Microsoft Research has developed a new reinforcement learning framework that trains large language models for complex reasoning tasks at a fraction of the usual computational cost. The framework, ...