Chain of Thought

Five words appended to a prompt, "Let's think step by step", took OpenAI's text-davinci-002 from 17.7% to 78.7% on the MultiArith word-problem benchmark and from 10.4% to 40.7% on GSM8K when Takeshi Kojima and colleagues tried them in 2022. Three years later, OpenAI's prompting guide for its o3 and o4-mini models carried a section titled "Avoid Chain of Thought Prompting", warning that asking these models to reason more "may actually hurt the performance".