SnowPro Specialty: Gen AI Question 98
Single answerEXTRACT_ANSWERA support analytics team stores product manuals in a Snowflake table and wants to build a SQL-based workflow that answers user questions such as "What is the maximum operating temperature?" They need the result to be a concise answer grounded in the manual text, not a generated summary that introduces extra details. Which Snowflake function is the best fit for this requirement?
- A
Use EXTRACT_ANSWER to return the answer span from the provided document text based on the question
- B
Use SUMMARIZE to generate a short version of the manual and then search the summary for the answer
- C
Use COMPLETE with a generic prompt, because it is designed to deterministically locate exact answer spans in source text
- D
Use CLASSIFY_TEXT to identify which section of the manual is relevant and treat the label as the final answer
Show answer and explanation
Correct answer: A
Explanation
The best choice is EXTRACT_ANSWER because the scenario requires question answering over existing document text with an answer grounded in the source. This is different from summarization, generation, or classification. In Snowflake Cortex AISQL, EXTRACT_ANSWER is the specialized function for extracting answers from text given a question, making it more appropriate than COMPLETE for this use case when precision and grounding are important. Best practice is to use a task-specific function when one matches the requirement, rather than relying on a general-purpose generation function that may introduce extra wording or unsupported information.
- A. Correct.
Correct. EXTRACT_ANSWER is intended for question-answering over supplied text and is designed to pull out an answer grounded in the provided content. In this scenario, the team wants a concise answer extracted from the manual rather than an open-ended generated response, which aligns directly with EXTRACT_ANSWER's purpose.
- B. Incorrect.
Incorrect. SUMMARIZE condenses text, but summarization is not optimized for precise question answering. A summary may omit the exact fact being asked about or rephrase content in a way that loses the specific answer. This option reflects the common misconception that summarization can replace targeted extraction.
- C. Incorrect.
Incorrect. COMPLETE can generate responses from prompts, but it is a general text generation capability rather than a specialized extraction function. While it can be prompted to answer questions, it is not the best fit when the requirement is to extract a concise answer grounded in provided text with minimal risk of adding unsupported details.
- D. Incorrect.
Incorrect. CLASSIFY_TEXT is for assigning text to categories or labels, not for extracting factual answers from a document. Identifying a relevant section might help as a preprocessing step in a larger pipeline, but the classification label itself would not satisfy the requirement to return the actual answer.