A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a novel approach to training AI models that significantly reduces computational costs. The method, called ‘Lazy Learning,’ allows models to selectively engage complex calculations only when necessary, rather than continuously. In tests on image classification tasks, this approach achieved comparable accuracy to traditional methods while using up to 80% less processing power. Researchers suggest this could make advanced AI more accessible and environmentally sustainable by lowering the energy barrier for development and deployment. Read the full article for detailed methodology and expert commentary.