Well, these world models represent a type of artificial intelligence system that is especially developed for different purposes, such as entertainment or education, because it can successfully simulate environments. So, users can type in any prompt and wait for the world model to generate a space similar to a video game.
Compared with the Genie 2 model that was officially released a couple of months ago, in December. Back then, the Genie 2 model was only capable of generating interactive scenes from images. Also, those videos had a time limitation of under one minute.
Now, Genie 3 is promised to be more capable because users can generate, based on their prompts, interactive environments that support more than one minute of continuous interaction. According to the Google giant, Genie 3 has a visual memory of up to one minute, which means that users can easily turn back to a change that they’ve made, while the rest of the scene remains the same.
The interesting thing about Genie 3 is that users can create world events based on their prompts. So, they will be able to make significant changes, such as adding new characters or even changing the weather in their generated world.
“The model is auto-regressive, meaning it generates one frame at a time. It has to look back at what was generated before to decide what’s going to happen next. That’s a key part of the architecture.”, the research director at DeepMind stated.
It’s important to mention that even if it seems like an interesting new AI world model, Google DeepMind announced that it will remain for now in a research preview version, and users need to wait a little longer to have it publicly available.
It’s true that at first glance it seems like an interesting artificial intelligence tool, and also a significant step ahead compared with Genie 2, but stay tuned to be the first to find out when Google Genie 3 will be released!