Chain-of-Thought: Teaching AI to Show Its Work

Adding 'Let's think step by step' to a math problem improved GPT-3's accuracy from 18% to 79%. One phrase. Same model. This is the power of chain-of-thought prompting.