Google

When a Chain-of-Thought prompt generates reasoning steps but no final answer, a second prompt can be used to extract the conclusion. This follow-up prompt is constructed by combining the original input with the model's generated reasoning from the first round. This combined input allows the LLM to continue its thought process and produce the correct final answer.

Answer Extraction via Second-Round Prompting

You provide the following query to a large language model:

**Initial Query:**
"A train leaves City A at 9:00 AM traveling at 60 mph. A second train leaves City B, 300 miles away, at the same time, traveling towards City A at 90 mph. At what time will they meet?"

The model provides the following incomplete response:

**Model's Reasoning:**
"Let's think step by step. The two trains are moving towards each other. Their combined speed is the sum of their individual speeds. So, the combined speed is 60 mph + 90 mph = 150 mph. The total distance to cover is 300 miles."

Your task is to write the complete text for a second-round prompt that will guide the model to provide the final answer, based on the principle of combining the original query with the model's generated reasoning.

Constructing a Follow-Up Prompt for Incomplete Reasoning

A developer is using a large language model to build a question-answering system. They observe that for some queries, the model provides a detailed reasoning process but fails to state the final answer. Based on the interaction log below, describe the specific content of the follow-up prompt the developer should send to elicit the final answer, and briefly explain the principle behind this approach.

Resolving an Incomplete LLM Response

A user provides the following query to a large language model: "A grocery store has 5 apples. They buy 3 more bags of apples, with 4 apples in each bag. They then sell 7 apples. How many apples do they have left?"

The model returns the following text, which includes reasoning steps but does not state the final answer:
"Okay, let's break this down. First, we calculate the total number of new apples: 3 bags * 4 apples/bag = 12 apples. The store started with 5 apples, so the new total is 5 + 12 = 

Learn Before

Related