📝The paper introduces a decoding technique for using large language models that involves an explicit tree search over model outputs.
🌈The technique is useful for tasks that require stepwise investigation and problem solving.
🤔The paper proposes new tasks where the decoding technique performs well.
📝Chain of Thought prompting is a similar technique that instructs the model to explicitly make intermediate steps in problem solving.
🎯The paper suggests combining Chain of Thought prompting with the proposed decoding technique for better results.