Google's game world falls apart within a minute: DeepMind admits Genie 3's struggle against traditional engines

Google DeepMind has provided an update on the current state of their much-talked-about technology, Genie 3. The short version? No miracle has occurred, and a replacement for traditional game engines isn't on the horizon anytime soon.

The key takeaway, which is sure to cool investor enthusiasm, is that the virtual spaces generated by the neural network start to "fall apart" roughly 60 seconds after interaction begins. According to Google's representatives, the algorithm simply hits technical limitations, and the image loses coherence. To be fair, just a couple of months ago, the model could only manage a few seconds, so progress is evident.

The Genie 3 generator operates on complex principles that many still misunderstand. It's not an engine in the classic sense that calculates physics and geometry. Instead, the model generates the world frame by frame, creating, in essence, an infinite interactive video that responds to player actions but doesn't store a complete 3D representation of what's happening in its memory.

On the sidelines of the conference, DeepMind representative Alexandre Moufarek confirmed to journalists that creating video games was never the team's end goal—it's a flashy demo, but the technology was actually developed for other purposes. DeepMind's primary focus is on creating Artificial General Intelligence (AGI), and Genie 3 is needed to generate unique virtual worlds where other AI agents could be trained and learn to navigate.

This statement sounds much more measured than the hype surrounding Genie 3 in January, when news of the technology caused the stocks of several game companies to plummet. Back then, many feared that AI would replace developers, but it's now becoming clear that the technology is still far from commercial use in game development. Simply put, Google showed off an impressive prototype, but it will be a long time before a neural network can maintain a stable world for even 10 minutes without glitches.

What do you think? Should we even be trying to replace game engines with neural networks, or could the controlled chaos of AI offer players a new, unconventional experience? Share your opinions in the comments.