I made a mistake in my previous post about Oasis, the AI that generates Minecraft in real time. I claimed it was the first public model of its kind, but I completely forgot about Microsoft Diamond, which simulates Counter-Strike in a similar way. Looking ahead, I see AI gaming taking an interesting path. In conversations with other developers, we’ve discussed how this technology could work alongside traditional game engines rather than replacing them entirely. The AI could handle specific tasks while game logic manages others – you don’t need the AI doing all the heavy lifting. Playing Oasis feels distinctly surreal – it’s like experiencing Minecraft through a dream state. Google’s work with Doom using AI is also fascinating, though it needs serious hardware to run. Where this leads long-term could be fascinating: Picture a future where Google or another tech giant creates a massive model that generates extensive gameplay based on detailed prompts. Smaller, specialized models would then train on that data to create specific games. Two key factors make this approach viable: 1. It could become the fastest, most cost-effective way to develop games 2. Integration with language models could enable truly dynamic experiences – players might create custom bosses or scenarios on the fly Developers could still maintain game balance through careful training data selection. And here’s an interesting twist – finding ways to trick these AI systems into doing unexpected things could become its own form of gameplay, similar to how speedrunners find creative exploits in traditional games. While this tech isn’t ready for mainstream gaming today, I believe we’ll see practical applications within the next decade. The ability to generate content dynamically based on player input opens up entirely new possibilities for gaming experiences. For more details on how Oasis works, check out my previous analysis here: https://adam.holter.com/oasis-first-publicly-available-ai-generated-playable-game-running-10x-faster/
Menu
