Google

When a Zero-Shot CoT prompt generates a chain of reasoning but no final answer, a second prompt can be employed to extract the conclusion. This follow-up prompt typically combines the original input question with the reasoning steps the model has already produced.

Using a Second Prompt to Extract Answers from Incomplete CoT Reasoning

You are interacting with a large language model to solve a problem. You provide the model with the initial prompt and receive the corresponding output, as shown below. The model has produced a correct line of reasoning but has failed to state the final answer explicitly. Your task is to design a second, follow-up prompt that uses the existing information to guide the model to provide only the final, conclusive answer.

Extracting a Final Answer from a Language Model

A language model was prompted with a question and produced a correct line of reasoning but did not state the final answer, as shown below. Write a single, effective follow-up prompt that incorporates both the original question and the model's reasoning to extract the final answer.

---
**Original Question:** "A recipe calls for 2 cups of flour to make 12 cookies. If you want to make 30 cookies, how much flour do you need?"

**Model's Reasoning:** "First, I need to find out how much flour is needed per cookie. That's 2 cups / 12 cookies = 1/6 cup of flour per cookie. To make 30 cookies, I need to multiply the amount of flour per cookie by the number of cookies I want to make. So, the calculation is (1/6) * 30."
---

Constructing a Follow-Up Prompt

A language model was prompted with a question. It produced a correct line of reasoning but did not state the final answer, as shown below.

---
**Original Question:** "A grocery store has 15 apples. They sell 7 and then buy 12 more. How many apples do they have now?"

**Model's Reasoning:** "Okay, let's break this down. The store starts with 15 apples. They sell 7, so we subtract 7 from 15, which is 15 - 7 = 8. Then, they buy 12 more, so we add 12 to the current number, which is 8 + 12 = 20."
--

Learn Before

Model's Reasoning: "Okay, let's break this down. The store starts with 15 apples. They sell 7, so we subtract 7 from 15, which is 15 - 7 = 8. Then, they buy 12 more, so we add 12 to the current number, which is 8 + 12 = 20."

Related