Google

A key technique for improving the accuracy of large language models on reasoning tasks is to include the step-by-step logical process, not just the final solution, within the prompt. This provides a clear template for the model to follow, guiding it to the correct answer.

Incorporating Reasoning Steps into Prompts for Improved Model Accuracy

A developer is trying to build a system that uses a large language model to solve simple logical ordering problems. To guide the model, they provide the following example in the prompt:

`Question: Sarah is older than Michael. Michael is older than David. Who is the youngest?
Answer: David.`

The developer finds that the model performs poorly when given new, slightly different problems (e.g., 'Who is the oldest?').

Evaluate the developer's example prompt. Explain the fundamental flaw in its design and rewrite the example to include the necessary components that will more effectively teach the model how to solve this class of problem.

Improving AI Reasoning with Better Prompts

A developer is using a large language model to solve simple physics word problems. They provide the following example in their prompt to guide the model:

`Question: A car accelerates from rest to 20 m/s in 5 seconds. What is its acceleration?`
`Answer: 4 m/s²`

When the developer then asks the model to solve a new problem, the model provides an incorrect answer. Based on principles of effective prompt design for reasoning tasks, analyze the developer's example and explain why it is likely failing to guide the model effectively. Then, describe how you would rewrite the example to improve the model's performance.

Diagnosing and Improving a Reasoning Prompt

A developer is trying to get a large language model to reliably solve multi-step arithmetic problems. Which of the following example prompts is best structured to guide the model to the correct reasoning process for a new, similar problem?

Learn Before

Related