Google DeepMind unveiled a new AI model called Genie 3. In demo videos posted on the company’s website, you can watch photorealistic 3D worlds, created from scratch from text prompts. It’s possible to navigate these worlds like you would in a videogame.
But there’s one profound element to the videos: Water. In one, ocean waves crash against a barrier, sending spray up into the air as seawater spills onto the street. It looks almost indistinguishable from real life. Fluid dynamics is one of the most difficult things to recreate on a computer. While Genie 3 doesn’t actually understand physics, it makes fairly accurate predictions by picking up on patterns it sees in video footage.
If you think AI models are expensive to run, the cost of modeling physics is exponentially higher (ask any Formula 1 team or weather forecaster). If physics modeling is ultimately replaced by neural networks, and AI can accurately recreate practically any physical environment, it has huge implications across many industries, including science and robotics. Based on the rate of progress, that day appears to be rapidly approaching.