MIT drops Ataraxos — an AI that just crushed the world's best Stratego player in a game where you can't even see most of your opponent's pieces.
The stats are brutal:
→ Demolished top human 15-1-4
→ Went 39-2 against elite players at world championship
→ Smoked DeepMind's DeepNash with 1/100th the training data and 1/30th the self-play games
The edge? Combines self-play with real-time planning + a generative model that predicts hidden pieces on the fly.
This isn't just a game flex — imperfect information games are closer to real markets than chess ever was. If AI can model hidden state this efficiently, think about what's coming for on-chain MEV, dark pool analysis, and predictive alpha in DeFi.
We're not ready.
The stats are brutal:
→ Demolished top human 15-1-4
→ Went 39-2 against elite players at world championship
→ Smoked DeepMind's DeepNash with 1/100th the training data and 1/30th the self-play games
The edge? Combines self-play with real-time planning + a generative model that predicts hidden pieces on the fly.
This isn't just a game flex — imperfect information games are closer to real markets than chess ever was. If AI can model hidden state this efficiently, think about what's coming for on-chain MEV, dark pool analysis, and predictive alpha in DeFi.
We're not ready.