🔥 We release Gamma-World from NVIDIA — a generative multi-agent world model that finally goes beyond 2 players. ⚡ 24 FPS real-time streaming 🧩 Simplex Rotary Agent Encoding — permutation-symmetric 🌐 Sparse Hub Attention: O(N²) → O(N) 🎯 2 → more players, zero-shot 🤖 Games → real multi-robot worlds... ━━━━━━━━━━━━━━━ 💥 THE SINGLE-AGENT ERA IS OVER. 💥 ━━━━━━━━━━━━━━━ 🔗show more

@nvidia Great work. Finally seeing multi-agent interaction in game world modeling. I also noticed ReactiveGWM doing the similar thing, but NPCs are not quite the same as players.

@nvidia i'd want to see what breaks first as player count grows: attention cost, identity tracking, or control.

@nvidia I really like the bypassed method of diffusion kv and the student. Really impressive with this.

@nvidia Wow this is pretty wild stuff! The Model is the product

@nvidia we got multiplayer in world model before gta 6

@nvidia The zero-shot scaling from 2 players to N is the real unlock here.

@nvidia finally a model that knows more than two people exist guess o(n) attention means i can afford to run it now

@nvidia Impressive work. Moving beyond two-player world models while maintaining scalable agent communication feels like an important milestone for the field.

@nvidia 非常酷炫啊

@JunGao33210520 @nvidia Neat!

@nvidia I’ve been using neural networks professionally for 8 years. Measuring your performance on your own data means nothing. Even most foundation models fail to work on real-world problems.

@nvidia Hey, can you check dm pls? It’s about what you are building, thanks @fangfu0830

@nvidia check DMs plz