolutionize game development by generating games based on user inputs. This technology could significantly impact the gaming industry, making game creation more accessible and efficient. The study highlights the advancements in AI's ability to understand and generate complex content like video games. Further research is needed to refine the model and overcome its current limitations.