Generative AIhas made further progress in simulating gameplay, but it still falls short of esports requirements. Eloi Alonso, the head of the open-source AI project DIAMOND, today released a demo video showcasing the world model of Counter-Strike: Global Offensive running on a neural network. Alonso also detailed the numerouschallenges involved, including a mere 10 FPS on an RTX 3090 platform.
It’s crucial to understand that this AI is not playing CS:GO in the traditional sense, nor is it runninga script to automate matches. The demonstration is entirely AI-generated, and the 10 FPS figure reflects the computational demands of running the model itself, not the game’s framerate.
Despite the limitations, the technological achievementis significant. The team trained a single GPU with enough Dust II Deathmatch footage to teach the AI model how to play the game, effectively porting CS:GO into the AI world.
The most intriguing aspect of the video is the AI’s left foot on right foot jumpingtechnique. The AI interprets pressing the jump key as a fixed action (moving a certain amount on the Y-axis), ignoring the physics of Valve’s Source engine. This leads to other hallucinations, such as weapons deforming under certain lighting conditions and the AI even teleporting through walls.
Whilethis AI technology could potentially have negative implications for gaming, its current capabilities are far from threatening. The AI’s world model is a fascinating glimpse into the future of AI and gaming, and it’s exciting to see how this technology will evolve.
References:
- IT之家. (2024, October 12). 开发者成功让 AI 学会打游戏，但用 RTX 3090 神经网络运行《CS:GO》“世界模型”仅有 10 帧. Retrieved from https://www.ithome.com/0/676/826.htm
Views: 0