Google

A Large Language Model's ability to reason can be enhanced by providing it with a detailed, step-by-step reasoning process for a comparable problem. This demonstration enables the model to learn how to construct its own logical problem-solving path, increasing its chances of reaching the correct solution for new tasks.

Improving LLM Reasoning with Step-by-Step Demonstrations

An analyst is struggling to get a large language model to produce structured summaries from customer feedback reports, as described in the case study below. Propose a new prompting strategy that incorporates a step-by-step demonstration to improve the model's reasoning. Describe what this new prompt should contain and explain why it would be more effective than the analyst's current approach.

Improving LLM Performance on a Complex Summarization Task

A developer is trying to teach a large language model to identify a main argument and two pieces of supporting evidence from a paragraph. They construct the following prompt, which includes an example before presenting the new task.

---
**Prompt:**

**Example Task:**
*Paragraph:* "The city council's new recycling program, launched last month, has been a resounding success. A report released today showed a 50% increase in recycled materials and a corresponding decrease in landfill waste. This initiative is a model for urban sustainability."
*Question:* What is the main idea of this paragraph?
*Answer:* The new recycling program is successful.

**New Task:**
*Paragraph:* "Recent studies indicate that regular physical activity is crucial for cognitive health. One major study tracked 5,000 adults and found that those who exercised at least three times a week had a 30% lower risk of developing dementia. Another experiment showed that even short bursts of activity can improve memory and focus."
*Question:* What is the main argument and two pieces of supporting evidence in this paragraph?

---

Critique the "Example Task" portion of the prompt. Explain why it is not an effective demonstration for teaching the model how to perform the "New Task." What specific changes would you make to the example to better guide the model's reasoning process?

Critique of a Reasoning Demonstration Prompt

An engineer is using a large language model to solve multi-step physics problems. The model consistently makes logical errors and provides incorrect final answers. The engineer's goal is to improve the model's ability to reason through the problems correctly. Which of the following prompt strategies would be the most effective way to achieve this?

Learn Before

Related