About @Google Genie and “AI-generated games.”
It looks incredible.
It is incredible.
But it’s important to understand what it actually is, and what it is not.
Genie is a world model.
It learns from video and predicts what the next frame should look like when you take an action.
It generates plausible visual continuations of a scene.
It does not run:
– game logic
– physics systems
– multiplayer sync
– progression
– economies
– persistent worlds
In other words:
It’s media generation, not world simulation.
As I explained in my thread yesterday 👇
World models are probabilistic (what might happen next).
Game engines are deterministic (what must happen given the rules).
That’s why Genie can look like a game…
… but it does not maintain persistent world state. It forgets the ledge you stood on 30 seconds ago.
A real metaverse, like @xSPECTAR as the 3D internet requires:
✔ shared state
✔ persistence
✔ reproducibility
✔ rules
✔ identity
✔ multiplayer consistency
Those don’t come from predicting pixels.
They come from engine logic + world rules + networking.
Long term, these technologies will merge:
Engines = structure
AI = content & behavior
But calling world models “game engines” confuses:
🎨 visual imagination
with
🌍 actual simulated worlds
Much of this distinction echoes what @TimSweeneyEpic has said about determinism and world state, and how the Genie team at @GoogleDeepMind frame world models.
Genie is powerful.
Genie is new.
Genie is exciting.
But it is not the same thing as a usable metaverse or the 3D internet.
And understanding that difference is exactly what matters next 🧬
1/ There’s growing excitement about “AI-generated games” after demos of Google Genie.
But Genie is not a game engine, and understanding why matters.
— DS 🪝🧬🟧🟦🍌🦍🧪 (@xspectDS) January 30, 2026
Source by DS 🪝🧬🟧🟦🍌🦍🧪
