AI's Game World Progress: A Step Forward, Miles to Go

Imagine a world where AI can create entire game worlds just from a few seconds of video. That's the dream, right? Well, Microsoft Research is getting closer to making that dream a reality with their new World and Human Action Model, or WHAM. But let's not get too excited just yet. First things first, WHAM is trained using real gameplay footage. This isn't some random data; it's actual gameplay from Bleeding Edge, a game released in 2020. Microsoft gathered enough data to equal seven years of player gameplay. That's a lot of data! Now, WHAM isn't perfect. At first, it struggled to make sense of the gameplay, creating confusing clips. But after a million training updates, things started to improve. The model began to understand complex gameplay interactions, like how a power cell explodes after three hits or how a character flies.

To test WHAM, Microsoft gave it one second of real gameplay footage and asked it to generate the next frames. They also tested its consistency by generating up to two minutes of new AI footage and comparing it to actual gameplay. The results? Well, they're promising, but there's still a long way to go. Think about it. We're talking about AI creating entire game worlds from just a few seconds of video. That's a huge leap from where we are now. But it's not just about the technology. It's about the potential for new games, new experiences, and new ways to play. But let's not forget, this is just the beginning. There's still a lot of work to be done. WHAM might be able to create basic gameplay interactions, but it's far from creating entire game worlds. And that's okay. Every journey starts with a single step, right? So, while WHAM is a step forward, it's also a reminder of how far we have to go. But who knows? Maybe one day, AI will be able to create entire game worlds from just a few seconds of video. Until then, we'll keep watching and waiting.

Actions