University of Geneva's DIAMOND AI Model Lets Users Play Counter-Strike Without Game at 10 FPS

Oct 12, 2024, 04:03 PM

Researchers at the University of Geneva have developed a new neural network model that allows users to play Counter-Strike: Global Offensive (CS:GO) without the actual game. Known as DIAMOND's diffusion world model, it was trained solely from game recordings and simulates CS:GO environments by generating future frames based on keyboard and mouse inputs. This technology enables real-time gameplay at approximately 10 frames per second on personal computers, including MacBooks and laptops with NVIDIA RTX 3090 and 4060 GPUs. A one-click launcher is available for users to download and play the AI-generated game, which also includes Atari games. The open-source project demonstrates significant advancements in world models and could have potential applications in robotics through real-world interactions.