Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 7·5 min read
Otter: A Time-Aware, History-Conditioned Human Chess AI
TL;DR
Otter is a novel human chess AI that predicts moves by considering the last 20 moves and time pressure, achieving 55.23% top-1 accuracy. It outperforms the previous model, Maia 2, with fewer parameters and less training data.
✦ Why It Matters
Explore Otter's code and models to enhance your own chess AI projects today.
Key Takeaways
- Utilize move history to improve AI predictions.
- Incorporate time pressure into decision-making models.
- Experiment with smaller models for efficiency.
- Access and adapt Otter's publicly available code.
How It Works
Otter's architecture includes a move history encoder that processes the last 20 moves to capture player tendencies and a time control module that adjusts predictions based on the player's remaining time. This dual conditioning allows Otter to make more informed predictions by understanding both the historical context of the game and the urgency of the current situation.