• Login
  • Register

Work for a Member company and need a Member Portal account? Register here with your company email address.

Project

The Indoor-Training Effect: Unexpected gains from distribution shifts in the transition function

Would you train for a tennis match in chaotic outdoor conditions from day one, or would you first perfect your fundamentals in a controlled indoor court? Turns out, AI benefits from the same principle. Our research uncovers the "indoor training effect"—showing that AI agents trained in stable, noise-free environments actually perform better in unpredictable real-world settings than those trained in chaotic conditions from the start. By first mastering core strategies in structured settings, AI develops more reliable decision-making abilities that hold up under uncertainty.