Researchers Unveil DIAMOND, Diffusion Model Trained on 5M Frames to Simulate Counter-Strike

Oct 12, 2024, 11:54 AM

Researchers have developed DIAMOND, a diffusion world model that allows users to play Counter-Strike: Global Offensive without the original game. This neural network generates future frames based on keyboard and mouse inputs, effectively simulating the game environment. Trained on 5 million frames, DIAMOND can be downloaded and run on personal computers, including MacBooks, achieving approximately 10 frames per second on a 3090 GPU. Developed by a team including Eloi Alonso, Adam Jelley, and Vincent Micheli from the University of Geneva, this advancement signifies a significant leap in world models with potential applications in robotics. As this technology progresses, open-source safety layers are anticipated to be needed.