SnowPro Specialty: Gen AI Question 104
Single answerSUMMARIZEA product support team stores long case notes in a Snowflake table and wants to generate a short, plain-language summary for each case so agents can review the issue faster. The team needs a solution that minimizes prompt engineering and uses a built-in Snowflake function intended specifically for summarization. Which approach best meets this requirement?
- A
Use the SUMMARIZE function on the case note text column to generate concise summaries directly in SQL.
- B
Use COMPLETE with a custom prompt because Snowflake does not provide a dedicated summarization function.
- C
Convert the case notes to vector embeddings first, then call SUMMARIZE on the embedding column for better summary quality.
- D
Use EXTRACT_ANSWER because it is designed to create a concise summary of an entire document.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use SUMMARIZE directly on the text column containing the support case notes. In Snowflake Cortex AISQL, SUMMARIZE is a built-in function specifically intended to generate concise summaries from text, making it the most appropriate choice when the goal is straightforward summarization with minimal prompt design. COMPLETE is more flexible but is a general text generation function, so it is better suited when customized instructions are needed rather than when a dedicated summarization function already exists. EXTRACT_ANSWER addresses question-answer extraction, not broad summarization, and embeddings support retrieval and similarity workflows rather than serving as direct input for summarization. This aligns with Snowflake best practices of choosing the most task-specific Cortex function available for the use case.
- A. Correct.
Correct. SUMMARIZE is the Snowflake Cortex AISQL function designed specifically to produce a concise summary from input text. In this scenario, it fits the requirement to generate short summaries of long support case notes while minimizing prompt engineering, because the function is purpose-built for summarization and can be invoked directly in SQL over text columns.
- B. Incorrect.
Incorrect. COMPLETE can be used for many generative tasks, including summarization, but this option does not best meet the stated requirement. The scenario explicitly asks for a built-in function intended specifically for summarization and wants to minimize prompt engineering. Using COMPLETE would require the team to craft and maintain prompts, making it less direct than using SUMMARIZE.
- C. Incorrect.
Incorrect. SUMMARIZE operates on text input, not on vector embedding columns. Embeddings are useful for semantic search, retrieval, and similarity use cases, but they are not the correct input for document summarization. This distractor reflects a common misconception that embeddings improve every LLM task, when in reality they serve a different purpose.
- D. Incorrect.
Incorrect. EXTRACT_ANSWER is intended for extracting an answer to a question from provided text, not for creating a general summary of an entire document. It is better suited to targeted question-answering scenarios, such as finding a specific policy detail in a long document, rather than summarizing all case notes.