Learn More Language models can generalize better when left to create their own solutions, a new study by Hong Kong University and University of California, Berkeley, shows. The findings ...