Question

A Generative AI Engineer is building a customer-facing math tutoring assistant on Databricks. Internal testing shows that when the model is prompted to reason step-by-step, its final answers are far more accurate. However, product requirements state that end users must only see a concise final answer, not the intermediate reasoning. Which prompt engineering approach best satisfies both accuracy and presentation requirements?

Accepted Answer

Instruct the model to perform step-by-step reasoning inside a delimited scratchpad section, then output only the final answer in a separate clearly labeled section that the application parses and displays.

Answer

Set temperature to 0 and remove all step-by-step instructions so the model produces only the concise answer directly.

Answer

Ask the model to reason step by step and display the entire response to the user, since accuracy is more important than the presentation requirement.

Answer

Reduce max_tokens to a very small value so the model is forced to skip reasoning and return only a short answer.