one-sample-polymath-learning

Demonstrate that a single strategically engineered training sample can improve reasoning across multiple domains. Polymath learning shows sample quality and multidisciplinary design matter more than quantity, enabling extreme data efficiency in RL training.

Independent GitHub snapshot · Automated classification · Original documentation governs