MIT researchers used a Battleship-style test to show how smaller AI models can improve by asking sharper questions, ...