New research shows large language models rival humans in learning logic-based rules, reshaping how we understand reasoning.