Much of the progress on LLMs has been through text or image modalities, but Google has put an entirely new paradigm into the hands of users.
Google has made its Genie 3 model available to Ultra G1 users in the US starting today. The model, which was first demoed in August last year, allows for the creation of video-game like 3D worlds in which users can move around and interact with elements. Unlike an AI-generated video which is static, Genie 3 allows people to actually interact with these worlds in real time.
And users seem to have lost no time in using Genie 3 for some interesting use-cases. People began creating game video game worlds that they could play in.
Users could simply “prompt” video-game like functionality of their choosing, which would typically take companies large teams and millions of dollars to create.
Genie 3 allows for the creation of entire worlds with just an image. This user managed to create a roller-coaster game by just giving the model a first-person’s view of a roller coaster.
Indie hacker Pieter Levels created a world model set in New Amsterdam in the 1600s.
And another user was able to walk through San Francisco in the 1900s.
Yet another user was able to walk around a scene depicting the crucifixion of Christ, just from a prompt with coordinates and the date 33 AD.
Users could also create some real-world use-cases, such as sample bodycam footage.
And it is possible to create some surreal worlds as well — it was possible for a user to create a first-person view of a pack of cigarettes moving around Penn Station.
And another user created a hall of mirrors, and was pleased with the results.
These are all very impressive experiences, and are a step up from what was able to do with AI so far. The results had been so impressive that several commentators had wondered if our own world was a simulation. And with the Genie 3 model in the hands of users, there’s no telling what sort of new innovations and products it might end up spawning.